**An Electrical Impedance Mammographic Scheme — Norms and Pathology**

Alexander Karpov, Andrey Kolobanov and Marina Korotkova

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/60014

#### **1. Introduction**

#### **1.1. The concept of a normal mammographic scheme: mammographic schemes in diagnostics**

A normal mammographic scheme in traditional diagnostic techniques (X-ray, US) is a set of anatomic structures regularly found on normal mammograms. Existing classifications of normal breast X-ray anatomic pictures help distinguish several structural types of the breast. Each structural type is presented by a certain morphological substrate (Figure 1). Wolfe's Classification (1976),

© 2015 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and eproduction in any medium, provided the original work is properly cited.

Tabar's Classification (1997), Boyd's Classification (1980) and the BI-RADS Classification (2000) are the most widely-known [1, 2, 3]. According to Wolfe, breast structural types are defined by the ratio of fat to ducts: N1 – the breast consists mainly of fat; P1 – linear densities occupy no more than 25% of the breast; P2 – linear densities occupy more than 25% of the breast; Dy – dense. Tabar relies on histomammographic correlations and distinguishes five types of breast structure: I – a balanced proportion of all components of the breast tissue; II – fat breasts; III – a predominance of fat tissue with retroareolar residual fibrous tissue; IV – predominantly nodular densities; V – dense breasts. Boyd takes a different approach, introducing a quantitative evaluation of the breast structure: A – 0% mammographic density; B - >0-10%; C - >10-25%; D - >25-50%; E - >50-75%; F - >75%. The BI-RADS classification modifies that worked out by Wolfe, expressing a breast density percentage and adding a division into quartiles: Type 1 – extremely fat, with parenchyma below 25%; Type 2- minimal density, parenchyma 25-50%; Type 3 – heterogeneous density, parenchyma 50-75%; Type 4 – extremely dense, parenchyma 75-100%. Thus, breast structural types have been defined in reliance on the correlation between the ductal component and fat lobules. The breast has a variable appearance in each case and the breast density decreases after the menopause. It should be noted that no similar system of breast structural types has been developed by reference to the US.

A distortion of the normal mammographic scheme is observed when structural changes appear, in the form of pathological shadows and microcalcinations. Such focal changes are easily detected by medical devices diagnosing using tissue density. However, focal changes are not the only mode of cancer manifestation. Diffuse changes of the breast structure may also stand for a malignant process and may equally result in the distortion of the mammographic scheme. Diffuse changes do not affect breast tissue density and present certain difficulties for X-ray diagnostics.

#### **2. Electrical impedance images**

Electrical impedance mammography is a medical imaging technique aimed at creating images of the breast by means of external scanning. Electrical impedance mammography measures electromagnetic phenomena and belongs to non-invasive techniques of image creation [4].

There exist two types of techniques creating tomographic images: local and non-local. X-ray computer tomography, magnetic resonance tomography and positron emission tomography belong to local techniques, implying the passage of one direct ray through the body and the creation of one pixel in the image (one ray – one pixel). The pixel value depends solely on the substance that the ray meets on its way. Using multiple angles for calculation captures twodimensional slices that can be reconstructed for obtaining three-dimensional images of the object scanned.

Electrical impedance mammography belongs to the category of soft-field tomography and uses a non-local technique for image creation, whereby all points on the object scanned affect the measurement result. This is why the value of the mean electrical conductivity is dominant in detecting focal changes.

Tabar's Classification (1997), Boyd's Classification (1980) and the BI-RADS Classification (2000) are the most widely-known [1, 2, 3]. According to Wolfe, breast structural types are defined by the ratio of fat to ducts: N1 – the breast consists mainly of fat; P1 – linear densities occupy no more than 25% of the breast; P2 – linear densities occupy more than 25% of the breast; Dy – dense. Tabar relies on histomammographic correlations and distinguishes five types of breast structure: I – a balanced proportion of all components of the breast tissue; II – fat breasts; III – a predominance of fat tissue with retroareolar residual fibrous tissue; IV – predominantly nodular densities; V – dense breasts. Boyd takes a different approach, introducing a quantitative evaluation of the breast structure: A – 0% mammographic density; B - >0-10%; C - >10-25%; D - >25-50%; E - >50-75%; F - >75%. The BI-RADS classification modifies that worked out by Wolfe, expressing a breast density percentage and adding a division into quartiles: Type 1 – extremely fat, with parenchyma below 25%; Type 2- minimal density, parenchyma 25-50%; Type 3 – heterogeneous density, parenchyma 50-75%; Type 4 – extremely dense, parenchyma 75-100%. Thus, breast structural types have been defined in reliance on the correlation between the ductal component and fat lobules. The breast has a variable appearance in each case and the breast density decreases after the menopause. It should be noted that no similar system of breast structural

A distortion of the normal mammographic scheme is observed when structural changes appear, in the form of pathological shadows and microcalcinations. Such focal changes are easily detected by medical devices diagnosing using tissue density. However, focal changes are not the only mode of cancer manifestation. Diffuse changes of the breast structure may also stand for a malignant process and may equally result in the distortion of the mammographic scheme. Diffuse changes do not affect breast tissue density and present certain difficulties for

Electrical impedance mammography is a medical imaging technique aimed at creating images of the breast by means of external scanning. Electrical impedance mammography measures electromagnetic phenomena and belongs to non-invasive techniques of image creation [4].

There exist two types of techniques creating tomographic images: local and non-local. X-ray computer tomography, magnetic resonance tomography and positron emission tomography belong to local techniques, implying the passage of one direct ray through the body and the creation of one pixel in the image (one ray – one pixel). The pixel value depends solely on the substance that the ray meets on its way. Using multiple angles for calculation captures twodimensional slices that can be reconstructed for obtaining three-dimensional images of the

Electrical impedance mammography belongs to the category of soft-field tomography and uses a non-local technique for image creation, whereby all points on the object scanned affect the

types has been developed by reference to the US.

X-ray diagnostics.

2 Mammography Techniques and Review

object scanned.

**2. Electrical impedance images**

We use the electrical impedance computer mammographer MEIK 5.6 for the creation of electrical impedance images [5, 6]. The mammographer has the following key characteristics:


Thus, the method of back-projection may be applied to the case when the power lines of electric field intensity are used instead of rays, the former being estimated for the case of homogeneous conductivity distribution [7].

A two-dimensional image is a good way to illustrate the method. The figure on the left (Figure 2) shows systems of equipotential lines for two injecting electrodes *i(A)*и*i(B)*.

**Figure 2.** Equipotential lines – pseudorays and image reconstruction.

Here, we present the case of a diametrical location of an injection electrode pair (the opposing referent electrode is not shown in the figure). The red spot stands for a non-homogeneous area in the object. The red dots on the surface of the object show pairs of measuring electrodes in which the electric potential differences u(A) and u(B) diverge maximally as compared to areas with a homogeneous distribution of electrical conductivity. The figure on the right demon‐ strates the principle of image reconstruction based on measurement results obtained from two injecting electrodes. The light blue lines show the noise components of the reconstructed image. The dark blue cross-section is a constructed image of the non-homogeneous area. The image will be of better quality when reconstructed by measurement results from a greater number of injecting electrodes.

The key differences between the three-dimensional image and the two-dimensional image involve the following:


Thus, the back-projection method in a three-dimensional image consists of the usual projection of the superficial potential difference change (as compared to the homogeneous tissues) into the object along the equipotential surfaces, but with a weight depending on the distance between the projected point and the point at which conductivity reconstruction occurs [7].

#### **3. Elements of electrical impedance in a mammographic scheme**

Here, we present a description of breast anatomic structures and comment on the concept of a mammographic scheme in terms of electrical impedance mammography [8].

#### **3.1. Terminology**


In the anatomy of the mammary gland, the following elements of the electrical mammographic scheme can be marked out: capsule, connective tissue carcass, parenchyma, secretory reservoir, nipple and areola.

#### **3.2. Capsule**

**Figure 2.** Equipotential lines – pseudorays and image reconstruction.

number of injecting electrodes.

4 Mammography Techniques and Review

involve the following:

Here, we present the case of a diametrical location of an injection electrode pair (the opposing referent electrode is not shown in the figure). The red spot stands for a non-homogeneous area in the object. The red dots on the surface of the object show pairs of measuring electrodes in which the electric potential differences u(A) and u(B) diverge maximally as compared to areas with a homogeneous distribution of electrical conductivity. The figure on the right demon‐ strates the principle of image reconstruction based on measurement results obtained from two injecting electrodes. The light blue lines show the noise components of the reconstructed image. The dark blue cross-section is a constructed image of the non-homogeneous area. The image will be of better quality when reconstructed by measurement results from a greater

The key differences between the three-dimensional image and the two-dimensional image

**•** The projection of an isolated non-homogeneity along the equipotential lines onto the object's surface has a spherical shape with a centre in the injecting electrode and a radius equal to

**•** The change in the superficial potential differences (as compared to the homogeneous tissues) is generally concentrated in the circle shown above, and decreases as the distance between

Thus, the back-projection method in a three-dimensional image consists of the usual projection of the superficial potential difference change (as compared to the homogeneous tissues) into the object along the equipotential surfaces, but with a weight depending on the distance between the projected point and the point at which conductivity reconstruction occurs [7].

**•** Equipotential surfaces have a spherical shape with a centre in the injecting electrode;

the distance from the injecting electrode to the non-homogeneity;

the point in the circle and the non-homogeneity becomes greater.

The mammary gland capsule consists of leaves of the superficial fascia and subcutaneous fat enveloping the mammary gland on all sides.

**Figure 3.** Electrical impedance tomogram of the mammary gland (EIM). Seven planes of scanning. In the sixth and the seventh scans, a retromammary tissue is visualized.

In the image, the capsule forms an important element of the mammographic scheme – the mammary gland contour – with fatty tissue intimately embracing the body of the mammary gland (capsulaadiposamammae). The front layer of the capsule ends in the behind-areola area where the ends of the milk duct are situated. In figure 3, one can see an electrical impedance tomogram where fatty tissue on the periphery of the mammary gland is represented in the shape of a hyperimpedance contour.

#### **3.3. Carcass**

The mammary gland is enclosed in a connective tissue capsule, sending septa into the strata of the gland. The septa consist of tender fibrillary tissues and are situated between the glandular elements. The septa that form the connective tissue framework of the mammary gland are in hyperimpedance and radically diverge from the areola. On the periphery of the mammary gland, the adipose capsule stands out in the form of a hyperimpedance contour (Figure 4).

**Figure 4.** EIM. Seven planes of scan. In the first, second and third scans, the connective tissues' septa are visualized along with the radially diverging areolae.

#### **3.4. Parenchyma**

The parenchyma is a structural component of the mammary gland made of alveotubular glands and the connective tissue stroma. Alveotubular glands consisting of ductal and secretary epithelium are gathered into small lobules forming lobes. The connective tissue stroma is represented by a small number of cells, gentle tissues and the base material. Figure 5 shows an electrical impedance tomogram, where the parenchyma is represented in the form of isoimpedance areas visible between the septa.

**Figure 5.** EIM. Seven scan planes. The parenchyma is visualized in the form of isoimpedance areas between the con‐ nective tissue septa.

#### **3.5. Secretory reservoir**

gland (capsulaadiposamammae). The front layer of the capsule ends in the behind-areola area where the ends of the milk duct are situated. In figure 3, one can see an electrical impedance tomogram where fatty tissue on the periphery of the mammary gland is represented in the

The mammary gland is enclosed in a connective tissue capsule, sending septa into the strata of the gland. The septa consist of tender fibrillary tissues and are situated between the glandular elements. The septa that form the connective tissue framework of the mammary gland are in hyperimpedance and radically diverge from the areola. On the periphery of the mammary gland, the adipose capsule stands out in the form of a hyperimpedance contour

**Figure 4.** EIM. Seven planes of scan. In the first, second and third scans, the connective tissues' septa are visualized

The parenchyma is a structural component of the mammary gland made of alveotubular glands and the connective tissue stroma. Alveotubular glands consisting of ductal and secretary epithelium are gathered into small lobules forming lobes. The connective tissue stroma is represented by a small number of cells, gentle tissues and the base material. Figure 5 shows an electrical impedance tomogram, where the parenchyma is represented in the form

**Figure 5.** EIM. Seven scan planes. The parenchyma is visualized in the form of isoimpedance areas between the con‐

shape of a hyperimpedance contour.

6 Mammography Techniques and Review

along with the radially diverging areolae.

of isoimpedance areas visible between the septa.

**3.4. Parenchyma**

nective tissue septa.

**3.3. Carcass**

(Figure 4).

Before opening into the nipple milk ducts enlarge and form a lactiferous sinus (sinus lactiferi). It which accumulates secretion or milk produced in alveoli and is characterized by low electrical impedance. There are 15-25 such sinuses in the area behind the nipple. In Figure 6, one can see an electrical impedance tomogram where the lactiferous sinus zone visualizes itself as a vast hypoimpedance area, situated in the centre of the mammogram.

**Figure 6.** EIM. Seven scan planes. In the centre of the tomogram there is a hypoimpedance area corresponding to the location of the lactiferous sinuses.

#### **3.6. Nipple and areola**

The nipple of the mammary gland and the mammary areola are areas of hairless pigmented epidermis. The nipple consists of the lactiferous sinus, an extension into which flows the excretory ducts of the mammary gland lobules, with fibrous tissue around them, and a large number of sebaceous glands. The absence in the nipple of the perspiratory gland's excretory ducts defines its high electrical impedance. The nipple is visible in the first and second scans of the electrical impedance tomogram, taking the form of a linear hyperimpedance formation located in the centre, near the zone of the lactiferous sinuses (Figure 7).

**Figure 7.** EIM. Seven scan planes. Outer segment. In the center of the 1st and the 2nd scans a linear hyperimpedance formation typical of the nipple is visualized.

The derma of the mammary areola contains a large number of pigmentary cells which condition its high electrical impedance. On the electrical impedance tomogram, the mammary areola is visualized as a hyperimpedance formation of a round shape around the area of the lactiferous sinuses (Figure 8).

**Figure 8.** Impedance area in the centre of the tomogram corresponding to the placement of the areola.

#### **4. Normal electrical impedance mammographic scheme**

As has been mentioned above, the breast may have a variable appearance on the tomogram. This is why the electrical impedance mammographic scheme depends on the type of breast structure. Earlier, we pointed out five types of breast structure from the point of view of electrical impedance mammography execution (Table 1).


**Table 1.** Mammary gland structure from the perspective of electrical impedance mammography execution, and breast density types according to the classification of the American College of Radiology (ACR).

#### **4.1. Normal mammographic scheme with a ductal-type mammary gland structure**

The prevailing ductal component in the parenchyma structure is typical of women during the early reproductive period. The high density of the ductal component is a significant obstacle in the way of electric charges, which adds specific features to the electrical impedance image: the prevalence of darker tones of greyscale; well-defined anatomical landmarks; low values of electrical conductivity. In the normal electrical impedance mammographic scheme, the anatomical landmarks are distinct and well-defined.

**•** The mammary gland contour is even and regular, without extrusions, retractions or unilateral thickening (Figure 9).

**Figure 8.** Impedance area in the centre of the tomogram corresponding to the placement of the areola.

As has been mentioned above, the breast may have a variable appearance on the tomogram. This is why the electrical impedance mammographic scheme depends on the type of breast structure. Earlier, we pointed out five types of breast structure from the point of view of

*EIM classification ACR classification*

Predominantly fat. Under 25% of the tissue is

Fat with some fibroglandular tissue. 25-50% of the

50-75% of the tissue is represented by the parenchyma.

tissue is represented by the parenchyma.

75-100% of the parenchyma tissue.

Heterogeneously dense.

Extremely dense.

**Table 1.** Mammary gland structure from the perspective of electrical impedance mammography execution, and breast

The prevailing ductal component in the parenchyma structure is typical of women during the early reproductive period. The high density of the ductal component is a significant obstacle

**4.1. Normal mammographic scheme with a ductal-type mammary gland structure**

represented by the parenchyma.

**4. Normal electrical impedance mammographic scheme**

electrical impedance mammography execution (Table 1).

Type Iа Amorphous type of mammary gland structure.

Type Ib Mixed type of mammary gland structure with amorphous component predominance.

Type II Mixed type of mammary gland structure.

Type III Mixed type of mammary gland structure with an

Type IV Acinal/ductal type of mammary gland structure. Extremely high density of the acinal/duct

acinal/ductal component predominance. High density of the acinal/ductal component.

density types according to the classification of the American College of Radiology (ACR).

IC = above 0.66 \*IC – index conductivity

8 Mammography Techniques and Review

IC=0.57-0.65.

IC=0.30-0.56

IC=0.22-0.29.

component. IC <0.22.

**Figure 9.** EIM. Seven scan planes. The mammary gland contour is distinct and without deformation.

**•** The areola is in the centre of the image, non-displaced, non-deformed, non-fragmented and coloured in black tones of greyscale (Figure 10).


**Figure 10.** EIM. Seven scan planes. Central location of the areola. The areola is black.

**•** The mammary gland structure is represented by the parenchyma and the septa without focal changes (Figure 11).

**Figure 11.** EIM. Seven scan planes. The parenchyma is without focal changes.

**•** The relative electrical conductivity of the left and right breasts is within normal range – the divergence of the electrical conductivity histograms is less than 20% (Figure 12).

**Figure 12.** EIM. Seven scan planes. The top row shows images of the left breast, the bottom shows images of the right breast. The middle row presents a comparison of conductivity distribution histograms.

#### **4.2. Normal mammographic scheme for amorphous-type breast structures**

The predominance of an amorphous substance in the breast parenchyma structure can be observed among peri-menopausal and menopausal women. A considerable amount of fibrous friable connecting tissue in the mammary gland with a predominance of the basic substance helps the electric charge to pass. This conditions the characteristic features of the electrical impedance image: the predominance of lighter tones of greyscale; an absence of anatomic landmarks; high values of electrical conductivity. A normal electrical impedance mammo‐ graphic scheme is expected to have all the characteristic features.

**•** The mammary gland contour is even and regular, without extrusions, retractions or unilateral thickening (Figure 13).

**Figure 13.** EIM. Seven scan planes. The mammary gland contour is distinct and without deformation.

**Figure 11.** EIM. Seven scan planes. The parenchyma is without focal changes.

10 Mammography Techniques and Review

**•** The relative electrical conductivity of the left and right breasts is within normal range – the

**Figure 12.** EIM. Seven scan planes. The top row shows images of the left breast, the bottom shows images of the right

The predominance of an amorphous substance in the breast parenchyma structure can be observed among peri-menopausal and menopausal women. A considerable amount of fibrous friable connecting tissue in the mammary gland with a predominance of the basic substance helps the electric charge to pass. This conditions the characteristic features of the electrical impedance image: the predominance of lighter tones of greyscale; an absence of anatomic landmarks; high values of electrical conductivity. A normal electrical impedance mammo‐

**•** The mammary gland contour is even and regular, without extrusions, retractions or

breast. The middle row presents a comparison of conductivity distribution histograms.

graphic scheme is expected to have all the characteristic features.

unilateral thickening (Figure 13).

**4.2. Normal mammographic scheme for amorphous-type breast structures**

divergence of the electrical conductivity histograms is less than 20% (Figure 12).

**•** The areola is in the centre of the image, non-displaced, non-deformed, non-fragmented and coloured in white tones of greyscale (Figure 14).

**Figure 14.** EIM. Seven scan planes. Central location of the areola. The areola is white.

**•** The mammary gland parenchyma is visualized as an unstructured mass, without septa and without focal changes (Figure 15).

**Figure 15.** EIM. Seven scan planes. The parenchyma is without focal changes.

**•** The relative electrical conductivity of the left and right breasts is within normal range – the divergence of electrical conductivity histograms is less than 20% (Figure 16).

**Figure 16.** EIM. Seven scan planes. The top row shows images of the left breast, the bottom – of the right. The middle row demonstrates the comparison of conductivity distribution histograms.

#### **4.3. Normal mammographic scheme for mixed-type breast structures**

A mixed-type breast structure combines the elements of the ductal and amorphous types. Different combinations of these structures define the differing electrical conductivity of tissues and affect the electrical impedance image. The breast parenchyma becomes 'poorer' and homogeneous.

**•** The mammary gland contour is even and regular, without extrusions, retractions or unilateral thickening (Figure 17).

**Figure 17.** EIM. Seven scan planes. The mammary gland contour is distinct and without deformation

**•** The areola is in the centre of the figure, non-displaced, non-deformed, non-fragmented and the colour is variable (Figure 18).

**Figure 18.** EIM. Seven scan planes. Central location of the areola.

**•** The relative electrical conductivity of the left and right breasts is within normal range – the

**Figure 16.** EIM. Seven scan planes. The top row shows images of the left breast, the bottom – of the right. The middle

A mixed-type breast structure combines the elements of the ductal and amorphous types. Different combinations of these structures define the differing electrical conductivity of tissues and affect the electrical impedance image. The breast parenchyma becomes 'poorer' and

**•** The mammary gland contour is even and regular, without extrusions, retractions or

**Figure 17.** EIM. Seven scan planes. The mammary gland contour is distinct and without deformation

**•** The areola is in the centre of the figure, non-displaced, non-deformed, non-fragmented and

row demonstrates the comparison of conductivity distribution histograms.

homogeneous.

12 Mammography Techniques and Review

unilateral thickening (Figure 17).

the colour is variable (Figure 18).

**4.3. Normal mammographic scheme for mixed-type breast structures**

divergence of electrical conductivity histograms is less than 20% (Figure 16).

**•** The mammary gland parenchyma is variable and without focal changes (Figure 19).


**Figure 19.** EIM. Seven scan planes. The parenchyma is without focal changes.

**•** The relative electrical conductivity of the left and right breasts is within normal range – the divergence of electrical conductivity histograms is less than 20% (Figure 20).

**Figure 20.** EIM. The top row shows images of the left breast, the bottom shows images of the right. The middle row demonstrates the comparison of the conductivity distribution histograms.

### **5. Electrical impedance mammographic scheme distortion in the presence of cancer**

Here, we present the prospective assessment results of 310 cases of verified breast cancer. Electrical impedance mammograms were received from oncological centres in Russia, Belarus, South Africa and Malaysia. As a result, the mammographic scheme distortion was revealed in 74% of cases, whereas the change in the general electrical conductivity was detected in 67% of cases. Notably, this does not depend on the tumour size.

The assessment criteria of the mammographic scheme distortion included:


#### **5.1. Mammary gland contour change**

The visual assessment of the mammary gland image should start with the analysis of the breast contour. Usually, the contour of the mammary gland is even and regular, non-deformed and without an isoimpedance structure. In the presence of breast cancer, there is a chance of contour deformation and its hyperimpedance.

#### *5.1.1. Contour deformation*

Mammary gland contour deformation is an important diagnosis criterion in the presence of certain breast diseases. More often than not, deformation is caused by volumetric processes in the mammary gland, such as cancer. Cancer infiltration of tissues causes local contour deformations in the form of extrusion or retraction. In the top row, one can see images of a malignant breast with local contour deformation in the form of extrusion. The bottom row shows a tomogram of the normal breast (Figure 21).

**Figure 21.** EIM. Seven scan planes. The top row shows images of a malignant breast.

A hyperimpedance contour with a protruding deformity is visible at 11 o'clock. The bottom row shows images of the normal breast.

#### *5.1.2. Thickening and hyperimpedance of the contour*

**5. Electrical impedance mammographic scheme distortion in the presence**

Here, we present the prospective assessment results of 310 cases of verified breast cancer. Electrical impedance mammograms were received from oncological centres in Russia, Belarus, South Africa and Malaysia. As a result, the mammographic scheme distortion was revealed in 74% of cases, whereas the change in the general electrical conductivity was detected in 67% of

The visual assessment of the mammary gland image should start with the analysis of the breast contour. Usually, the contour of the mammary gland is even and regular, non-deformed and without an isoimpedance structure. In the presence of breast cancer, there is a chance of contour

Mammary gland contour deformation is an important diagnosis criterion in the presence of certain breast diseases. More often than not, deformation is caused by volumetric processes in the mammary gland, such as cancer. Cancer infiltration of tissues causes local contour deformations in the form of extrusion or retraction. In the top row, one can see images of a malignant breast with local contour deformation in the form of extrusion. The bottom row

cases. Notably, this does not depend on the tumour size.

**1.** Alteration of the mammary gland contour

**2.** Alteration of the breast anatomy

**5.1. Mammary gland contour change**

deformation and its hyperimpedance.

shows a tomogram of the normal breast (Figure 21).

**Figure 21.** EIM. Seven scan planes. The top row shows images of a malignant breast.

*5.1.1. Contour deformation*

**3.** Local electrical impedance alterations

**4.** Change of relative electrical conductivity

The assessment criteria of the mammographic scheme distortion included:

**of cancer**

14 Mammography Techniques and Review

By contour hyperimpedance, one should understand a significant increase of electrical impedance on the periphery of the mammary gland. This phenomenon – unilateral as a rule – should be considered as a reaction of the breast tissues solely in response to a malignant process (Figure 22).

**Figure 22.** EIM. Seven scan planes. The top row shows images of a malignant breast. The bottom row shows images of the normal breast. A unilateral thickened hyperimpedance contour can clearly be seen.

In the electrical impedance image, a hyperimpedance contour is thickened and intensively black in colour. The top rows of the figures present tomograms with hyperimpedance contours of malignant breasts. In the bottom rows, tomograms of normal breasts are given. The contour hyperimpedance is revealed through the change in its colour and its significant thickening compared to the normal breast (Figure 23).

**Figure 23.** EIM. Seven scan planes. The top row shows images of a malignant breast. The bottom row shows images of the normal breast. A unilateral thickened hyperimpedance contour can be clearly seen.

#### **5.2. Breast anatomy changes**

Normally, an image of the mammary gland has an anatomy corresponding to the age-norm, with no shift of the inner structures.

In the presence of breast diseases, the breast anatomy undergoes changes. In Figures 24 and 25, we can clearly see the alteration of the mammary gland anatomy in the presence of cancer (top row) as compared to the images of the normal breast (bottom row).

**Figure 24.** EIM. Seven scan planes. The top row shows images of a malignant breast, the bottom row shows images of the normal breast. A substantial change in the anatomy of the malignant breast accompanied by numerous hyperimpe‐ dance enclosures can clearly be seen.

**Figure 25.** EIM. Seven scan planes. The top row shows images of a malignant breast, the bottom row shows images of the normal breast. No anatomic landmarks are visible in the malignant breast.

#### **5.3. Local electrical conductivity changes**

Some of the most important elements of image assessment are local electroconductivity changes outside the lacteous sinus area, and are uncommon in relation to the norm. In the presence of breast diseases, both areas with high impedance (i.e., hyperimpedance area) and areas with low impedance (i.e., hypoimpedance area) can be visualized.

#### *5.3.1. Local hyperimpedance*

**5.2. Breast anatomy changes**

16 Mammography Techniques and Review

dance enclosures can clearly be seen.

with no shift of the inner structures.

Normally, an image of the mammary gland has an anatomy corresponding to the age-norm,

In the presence of breast diseases, the breast anatomy undergoes changes. In Figures 24 and 25, we can clearly see the alteration of the mammary gland anatomy in the presence of cancer

**Figure 24.** EIM. Seven scan planes. The top row shows images of a malignant breast, the bottom row shows images of the normal breast. A substantial change in the anatomy of the malignant breast accompanied by numerous hyperimpe‐

**Figure 25.** EIM. Seven scan planes. The top row shows images of a malignant breast, the bottom row shows images of

Some of the most important elements of image assessment are local electroconductivity changes outside the lacteous sinus area, and are uncommon in relation to the norm. In the presence of breast diseases, both areas with high impedance (i.e., hyperimpedance area) and

the normal breast. No anatomic landmarks are visible in the malignant breast.

areas with low impedance (i.e., hypoimpedance area) can be visualized.

**5.3. Local electrical conductivity changes**

(top row) as compared to the images of the normal breast (bottom row).

This figure presents an image of a malignant breast. A focal change of electric conductivity in the form of a non-homogeneous hyperimpedance area with sharp contours and local contourthickening is visible at eight o'clock. Such changes are typical of an infiltrative process. The bottom row shows images of the normal breast (Figure 26).

**Figure 26.** EIM. Seven scan planes. The top row shows images of a malignant breast. A focal change of electric conduc‐ tivity in the form of a non-homogeneous hyperimpedance area is visible at eight o'clock. The bottom row shows im‐ ages of the normal breast.

#### *5.3.2. Local hypoimpedance*

Focal changes of electric conductivity in the form of a hypoimpedance area with indistinct contours are visible at 11 o'clock. This is also typical of cancer (Figure 27).

**Figure 27.** EIM. Seven scan planes. Focal changes of electric conductivity in the form of a hypoimpedance area with indistinct contours are visible at 11 o'clock.

In addition, in the presence of breast cancer, a hyperimpedance contour can be visualized around the infiltration zone. The top row shows images of the malignant breast. A focal change of electric conductivity visualizing itself as an irregularly-shaped area with a hyperimpedance contour and a hypoimpedance structure can be found at two o'clock. The bottom row shows images of the normal breast (Figure 28).

#### *5.3.3. Total hyperimpedance*

There are cases of electrical conductivity disturbance that reveal themselves as total impedance increases. This can be observed during the infiltrative oedematous form of breast cancer. In

**Figure 28.** EIM. Seven scan planes. The top row shows images of a malignant breast. A focal change of electric conduc‐ tivity visualizing itself as an irregularly-shaped area with a hyperimpedance contour and a hypoimpedance structure can be found at two o'clock. The bottom row shows images of the normal breast.

the figures, the top row presents a breast with total high impedance, which is common for cancer. In the bottom row, there is a tomogram of a normal breast (Figure 29).

**Figure 29.** EIM. Seven scan planes. The top row shows images of a malignant breast. The bottom row shows images of a normal breast. The image of the malignant breast is domineered by hyperimpedance tones due to the decrease in electric conductivity.

#### **5.4. Change of relative electrical conductivity**

Unlike other methods of visualization, the electrical impedance method makes the quantitative analysis of the image possible. The quantitative analysis of the electrical impedance image implies the assessment of the following parameters: mean electrical conductivity index; histogram of conductivity distribution; comparison with the reference data. To attribute patients to a certain class (healthy or diseased), we have applied the following: the criterion of difference in the form of distributions; the λ criterion; the Kholmogorov-Smirnov criterion (or the DX statistics); an intermediate index in Kholmogorov-Smirnov's computations. This criterion, relating to nonparametric criteria, allows the assessment of the statistical significance of divergences in the distribution of any mark of the norm or pathology, including the electrical conductivity distribution on electrical impedance tomograms. The Dx statistics permit the assessment of the surface of one of the distributions which is not common with the second distribution. The Dx value reflects the share of observations or data which distinguish the experience (patient) from the control (norm) group. This value is important, both for the substantiation of the diagnosis and the assessment of the information value of the index. A high informative value of differences revealed allows the attribution of the patient to one class or another (e.g., norm or cancer). In order to determine the informative capacity of distribution divergence, we used the information measure of S. Kullback, which demonstrated the informativity of the applied Dx statistics and the input of this index in disease diagnosis, for example, cancer. The assessment of the distribution divergence yielded results which are in direct dependency on the value according to S. Kullback [9] (Table 2).


**Table 2.** Informative content and accuracy for different values (%) of conductivity distribution diversion.

the figures, the top row presents a breast with total high impedance, which is common for

**Figure 28.** EIM. Seven scan planes. The top row shows images of a malignant breast. A focal change of electric conduc‐ tivity visualizing itself as an irregularly-shaped area with a hyperimpedance contour and a hypoimpedance structure

**Figure 29.** EIM. Seven scan planes. The top row shows images of a malignant breast. The bottom row shows images of a normal breast. The image of the malignant breast is domineered by hyperimpedance tones due to the decrease in

Unlike other methods of visualization, the electrical impedance method makes the quantitative analysis of the image possible. The quantitative analysis of the electrical impedance image implies the assessment of the following parameters: mean electrical conductivity index; histogram of conductivity distribution; comparison with the reference data. To attribute patients to a certain class (healthy or diseased), we have applied the following: the criterion of difference in the form of distributions; the λ criterion; the Kholmogorov-Smirnov criterion (or the DX statistics); an intermediate index in Kholmogorov-Smirnov's computations. This criterion, relating to nonparametric criteria, allows the assessment of the statistical significance of divergences in the distribution of any mark of the norm or pathology, including the electrical conductivity distribution on electrical impedance tomograms. The Dx statistics permit the assessment of the surface of one of the distributions which is not common with the second distribution. The Dx value reflects the share of observations or data which distinguish the

cancer. In the bottom row, there is a tomogram of a normal breast (Figure 29).

can be found at two o'clock. The bottom row shows images of the normal breast.

electric conductivity.

18 Mammography Techniques and Review

**5.4. Change of relative electrical conductivity**

In the presence of diseases, mammary gland histogram offset takes place (Figure 30).


**Figure 30.** Upper row – EIM. Seven scan planes. Breast cancer. Bottom row - EIM. Seven scan planes. Healthy gland. The second row shows the divergence between the histograms of the electrical conductivity distribution of the affected and healthy glands.Table 3, below, contains the data on the relative conductivity of malignant breasts as compared to breasts undergoing benign changes, normal breasts and breasts with different anatomies.


**Table 3.** Relative breast conductivity for the norm and pathology.

As the oncological process develops, the overall and local conductivity changes. At the same time, the distortion of the normal mammographic scheme can already be observed at the initial stages of the disease. It is for this reason that the given criterion is included in the EIM scale of breast cancer diagnosis (Table 4).



**Table 4.** Diagnostic criteria for differentiating volumetric lesions in electrical impedance mammography.

#### **6. Conclusion**

**Number of patients**

20

32

of breast cancer diagnosis (Table 4).

101 (33%)

157 (98%)

18 (90%)

28 (88%)

59 (87%)

**Table 3.** Relative breast conductivity for the norm and pathology.

67 (22%)

4 (2%)

1 (5%)

2 (6%)

7 (10%)

**Diagnostic criteria Electrical impedance**

As the oncological process develops, the overall and local conductivity changes. At the same time, the distortion of the normal mammographic scheme can already be observed at the initial stages of the disease. It is for this reason that the given criterion is included in the EIM scale

Cancer 310

20 Mammography Techniques and Review

Healthy 161

Healthy Acinal/ductaltype mammary gland

Benign 68

structure.

Healthy Amorphous-type mammary gland structure.

**Shape** • round, oval • lobular, irregular

**Contour** • no • sharp

• hyperimpedance, indistinct

**Internal electrical structure** • hyperimpedance (ICroi<ICav+2std) • isoimpedance ICroi=ICav±2std) • hypoimpedance (ICroi>ICav+2std) • animpedance (ICroi>ICav+3std)

• structure alteration/displacement • thickening/extrusion/retraction

**Surrounding tissues** • preserved

**Relative conductivity (diseased - healthy mammary gland)**

44 (14%)

1 (5%)

2 (6%)

2 (3%)

**< 20% 20-30% 30-40% 40-50% 50-60% >60%**

37 (12%)

0 0 0 0

26 (8%)

0 0 0

0 0 0

0 0 0

**mammography points**

1 2

0 1 2

0 1 2

35 (11%)

> The alteration of the normal mammographic scheme along with the abnormal changes of local conductivity is one of the early diagnostic criteria for breast cancer.

> Electrical impedance mammography represents a method allowing the formation of observa‐ tion and risk groups for the development of breast cancer through the use of data on relative electrical conductivity and the age-related electrical conductivity scale.

> Complications of the oncological process with oedema are a serious obstacle for X-ray and US diagnostics. Unlike the aforementioned methods of diagnosis, electrical impedance tomogra‐ phy makes it possible to visualize pathophysiological changes in such states as cancer, mastitis and lymphostasis, etc.

#### **Author details**

Alexander Karpov1\*, Andrey Kolobanov2 and Marina Korotkova1


#### **References**

[1] Boyd NF, Byng JW, Jong RA, Fishell EK, Little LE, Miller AB, Lockwood GA, Tritcher DL, Yaffe MJ. Quantitative classification of mammographic densities and breast can‐ cer risk: results from Canadian National Breast Screening Study. *J. Natl. Cancer Inst.* 1995; 87; 670-675.


### **Chapter 2**

## **Ultrasound Axillary Imaging**

#### Nastasia Serban

[2] Wolfe JN, Saftias AF, Salane M. Mammographic parenchymal patterns and quantita‐ tive evaluation of mammographic densities: a case-control study. *AJR Am J Roentgen‐*

[3] American College of Radiology. BI-RADS Breast Imaging Reporting and Data Sys‐

[5] Cherepenin V, Karpov A, Korjenevsky A, Kornienko V, Mazaletskaya A, Mazurov D. A 3D electrical impedance tomography (EIT) system for breast cancer detection.

[6] Karpov A, Korjenevsky A, Mazurov D, Mazaletskaya A. 3D Electrical Impedance Scanning of Breast Cancer. *World Congress on Medical Physics and Biomedical Engineer‐*

[7] Dunaeva O, Gerasimov D, Karpov A, Machin M, Tchayev A, Tsofin Yu, Tsyplyonkov V. Using Backprojection Algorithm for 3D Image Reconstruction in EIT. *World Con‐*

[8] Korotkova M, Karpov A. Procedure for assessment of the mammary gland electrical impedance images. *XIII international conference on electrical bio-impedance*. Graz, Aus‐

[9] Gubler E. Quantitative methods for analysis and identification of pathology. Lenin‐

*gress on Medical Physics and Biomedical Engineering*, Munich, Germany, 2009.

*ol.* 1987, 148, 1087-1092.

22 Mammography Techniques and Review

*ing, Chicago*, 2000, 62.

tria, 2007.

grad, 1978.

tem. Virginia: Reston; 2003.

[4] Electrical Impedance Tomography. IOP, 2005.

*Physiological Measurement.* 2001, 22, 9-18.

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/59730

#### **1. Introduction**

The most significant prognostic factors in breast cancer are the tumoral diameter, tumor grading and the status of the axillary lymph nodes. The presence of nodal metastases decreases 5-year survival by approximately 40% compared to node-negative patients, in reference [1]. Lymph node status is of particular value in choosing further therapy. Lymph node metastatic disease is an indication for skipping sentinel node biopsy (SLNB) (and proceeding to complete axillary dissection) and/or for adjuvant systemic chemotherapy, which may be of benefit if administered as preoperative treatment.

#### **2. The anatomy of the axillary lymph node**

The anatomy of the axillary lymph node includes the cortex and the medulla. The highfrecquency probes allow the differentiation of the central echogenic hilum and the peripheral hypoechoic cortex. The cortex, which includes the marginal sinus and the lymphoid follicles is hypoechoic and thin, and has a fusiform shape with smooth edge. The hilum is the hyperechoic, its echogenity being attributable to multiple reflective interfaces of blood vessels, fat, and the central sinus, in reference [2,3].

Carcinoma from the breast enters the lymph node via the afferent lymphatics, penetrates the capsule, and enters the subcapsular sinus, in reference [4]. Metastatic cells firstly stop in the periphery (cortex) of the nodes, causing cortical enlargement. Then generalized cortical enlargement and destruction of the nodal architecture occurs, with compression and, eventu‐ ally, loss of the hilum, in reference [2].

© 2015 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### **3. Assessment of axillary lymph nodes status**

Grossly involvement of axillary lymph nodes can be detected by clinical examination, ultrasound or axilla MRI. However, introduction of screening mammography led to earlier diagnosis of breast cancer, in which axillary involvement is frequently absent. The challenge of imaging technique is to differentiate the normal lymph nodes from the nodes with minimal metastatic disease, which do not change the size and shape of the lymph node, in patients with small primary breast tumors.

The "golden standard" for axillary lymph node status is pathological examination of lymph nodes. There are three possibilities to obtain information regarding the axillary lymph nodes status: complete axillary lymph node dissection, biopsy of the sentinel lymph node (SLN) and pretreatment imaging of the axillary lymph nodes, associated or not with fine-needle aspira‐ tion cytology or core biopsy of the suspicious nodes.

Complete lymph node dissection represents the classic approach that allows pathological examination of all the lymph nodes in the axilla. However, complete axillary lymph node dissection is accompanied by complications like seroma formation, numbness, limitation of shoulder movement, and lymphedema, in reference [5].

SLN biopsy (SLNB) represents the biopsy of that lymph node, which first collects the lymph from the breast. It is a surgical procedure, requiring preoperative administration of a dye and/ or radionuclide tracer.

Pretreatment imaging of the axillary lymph nodes must closely match the pathological findings in order to have any value for clinical decision making. Many studies suggest that patients with axillary involvement may benefit from preoperative systemic treatment. Imaging techniques for axilla include ultrasound, MRI enhanced or nonenhanced, FDG-PET scan, 99mTc-sestamibi scintigraphy.

#### **4. Ultrasound evaluation of axilla**

The most available imaging technique for axilla is ultrasound. Ultrasound has two roles in visualizing the axilla: a) to characterize the abnormal lymph nodes, either identified by US or by clinical examination or other imaging technique and b) to help axillar SLN identification. In both circumstances, ultrasound helps the biopsy of the nodes.

Afferent lymphatic channels enter a node through the periphery of the cortex, so the malignant cells travelling the lymphatic vessel will first stop in the cortical region of the lymph node. Most of the US signs of lymph nodes metastasis will refer to the abnormalities of the cortex. Subtle abnormalities of the cortex can indicate early metastatic involvement.

For the assessment of a lymph node by US, quantitative or qualitative methods have been used.

#### **4.1. The qualitative features of a metastatic lymph node on US**

**3. Assessment of axillary lymph nodes status**

tion cytology or core biopsy of the suspicious nodes.

shoulder movement, and lymphedema, in reference [5].

small primary breast tumors.

24 Mammography Techniques and Review

or radionuclide tracer.

99mTc-sestamibi scintigraphy.

**4. Ultrasound evaluation of axilla**

In both circumstances, ultrasound helps the biopsy of the nodes.

Subtle abnormalities of the cortex can indicate early metastatic involvement.

Grossly involvement of axillary lymph nodes can be detected by clinical examination, ultrasound or axilla MRI. However, introduction of screening mammography led to earlier diagnosis of breast cancer, in which axillary involvement is frequently absent. The challenge of imaging technique is to differentiate the normal lymph nodes from the nodes with minimal metastatic disease, which do not change the size and shape of the lymph node, in patients with

The "golden standard" for axillary lymph node status is pathological examination of lymph nodes. There are three possibilities to obtain information regarding the axillary lymph nodes status: complete axillary lymph node dissection, biopsy of the sentinel lymph node (SLN) and pretreatment imaging of the axillary lymph nodes, associated or not with fine-needle aspira‐

Complete lymph node dissection represents the classic approach that allows pathological examination of all the lymph nodes in the axilla. However, complete axillary lymph node dissection is accompanied by complications like seroma formation, numbness, limitation of

SLN biopsy (SLNB) represents the biopsy of that lymph node, which first collects the lymph from the breast. It is a surgical procedure, requiring preoperative administration of a dye and/

Pretreatment imaging of the axillary lymph nodes must closely match the pathological findings in order to have any value for clinical decision making. Many studies suggest that patients with axillary involvement may benefit from preoperative systemic treatment. Imaging techniques for axilla include ultrasound, MRI enhanced or nonenhanced, FDG-PET scan,

The most available imaging technique for axilla is ultrasound. Ultrasound has two roles in visualizing the axilla: a) to characterize the abnormal lymph nodes, either identified by US or by clinical examination or other imaging technique and b) to help axillar SLN identification.

Afferent lymphatic channels enter a node through the periphery of the cortex, so the malignant cells travelling the lymphatic vessel will first stop in the cortical region of the lymph node. Most of the US signs of lymph nodes metastasis will refer to the abnormalities of the cortex.

For the assessment of a lymph node by US, quantitative or qualitative methods have been used.

The qualitative features of a metastatic lymph node on US include shape (round morphology), asymmetric cortical thickening (Figure 1), loss of central hilum, loss or compression of the hyperechoic medullary region (Figure 2), relationship with neighbouring lymph nodes, leftto-right asimetry and increased peripheral blood flow. Even more, lymph nodes can exhibit ultrasound malignancy signs characteristic to the primary breast tumor, such as angular margins, "taller than wide" (Figure 3) or they appears intensely vascularisated on Doppler ultrasound. Reactive changes associated with inflamation produce an increase of blood flux in the preexisting blood vessels, but do not generate new vessel formation or vessels that can penetrate the capsule. Metastatic disease stimulates new vessels formation, so Doppler examination can reveal an intense Doppler signal inside the lymph node or blood vessel penetrating the capsule. The qualitative features of a metastatic lymph node on US include shape (round morphology), asymmetric cortical thickening (Figure 1), loss of central hilum, loss or compression of the hyperechoic medullary region (Figure 2), relationship with neighbouring lymph nodes, left-to-right asimetry and increased peripheral blood flow. Even more, lymph nodes can exhibit ultrasound malignancy signs characteristic to the primary breast tumor, such as angular margins, "taller than wide" (Figure 3) or they appears intensely vascularisated on Doppler ultrasound. Reactive changes associated with inflamation produce an increase of blood flux in the preexisting blood vessels, but do not generate new vessel formation or vessels that can penetrate the capsule. Metastatic disease stimulates new vessels formation, so Doppler examination can reveal an intense Doppler signal inside the lymph node or blood vessel penetrating the capsule. The qualitative features of a metastatic lymph node on US include shape (round morphology), asymmetric cortical thickening (Figure 1), loss of central hilum, loss or compression of the hyperechoic medullary region (Figure 2), relationship with neighbouring lymph nodes, left-to-right asimetry and increased peripheral blood flow. Even more, lymph nodes can exhibit ultrasound malignancy signs characteristic to the primary breast tumor, such as angular margins, "taller than wide" (Figure 3) or they appears intensely vascularisated on Doppler ultrasound. Reactive changes associated with inflamation produce an increase of blood flux in the preexisting blood vessels, but do not generate new vessel formation or vessels that can penetrate the capsule. Metastatic disease stimulates new vessels formation, so Doppler examination can reveal an intense Doppler signal inside the lymph node or blood vessel penetrating the capsule.

Figure 1. Left: Metastatic lymph node, with a round shape. Invasive ductal carcinoma (primary tumor of 1.8 cm, three SLNs identified, all of them metastatic, and complete axillary dissection). Right: asymmetrical cortical thickening in a metastatic axillary lymph **Figure 1.** Left: Metastatic lymph node, with a round shape. Invasive ductal carcinoma (primary tumor of 1.8 cm, three SLNs identified, all of them metastatic, and complete axillary dissection). Right: asymmetrical cortical thickening in a metastatic axillary lymph node (with normal the cortex-hilum area ratio). (primary tumor of 1.8 cm, three SLNs identified, all of them metastatic, and complete axillary dissection). Right: asymmetrical cortical thickening in a metastatic axillary lymph node (with normal the cortex-hilum area ratio).

Figure 1. Left: Metastatic lymph node, with a round shape. Invasive ductal carcinoma

Figure 2. Metastatic lymph node. Left: loss of the hyperechoic medullary region. Right: on Figure 2. Metastatic lymph node. Left: loss of the hyperechoic medullary region. Right: on power Doppler examination, this node is displaying both central and peripheral blood flow. **Figure 2.** Metastatic lymph node. Left: loss of the hyperechoic medullary region. Right: on power Doppler examina‐ tion, this node is displaying both central and peripheral blood flow.

power Doppler examination, this node is displaying both central and peripheral blood flow.

Figura 3. Left, metastatic lymphadenopathy, exhibiting "taller than wide" property and angular margins. Pathology showed capsular efraction. Right, gigantic metastatic lymphadenopathy with increased vascularisation and intact capsule. **Figure 3.** Left, metastatic lymphadenopathy, exhibiting "taller than wide" property and angular margins. Pathology showed capsular efraction. Right, gigantic metastatic lymphadenopathy with increased vascularisation and intact cap‐ sule.

#### **4.2. The quantitative indicators of a metastatic lymph node on US**

**4.2 The quantitative indicators of a metastatic lymph node on US**  The quantitative indicators of a metastatic lymph node on US include the size (Figure 4, left), maximum thickness of the cortex, in reference [6] (Figure 4, right; Figure 5), the cortexhilum (CH) area ratio, in reference [7] (Figure 5), the longitudinal-transverse (LT) axis ratio, The quantitative indicators of a metastatic lymph node on US include the size (Figure 4, left), maximum thickness of the cortex, in reference [6] (Figure 4, right; Figure 5), the cortex-hilum (CH) area ratio, in reference [7] (Figure 5), the longitudinal-transverse (LT) axis ratio, in reference [7], the number of peripheral blood vessels.

in reference [7], the number of peripheral blood vessels. Lymph nodes can be enlarged, either by metastatic disease or reactive changes, including fat degeneration. Reactive changes in lymph nodes increase all dimensions, keeping the eliptical shape and a normal cortical/medullar index. Metastatic disease will increase the lymph node small diameter, will replace the medullar hyperechogenic region with hypoechogenic tumoral tissue, so the maximum thickness of the cortex and cortex-hilum area ratio will increase, eventually completely destroying the hilum (absent hilum). Lymph nodes can be enlarged, either by metastatic disease or reactive changes, including fat degeneration. Reactive changes in lymph nodes increase all dimensions, keeping the eliptical shape and a normal cortical/medullar index. Metastatic disease will increase the lymph node small diameter, will replace the medullar hyperechogenic region with hypoechogenic tumoral tissue, so the maximum thickness of the cortex and cortex-hilum area ratio will increase, eventually completely destroying the hilum (absent hilum).

A small study of the author, in reference [8], evaluating 21 consecutive breast cancer patients, in which SLNB was performed, suggest that ultrasound size > 1 cm of lymph nodes correlates with invasion of SLN (Figure 4, left). A small study of the author, in reference [8], evaluating 21 consecutive breast cancer patients, in which SLNB was performed, suggest that ultrasound size > 1 cm of lymph nodes correlates with invasion of SLN (Figure 4, left).

Cortical region evaluation is more important in lymph nodes assessment than size. The absolute cortical thickness is predictive for axillary metastatic disease (Figure 4, right), a cortical thickness more than 2.5 mm being associated in 70 percent of cases with lymph node metastasis (Cho N et al, 2009, in reference [6]).

Song SE et al, in reference [7], evaluated the diagnostic performance for their own positive criteria for lymph node metastasis, such as CH area ratio >2 (Figure 5), LT axis ratio <2 or peripheral type of vascularisation on power Doppler imaging. They found that the sensitivity of the CH area ratio was superior to that of the LT axis ratio (94.1% vs. 82.3%, p=0.031) and to that of the peripheral blood flow pattern (94.1% vs. 29.4%, p=0.009) (Figure 6). For specificity, all three parameters had the same high values (89.1-95.6%; NS).

cm), in a case of advanced invasive ductal carcinoma. Core biopsy of this node showed invasive duct carcinoma. Right: maximum thickness of the cortex of 4 mm in a metastatic lymph node (invasive ductal carcinoma). **Figure 4.** Left: metastatic lymph node, showing increased size (longest dimension of 2,29 cm), in a case of advanced invasive ductal carcinoma. Core biopsy of this node showed invasive duct carcinoma. Right: maximum thickness of the cortex of 4 mm in a metastatic lymph node (invasive ductal carcinoma).

Figure 4. Left: metastatic lymph node, showing increased size (longest dimension of 2,29

Figura 3. Left, metastatic lymphadenopathy, exhibiting "taller than wide" property and angular margins. Pathology showed capsular efraction. Right, gigantic metastatic lymphadenopathy with increased vascularisation and intact capsule.

**Figure 3.** Left, metastatic lymphadenopathy, exhibiting "taller than wide" property and angular margins. Pathology showed capsular efraction. Right, gigantic metastatic lymphadenopathy with increased vascularisation and intact cap‐

The quantitative indicators of a metastatic lymph node on US include the size (Figure 4, left), maximum thickness of the cortex, in reference [6] (Figure 4, right; Figure 5), the cortexhilum (CH) area ratio, in reference [7] (Figure 5), the longitudinal-transverse (LT) axis ratio,

The quantitative indicators of a metastatic lymph node on US include the size (Figure 4, left), maximum thickness of the cortex, in reference [6] (Figure 4, right; Figure 5), the cortex-hilum (CH) area ratio, in reference [7] (Figure 5), the longitudinal-transverse (LT) axis ratio, in

Lymph nodes can be enlarged, either by metastatic disease or reactive changes, including fat degeneration. Reactive changes in lymph nodes increase all dimensions, keeping the eliptical shape and a normal cortical/medullar index. Metastatic disease will increase the lymph node small diameter, will replace the medullar hyperechogenic region with hypoechogenic tumoral tissue, so the maximum thickness of the cortex and cortex-hilum

Lymph nodes can be enlarged, either by metastatic disease or reactive changes, including fat degeneration. Reactive changes in lymph nodes increase all dimensions, keeping the eliptical shape and a normal cortical/medullar index. Metastatic disease will increase the lymph node small diameter, will replace the medullar hyperechogenic region with hypoechogenic tumoral tissue, so the maximum thickness of the cortex and cortex-hilum area ratio will increase,

A small study of the author, in reference [8], evaluating 21 consecutive breast cancer patients, in which SLNB was performed, suggest that ultrasound size > 1 cm of lymph

A small study of the author, in reference [8], evaluating 21 consecutive breast cancer patients, in which SLNB was performed, suggest that ultrasound size > 1 cm of lymph nodes correlates

Cortical region evaluation is more important in lymph nodes assessment than size. The absolute cortical thickness is predictive for axillary metastatic disease (Figure 4, right), a cortical thickness more than 2.5 mm being associated in 70 percent of cases with lymph node

Song SE et al, in reference [7], evaluated the diagnostic performance for their own positive criteria for lymph node metastasis, such as CH area ratio >2 (Figure 5), LT axis ratio <2 or peripheral type of vascularisation on power Doppler imaging. They found that the sensitivity of the CH area ratio was superior to that of the LT axis ratio (94.1% vs. 82.3%, p=0.031) and to that of the peripheral blood flow pattern (94.1% vs. 29.4%, p=0.009) (Figure 6). For specificity,

area ratio will increase, eventually completely destroying the hilum (absent hilum).

**4.2 The quantitative indicators of a metastatic lymph node on US** 

**4.2. The quantitative indicators of a metastatic lymph node on US**

in reference [7], the number of peripheral blood vessels.

reference [7], the number of peripheral blood vessels.

sule.

26 Mammography Techniques and Review

nodes correlates with invasion of SLN (Figure 4, left).

metastasis (Cho N et al, 2009, in reference [6]).

with invasion of SLN (Figure 4, left).

eventually completely destroying the hilum (absent hilum).

all three parameters had the same high values (89.1-95.6%; NS).

moderate and simetrical increase of cortical thickness, but normal cortical/medullar index. Below: metastatic axillary lymph node with cortical/medullar index of 1.82 and an absolute increase of cortical thickness > 2.5 mm. **Figure 5.** Above: reactive lymphadenopathy in patient with postpartum mastitis. There is moderate and simetrical in‐ crease of cortical thickness, but normal cortical/medullar index. Below: metastatic axillary lymph node with cortical/ medullar index of 1.82 and an absolute increase of cortical thickness > 2.5 mm.

Song SE et al, in reference [7], evaluated the diagnostic performance for their own positive criteria for lymph node metastasis, such as CH area ratio >2 (Figure 5), LT axis ratio <2 or peripheral type of vascularisation on power Doppler imaging. They found that the sensitivity of the CH area ratio was superior to that of the LT axis ratio (94.1% vs. 82.3%, p= 0.031) and to that of the peripheral blood flow pattern (94.1% vs. 29.4%, p=0.009) (Figure 6).

For specificity, all three parameters had the same high values (89.1-95.6%; NS).

Figure 5. Above: reactive lymphadenopathy in patient with postpartum mastitis. There is

**Figure 6.** Metastatic axillary lymph node with LT axis ratio of 1.50; blood flow was absent on power Doppler.

#### **4.3. Sonoelastography**

Sonoelastography can be added to axillary lymph nodes ultrasound evaluation for further increase the precision of identification of metastatic lymph nodes. At present, there are not many studies trying to establish the place of sonoelastography in evaluation of axillary lymph nodes status. Choi (2011, 64 patients, in reference [9]), Taylor (2011, 50 patients, in reference [10]), Wojcinski (2012, 180 patients, in reference [11]) found that sonoelastography is capable of detecting elasticity differences between the cortex and medulla, and between metastatic and healthy LNs.

Wojcinski et al (2012) found that the highest sensibility (73.3%) is obtained when cortex >3mm in B-mode OR blue cortex in the elastogram, while, when these two features are found together (cortex >3mm in B-mode AND blue cortex in the elastogram (Figure 7)), the highest specificity is obtained (99.3%).

#### **4.4. The role of ultrasound in sentinel lymph node identification and biopsy**

Ultrasound has a role in sentinel lymph node identification. With introduction of indocyanine green for sentinel lymph node biopsy (SLNB), Tagaya et al (2010) were able to visualize the fluorescence of lymphatic vessels on the skin. The authors performed firstly intraoperative ultrasonography to identify a SLN as the first lymph node recognized during ultrasonography scanning from the edge of the breast gland in the direction of the axilla and they marked its position on the axillary skin. After indocyanine green dye injection, lymphatic ducts were visualized towards the axilla and the fluorescence stream disappeared aproximatively 1 cm

**Figure 7.** Wojcinski et al, 2012, in reference [11]. Open Access. Example for B-mode ultrasound and elastogram of a metastatic LN. In B-mode ultrasound, the cortex of the LN is slightly enlarged (maximum ~3.5mm). The predominant color of the medulla is turquoise (to green) and the cortex is mainly blue. Meeting both criteria of cortex >3mm in Bmode AND blue cortex in the elastogram, this case would be a true-positive.

before the line marked on the skin for ultrasound SLN location. In this study, the sites of skin incision for SLNB were also identical with the LN that had been demonstrated by ultrasonog‐ raphy in all patients.

Ultrasound signs of SLN involvement could be very subtle, with only a minimal focal cortical thickness increase.

By recognizing the first lymph node during scanning towards axilla (Figure 8), ultrasound may help SLN identification and decrease the operation time, an important fact because as the identification time increase, more SLNs are found.

However, in case of axillary metastases, identification of SLN may be impaired (Esen G, Gurses B, 2005, in [12]).

#### **4.5. The role of ultrasound in imagistic staging of breast cancer**

**Figure 6.** Metastatic axillary lymph node with LT axis ratio of 1.50; blood flow was absent on power Doppler.

Sonoelastography can be added to axillary lymph nodes ultrasound evaluation for further increase the precision of identification of metastatic lymph nodes. At present, there are not many studies trying to establish the place of sonoelastography in evaluation of axillary lymph nodes status. Choi (2011, 64 patients, in reference [9]), Taylor (2011, 50 patients, in reference [10]), Wojcinski (2012, 180 patients, in reference [11]) found that sonoelastography is capable of detecting elasticity differences between the cortex and medulla, and between metastatic and

Wojcinski et al (2012) found that the highest sensibility (73.3%) is obtained when cortex >3mm in B-mode OR blue cortex in the elastogram, while, when these two features are found together (cortex >3mm in B-mode AND blue cortex in the elastogram (Figure 7)), the highest specificity

Ultrasound has a role in sentinel lymph node identification. With introduction of indocyanine green for sentinel lymph node biopsy (SLNB), Tagaya et al (2010) were able to visualize the fluorescence of lymphatic vessels on the skin. The authors performed firstly intraoperative ultrasonography to identify a SLN as the first lymph node recognized during ultrasonography scanning from the edge of the breast gland in the direction of the axilla and they marked its position on the axillary skin. After indocyanine green dye injection, lymphatic ducts were visualized towards the axilla and the fluorescence stream disappeared aproximatively 1 cm

**4.4. The role of ultrasound in sentinel lymph node identification and biopsy**

**4.3. Sonoelastography**

28 Mammography Techniques and Review

healthy LNs.

is obtained (99.3%).

Ultrasound could have a role in imagistic staging of breast cancer. Knowledge of axillary lymph node involvement before surgery may allow for individualization of multimodal treatment. This may include preoperative chemotherapy, intraoperative breast radiotherapy or plastic surgery for immediate reconstruction.

The future protocols of breast cancer treatment will probably include ultrasound as a step in preoperative sentinel node mapping. Ultrasound may reveal abnormalities of axillary lymph nodes and guide biopsy of these nodes.

Figure 8. Above: preoperative assessment of axilla in a breast cancer patient with preoperative chemotherapy. The first lymph node during scanning towards axilla: sentinel lymph node with normal size (longest dimension 5.9 mm) and shape, but with smal focal cortical thickness. Below: intraoperative identification of SLN. After complete axillary **Figure 8.** Above: preoperative assessment of axilla in a breast cancer patient with preoperative chemotherapy. The first lymph node during scanning towards axilla: sentinel lymph node with normal size (longest dimension 5.9 mm) and shape, but with smal focal cortical thickness. Below: intraoperative identification of SLN. After complete axillary dis‐ section, the sentinel lymph and all the other nodes were negative.

Patients with either normal or abnormal ultrasound exams, but negative cytology, underwent sentinel node mapping. Patients with abnormal ultrasound and positive cytology proceeded to complete axillary dissection, in reference [13]. dissection, the sentinel lymph and all the other nodes were negative. By recognizing the first lymph node during scanning towards axilla (Figure 8), ultrasound may help SLN identification and decrease the operation time, an important fact because as

There are studies trying to assess the tumoral burden in patient with positive nodes. The study of Moore A et al, in [13], indicates that abnormalities limited to the lymph node cortex (Figure 9) were indicative of N1 disease. the identification time increase, more SLNs are found. However, in case of axillary metastases, identification of SLN may be impaired (Esen G, Gurses B, 2005, in [12]).

Ultrasound features of axilla, suggesting metastasis in lymph nodes, combined with results of cytology or biopsy, could modify the surgical approach to the axilla, eliminating the need for sentinel node mapping in a significant proportion of patients, in reference [13]. **4.5 The role of ultrasound in imagistic staging of breast cancer**  Ultrasound could have a role in imagistic staging of breast cancer. Knowledge of axillary lymph node involvement before surgery may allow for individualization of multimodal

Loss or compression of the hyperechoic medullary region, absence of fatty hilum, abnormal lymph node shape and increased peripheral blood flow are predictive of N2–3 disease, in reference [13] (Figure 10). treatment. This may include preoperative chemotherapy, intraoperative breast radiotherapy or plastic surgery for immediate reconstruction.

Preoperative ultrasound associated with ultrasound-guided biopsy can be used for preoper‐ ative axillary staging in patients who will be referred to preoperative systemic therapy. Study of Joh et al, in reference [14], showed that planning and initiation of preoperative systemic therapy can reliably be done using ultrasound axillary evaluation and biopsy. Patients with either normal or abnormal ultrasound exams, but negative cytology, underwent sentinel node mapping. Patients with abnormal ultrasound and positive cytology proceeded to complete axillary dissection, in reference [13]. There are studies trying to assess the tumoral burden in patient with positive nodes. The study of Moore A et al, in [13], indicates that abnormalities limited to the lymph node cortex

lymph nodes and guide biopsy of these nodes.

(Figure 9) were indicative of N1 disease.

Figure 9. Postoperative assessment of ultrasound. The first lymph node during scanning towards axilla: metastatic sentinel lymph node with normal size and shape, but with smal focal cortical thickness. After complete axillary dissection, the sentinel lymph node was the **Figure 9.** Postoperative assessment of ultrasound. The first lymph node during scanning towards axilla: metastatic sen‐ tinel lymph node with normal size and shape, but with smal focal cortical thickness. After complete axillary dissection, the sentinel lymph node was the only metastatic lymph node.

only metastatic lymph node.

Figure 10. Left and right: same case – advanced ductal carcinoma (T3 N2 M0). Three lymph nodes displaying the features of metastatic disease. Axillary metastases were confirmed by core biopsy, and the patient was referred to preoperative chemotherapy. **Figure 10.** Left and right: same case – advanced ductal carcinoma (T3 N2 M0). Three lymph nodes displaying the fea‐ tures of metastatic disease. Axillary metastases were confirmed by core biopsy, and the patient was referred to preop‐ erative chemotherapy.

#### Preoperative ultrasound associated with ultrasound-guided biopsy can be used for **4.6. Percutaneous biopsy procedures**

**4.6 Percutaneous biopsy procedures** 

biopsy.

axillary dissection.

Patients with either normal or abnormal ultrasound exams, but negative cytology, underwent sentinel node mapping. Patients with abnormal ultrasound and positive cytology proceeded

By recognizing the first lymph node during scanning towards axilla (Figure 8), ultrasound may help SLN identification and decrease the operation time, an important fact because as

Figure 8. Above: preoperative assessment of axilla in a breast cancer patient with preoperative chemotherapy. The first lymph node during scanning towards axilla: sentinel lymph node with normal size (longest dimension 5.9 mm) and shape, but with smal focal cortical thickness. Below: intraoperative identification of SLN. After complete axillary dissection, the sentinel lymph and all the other nodes were negative.

**Figure 8.** Above: preoperative assessment of axilla in a breast cancer patient with preoperative chemotherapy. The first lymph node during scanning towards axilla: sentinel lymph node with normal size (longest dimension 5.9 mm) and shape, but with smal focal cortical thickness. Below: intraoperative identification of SLN. After complete axillary dis‐

There are studies trying to assess the tumoral burden in patient with positive nodes. The study of Moore A et al, in [13], indicates that abnormalities limited to the lymph node cortex (Figure

However, in case of axillary metastases, identification of SLN may be impaired (Esen G,

Ultrasound features of axilla, suggesting metastasis in lymph nodes, combined with results of cytology or biopsy, could modify the surgical approach to the axilla, eliminating the need for

Ultrasound could have a role in imagistic staging of breast cancer. Knowledge of axillary lymph node involvement before surgery may allow for individualization of multimodal treatment. This may include preoperative chemotherapy, intraoperative breast radiotherapy

Loss or compression of the hyperechoic medullary region, absence of fatty hilum, abnormal lymph node shape and increased peripheral blood flow are predictive of N2–3 disease, in

sentinel node mapping in a significant proportion of patients, in reference [13].

**4.5 The role of ultrasound in imagistic staging of breast cancer** 

to complete axillary dissection, in reference [13].

section, the sentinel lymph and all the other nodes were negative.

or plastic surgery for immediate reconstruction.

the identification time increase, more SLNs are found.

9) were indicative of N1 disease.

Gurses B, 2005, in [12]).

30 Mammography Techniques and Review

reference [13] (Figure 10).

preoperative axillary staging in patients who will be referred to preoperative systemic therapy. Study of Joh et al, in reference [14], showed that planning and initiation of preoperative systemic therapy can reliably be done using ultrasound axillary evaluation and Unfortunetly, no imaging technique has enough reliability to attribute patients directly to complete axillary dissection, without first performing SLNB. The study of Valente SA, Sener

Unfortunetly, no imaging technique has enough reliability to attribute patients directly to complete axillary dissection, without first performing SLNB. The study of Valente SA, Sener SF et al, in [15], evaluated retrospectively 244 consecutive patients diagnosed with invasive breast carcinoma, by physical examination of the axilla, digital mammography, axillary ultrasonography, and contrast enhanced breast magnetic resonance imaging. The authors found that from the patients who had all four modalities negative, 14% were ultimately

The role of ultrasound in staging breast cancer differs with stage of disease, helping

In operable breast cancer, ultrasound helps identification of sentinel lymph node and of suspicious nodes, that warrant biopsy. Ultrasound alone has modest accuracy in detecting axillary metastasis, not being reliable, on its own, to make a decision in surgical treatment of the axilla. Ultrasound does not provide enough information to refer patients to complete

The reported a median ultrasound sensitivity, in a meta-analysis of 21 studies, including 4313 patients, made by Houssami et al, was 61.4% [51.2% - 79.4%], and the median ultrasound specificity was 82.0% (76.9%-89.0%), in reference [16]. Adding a axillary biopsy procedure to ultrasound, to assess patients with abnormal or suspicious axillary nodes,

found to have histologically positive nodes at the time of surgery**.** 

treatment decisions for surgery, chemotherapy, and radiation therapy.

**4.6.1 Percutaneous biopsy procedures in operable breast cancer** 

SF et al, in [15], evaluated retrospectively 244 consecutive patients diagnosed with invasive breast carcinoma, by physical examination of the axilla, digital mammography, axillary ultrasonography, and contrast enhanced breast magnetic resonance imaging. The authors found that from the patients who had all four modalities negative, 14% were ultimately found to have histologically positive nodes at the time of surgery.

The role of ultrasound in staging breast cancer differs with stage of disease, helping treatment decisions for surgery, chemotherapy, and radiation therapy.

#### *4.6.1. Percutaneous biopsy procedures in operable breast cancer*

In operable breast cancer, ultrasound helps identification of sentinel lymph node and of suspicious nodes, that warrant biopsy. Ultrasound alone has modest accuracy in detecting axillary metastasis, not being reliable, on its own, to make a decision in surgical treatment of the axilla. Ultrasound does not provide enough information to refer patients to complete axillary dissection.

The reported a median ultrasound sensitivity, in a meta-analysis of 21 studies, including 4313 patients, made by Houssami et al, was 61.4% [51.2%-79.4%], and the median ultrasound specificity was 82.0% (76.9%-89.0%), in reference [16]. Adding a axillary biopsy procedure to ultrasound, to assess patients with abnormal or suspicious axillary nodes, leads to a good sensitivity and excellent specificity (nearly 100%). The same meta-analysis, made by Houssami et al, in [16], evaluated 1733 patients, in whom needle biopsy was added and guided by ultrasound, because of abnormal findings. In these patients, the ultrasound-guided biopsy had median sensitivity of 79.4% (68.3%-8.9%) and a median specificity of 100% (100%-100%).

The study of Holwit DM, Margenthaler JA, in [17], retrospectively performed on 256 patients with clinically node-negative breast cancer, who underwent axillary ultrasound (AUS) evaluation and ultrasound-guided FNAB/needle core biopsy only in suspicious-appearing lymph nodes, found that the sensitivity and specificity of axillary ultrasound alone were 79% and 81%, respectively. The overall combined sensitivity and specificity for AUS-guided FNAB/ needle core biopsy were 71% and 99%, respectively, with a negative predictive value of 84% and a positive predictive value of 97%.

Axillary UNB has a good clinical utility, based on a meta-analysis of Houssami N, Diepstraten SCE et al, in [18], on 7097 patients, with a percent of 18.4% of patients effectively referred to axillary treatment thus avoiding SNB.

#### *4.6.2. Percutaneous biopsy procedures in locally advanced breast cancer*

Locally advanced stages of the disease are usually associated with obvious ultrasound features of axillary node involvement, and ultrasound helps the biopsy of these nodes, in most cases reffering the patient to systemic preoperative treatment.

Ultrasound examination and US-guided biopsy may the only possibility to diagnose the breast cancer that presents with no identifiable breast tumor and clinically positive axillary metastasis only. When mammography is negative, biopsy of the clinically positive lymph node is the only way to obtain a specimen for pathology and ultrasound could help localization and guiding the procedure.

SF et al, in [15], evaluated retrospectively 244 consecutive patients diagnosed with invasive breast carcinoma, by physical examination of the axilla, digital mammography, axillary ultrasonography, and contrast enhanced breast magnetic resonance imaging. The authors found that from the patients who had all four modalities negative, 14% were ultimately found

The role of ultrasound in staging breast cancer differs with stage of disease, helping treatment

In operable breast cancer, ultrasound helps identification of sentinel lymph node and of suspicious nodes, that warrant biopsy. Ultrasound alone has modest accuracy in detecting axillary metastasis, not being reliable, on its own, to make a decision in surgical treatment of the axilla. Ultrasound does not provide enough information to refer patients to complete

The reported a median ultrasound sensitivity, in a meta-analysis of 21 studies, including 4313 patients, made by Houssami et al, was 61.4% [51.2%-79.4%], and the median ultrasound specificity was 82.0% (76.9%-89.0%), in reference [16]. Adding a axillary biopsy procedure to ultrasound, to assess patients with abnormal or suspicious axillary nodes, leads to a good sensitivity and excellent specificity (nearly 100%). The same meta-analysis, made by Houssami et al, in [16], evaluated 1733 patients, in whom needle biopsy was added and guided by ultrasound, because of abnormal findings. In these patients, the ultrasound-guided biopsy had median sensitivity of 79.4% (68.3%-8.9%) and a median specificity of 100% (100%-100%).

The study of Holwit DM, Margenthaler JA, in [17], retrospectively performed on 256 patients with clinically node-negative breast cancer, who underwent axillary ultrasound (AUS) evaluation and ultrasound-guided FNAB/needle core biopsy only in suspicious-appearing lymph nodes, found that the sensitivity and specificity of axillary ultrasound alone were 79% and 81%, respectively. The overall combined sensitivity and specificity for AUS-guided FNAB/ needle core biopsy were 71% and 99%, respectively, with a negative predictive value of 84%

Axillary UNB has a good clinical utility, based on a meta-analysis of Houssami N, Diepstraten SCE et al, in [18], on 7097 patients, with a percent of 18.4% of patients effectively referred to

Locally advanced stages of the disease are usually associated with obvious ultrasound features of axillary node involvement, and ultrasound helps the biopsy of these nodes, in most cases

Ultrasound examination and US-guided biopsy may the only possibility to diagnose the breast cancer that presents with no identifiable breast tumor and clinically positive axillary metastasis only. When mammography is negative, biopsy of the clinically positive lymph node is the only

to have histologically positive nodes at the time of surgery.

decisions for surgery, chemotherapy, and radiation therapy.

*4.6.1. Percutaneous biopsy procedures in operable breast cancer*

axillary dissection.

32 Mammography Techniques and Review

and a positive predictive value of 97%.

axillary treatment thus avoiding SNB.

*4.6.2. Percutaneous biopsy procedures in locally advanced breast cancer*

reffering the patient to systemic preoperative treatment.

The advantages of preoperative systemic therapy include the potential downsizing of large tumors for either conversion of inoperable disease to resectable lesions or conversion of patients to breast conservation therapy, and in vivo assessment of the response of the tumor to chemotherapy, in reference [19]. Algorithms were issued for attributing patients to preop‐ erative systemic therapy.

biopsy, and the palpable node was removed by open surgery. Pathology showed axillary metastasis of invasive ductal carcinoma, with areas of mucinous carcinoma and failed to confirm the presence of the disease at the breast level. **Figure 11.** Patient presenting with palpable axillary lymph node. No breast tumor could be identified (mammography negative). Multiple passes were performed on the breast for core-biopsy, and the palpable node was removed by open surgery. Pathology showed axillary metastasis of invasive ductal carcinoma, with areas of mucinous carcinoma and failed to confirm the presence of the disease at the breast level.

Figure 11. Patient presenting with palpable axillary lymph node. No breast tumor could be identified (mammography negative). Multiple passes were performed on the breast for core-

tumors for either conversion of inoperable disease to resectable lesions or conversion of patients to breast conservation therapy, and in vivo assessment of the response of the tumor to chemotherapy, in reference [19]. Algorithms were issued for attributing patients to Lee at al, in [20], consider sonographically detected axillary metastases as a clinically positive axilla, so complete ALND is recommended for patients with positive axillary biopsy, even with a clinically negative axilla, after neoadjuvant chemotherapy.

preoperative systemic therapy.

The advantages of preoperative systemic therapy include the potential downsizing of large

Figure 12. Algorithm for axillary assessment in patients with locally advanced invasive breast cancer (adapted from Lee MC et al, in [20]). **Figure 12.** Algorithm for axillary assessment in patients with locally advanced invasive breast cancer (adapted from Lee MC et al, in [20]).

Lee at al, in [20], consider sonographically detected axillary metastases as a clinically positive axilla, so complete ALND is recommended for patients with positive axillary

#### biopsy, even with a clinically negative axilla, after neoadjuvant chemotherapy. **5. Conclusion**

**5. Conclusion**  Axillary staging for breast cancer evolved from axillary lymph node dissection towards the lesser invasive sentinel lymph node biopsy. Nowadays, although SLNB remains the standard procedure for diagnosing axillar involvement, axillary ultrasonography is Axillary staging for breast cancer evolved from axillary lymph node dissection towards the lesser invasive sentinel lymph node biopsy. Nowadays, although SLNB remains the standard procedure for diagnosing axillar involvement, axillary ultrasonography is performed as the initial staging examination breast cancer patients.

performed as the initial staging examination breast cancer patients. In operable breast cancer, ultrasound helps identification and guide biopsy of sentinel lymph node, and/or other imaging suspicious-appearing lymph nodes. As axillary ultrasonography with either FNAB or core-biopsy is a far less invasive approach to diagnose lymph node metastasis, approximately 15 % of breast cancer patients will avoid an unnecessary SLNB and proceed directly to complete axillary dissection.

For patients with locally advanced invasive breast cancer, the recent years brought a growing practice of the routine axillary ultrasound imaging, with early referral of patients to preoper‐ ative systemic chemotherapy.

#### **Author details**

Nastasia Serban\*

Address all correspondence to: serban\_nastasia@yahoo.com

"Carol Davila" University of Medicine and Pharmacy, "Dr. Ion Cantacuzino"Department of Obstetrics and Gynecology, Bucharest, Romania

#### **References**

Figure 12. Algorithm for axillary assessment in patients with locally advanced invasive breast cancer (adapted from Lee MC et al, in [20]).

**Figure 12.** Algorithm for axillary assessment in patients with locally advanced invasive breast cancer (adapted from

Positive Negative Negative Positive

Invasive breast cancer T2 or above

Axillary ultrasound imaging

Normal Abnormal or

SLNB FNAB or core biopsy

suspicious

Complete axillary dissection usually post chemotherapy

Lee at al, in [20], consider sonographically detected axillary metastases as a clinically positive axilla, so complete ALND is recommended for patients with positive axillary

Axillary staging for breast cancer evolved from axillary lymph node dissection towards the lesser invasive sentinel lymph node biopsy. Nowadays, although SLNB remains the standard procedure for diagnosing axillar involvement, axillary ultrasonography is

In operable breast cancer, ultrasound helps identification and guide biopsy of sentinel lymph node, and/or other imaging suspicious-appearing lymph nodes. As axillary ultrasonography

Axillary staging for breast cancer evolved from axillary lymph node dissection towards the lesser invasive sentinel lymph node biopsy. Nowadays, although SLNB remains the standard procedure for diagnosing axillar involvement, axillary ultrasonography is performed as the

biopsy, even with a clinically negative axilla, after neoadjuvant chemotherapy.

performed as the initial staging examination breast cancer patients.

initial staging examination breast cancer patients.

Complete axillary No axillary dissection

**5. Conclusion** 

**5. Conclusion**

Lee MC et al, in [20]).

dissection pre- or post chemotherapy

34 Mammography Techniques and Review


[18] Houssami N, Diepstraten SCE, Cody H, Turner RM, Sever AM. Clinical utility of ul‐ trasound-needle biopsy for preoperative staging of the axilla in invasive breast can‐ cer. Anticancer Research 2014;34 1087-1098.

[8] Nastasia S, Bordea C, Russu MC, Blidaru Al, Hudita D. Axillary ultrasound and the concept of sentinel lymph node in breast cancer. Revista Societăţii Române de Ob‐

[9] Choi JJ, Kang BJ, Kim SH, Lee JH, Jeong SH, Yim HW, Song BJ, Jung SS. Role of sono‐ graphic elastography in the differential diagnosis of axillary lymph nodes in breast

[10] Taylor K, OKeeffe S, Britton PD, Wallis MG, Treece GM, Housden J, Parashar D, Bond S, Sinnatamby R. Ultrasound elastography as an adjuvant to conventional ul‐ trasound in the preoperative assessment of axillary lymph nodes in suspected breast

[11] Wojcinski S, Dupont J, Schmidt W, Cassel M, Hillemanns P. Real-time ultrasound elastography in 180 axillary lymph nodes: elasticity distribution in healthy lymph nodes and prediction of breast cancer metastases. BMC Medical Imaging 2012;19 12-35. Open Access. http://www.biomedcentral.com/1471-2342/12/35. DOI:

[12] Esen G, Gurses B, Yilmaz MH, Ilvan S, Ulus S, Celik V, Farahmand M, Calay OO. Gray scale and power Doppler US in the preoperative evaluation of axillary metasta‐ ses in breast cancer patients with no palpable lymph nodes. European Radiology

[13] Moore A, Hester M, Nam MW, Brill Y M, McGrath P, Wright H, Weisinger K, Ro‐ mond E, Samayoa L M. Distinct lymph nodal sonographic characteristics in breast cancer patients at high risk for axillary metastases correlate with the final axillary

[14] Joh JE, Han G, Kiluk JV, Laronga C, Khakpour N, Lee MC. Indications for axillary ultrasound use in breast cancer patients. Clinical Breast Cancer 2012;12 433-437.

[15] Valente SA, Levine GM, Silverstein MJ, Rayhanabad JA, Weng-Grumley JG, Ji L, Holmes DR, Sposto R, Sener SF. Accuracy of predicting axillary lymph node positivi‐ ty by physical examination, mammography, ultrasonography, and magnetic reso‐ nance imaging. Annals of Surgical Oncology 2012;19(6):1825-30. DOI: 10.1245/

[16] Houssami N, Ciatto S, Turner RM, Cody HS III, Macaskill P. Preoperative ultra‐ sound-guided needle biopsy of axillary nodes in invasive breast cancer: meta-analy‐ sis of its accuracy and utility in staging the axilla. Annals of Surgery 2011;254

[17] Holwitt DM, Swatske ME, Gillanders WE, Monsees BS, Gao F, Aft RL, Eberlein TJ, Margenthaler JA. The combination of axillary ultrasound and ultrasound-guided bi‐ opsy is an accurate predictor of axillary stage in clinically node-negative breast can‐

cer patients. The American Journal of Surgery 2008;196 477–482.

cancer. Journal of Ultrasound in Medicine 2011;30(4) 429–36.

cancer: a pilot study. Clinical Radiology 2011;66(11) 1064–71.

stage. The British Journal of Radiology 2008;81(968) 630-636.

stetrică-Ginecologie 2008;2(LVI) 117-123.

36 Mammography Techniques and Review

10.1186/1471-2342-12-35.

2005;15(6) 1215–23.

s10434-011-2200-7.

243-251.


#### **Chapter 3**

## **Positron Emission Mammography**

### Mónica Vieira Martins

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/60452

#### **1. Introduction**

#### **1.1. Motivation**

Breast cancer imaging is a good example of how medical imaging modalities can diminish the number of patients suffering from this highly prevalent and deadly disease. X-ray mammog‐ raphy is the most used technique for breast cancer imaging. This technique is used either as a complementary tool to clinical diagnosis or as an irreplaceable screening tool for the early detection of the disease. However, it suffers from low specificity in the detection of malignancy and low sensitivity in women with dense breast tissue. Other imaging methods such as breast Ecography and breast Magnetic Resonance Imaging play important roles as adjunct techniques to X-ray mammography. The information provided by the aforementioned techniques is, however, mainly anatomical, thus leaving space for imaging methods that are able to obtain information regarding functional or metabolic changes in tissues.

Among these, molecular imaging methods using labeled radiotracers such as Scintigraphy, SPECT, Positron Emission Tomography (PET) and PET-CT, have been found to provide useful complementary information to the anatomical methods regarding detection, diagnosis and staging of breast cancer. However, the standard technology of whole body scanners, due to its limited spatial resolution, and to its disadvantageous geometry, which limits sensitivity, have in part precluded molecular imaging using radiotracers from contributing with its full potential to the imaging of the breast.

These limitations have prompted an active interest in the development of compact positron emission tomography cameras dedicated for breast imaging, a technique named Positron Emission Mammography. In the last 20 years there has been a tremendous effort from the industry and the scientific community to develop such devices, with a variety of detector designs and geometries, innovative radiation detection schemes, new scintillation crystals and adapted image reconstruction algorithms being studied in order to optimize the technique.

© 2015 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Either as prototypes or as commercial equipment, PEM scanners have provided data that confirms a huge improvement in technical characteristics with regards to whole body scanner, thus showing great promise of becoming a valuable modality in the clinical practice. In fact, Positron Emission Mammography, for which there are now two commercial equipment available, has demonstrated higher detectability than PET/CT and comparable or better sensitivity than MRI. It seems now to be clear that PEM is a valuable technique when MRI cannot be used.

In this chapter we will present a review from the literature of the main equipment that have been developed for Positron Emission Mammography, emphasizing the different approaches that have been followed in the design of the scanners, the main technical outcomes and, when such information is available, the most significant results obtained in the clinical trials.

The chapter will begin with a brief description of the principles behind the tomographic method for molecular imaging of positron emission isotopes. We will then review the per‐ formance parameters and instrumentation issues more challenging in the design of dedicated PEM scanners. We will review the PEM clinical results available so far, which allow a direct comparison between PEM, PET and MRI, and which give some clues regarding the clinical utility of PEM.

#### **2. Positron emission mammography technology**

The principles underlying Positron Emission Mammography are similar to those of Positron Emission Tomography. However, since there is a requirement for high resolution and high sensitivity, specific instrumentation issues are raised during the design and development of PEM scanners. This section will begin by presenting the principles of positron emission detection for medical imaging purposes; then, the key factors regarding the technical per‐ formance of the scanner will be presented. A very brief overview on the most important issues regarding the image reconstruction will also be presented.

#### **2.1. Positron emission and detection**

Positron Emission Mammography, as well as Positron Emission Tomography, uses radio‐ pharmaceuticals that are labeled with a positron emitting radionuclide.

The most used radiopharmaceutical for imaging cancer in PET is 18F-fluorodeoxyglucose, or 18F-FDG, a glucose analogue that is labeled with the positron emitter 18F. This radiotracer is used to detect glucose consumption, which is known to be increased in cancerous cells compared to normal cells. This is due to the higher metabolism of cancerous cells. The advantage of being able to image metabolism instead of anatomy, like in CT, is that the accelerated metabolic activity of cancerous cells occurs before the changes in the anatomical structures are detectable.

The tracking mechanism of 18FDG inside the cancerous cells is as follows. FDG is a glucose analog and, just like glucose, is transported into the cells by glucose transporters named GLUT1. These transporters are known to be overexpressed in breast cancerous cells, thus contributing to increased FDG uptake [1]. Once inside the cell, FDG is phosphorylated into deoxyglucose-6-phosphate (FDG-6-PO4) by an enzyme named hexokinase. Hexokinase is also thought to be overexpressed in cancerous cells. Unlike glucose, FDG-6-PO4 does not enter further enzymatic reactions and, due to its negative charge, it remains trapped inside the cell [2]. This metabolic trapping of FDG inside the cell constitutes the basis for imaging the in vivo distribution of the tracer.

Besides 18F, other radionuclides, such as 82Rb, 15O, 13N and 11C, among others, can be used to label the molecular probes used in PET.

Once emitted from the radiopharmaceutical, the positron travels a given amount of space in matter, while dissipating its kinetic energy through interactions with electrons and surround‐ ing nuclei. At the end of its path, the positron combines with an electron in its vicinity, in an annihilation reaction whereby the total mass of the electron and the positron is converted into high energy photons. The rest-mass energy of both positron and electron is 511 keV. If both particles are at rest at the time of annihilation, the two resulting 511 keV photons will be emitted in opposite directions.

The gamma rays thus emitted can interact with the tissues of the patient either by photoelectric effect or by Compton scatter. Both photon attenuation and scatter result in image degradation since, in the first case, emission counts are lost and, in the second, the measured spatial information is inaccurate. Whereas the correction for the effect of attenuation is fairly simple if a transmission scan is available, the correction for the scatter effect not as straightforward.

If the photons are not absorbed by the matter in the patient body, they can be detected externally in coincidence by using opposed pairs of scintillator crystals. Each luminous signal produced in the crystals is transformed into an electrical signal. A coincidence event happens when such two electrical signals are registered in a coincidence electronic circuit within a time frame that is defined by a coincidence time window.

The imaginary line that unites the two activated crystals in a coincidence event is called a line of response (LOR). The number of counts that are detected along the several LORs during an exam is stored in an histogram and used for image reconstruction purposes. Alternatively, list mode format can be used. In list mode format, the relevant information regarding the event, such as the activated crystals, the deposited energy and a time stamp, is stored sequentially on disk and used directly for image reconstruction.

#### **2.2. The performance of a scanner**

Either as prototypes or as commercial equipment, PEM scanners have provided data that confirms a huge improvement in technical characteristics with regards to whole body scanner, thus showing great promise of becoming a valuable modality in the clinical practice. In fact, Positron Emission Mammography, for which there are now two commercial equipment available, has demonstrated higher detectability than PET/CT and comparable or better sensitivity than MRI. It seems now to be clear that PEM is a valuable technique when MRI

In this chapter we will present a review from the literature of the main equipment that have been developed for Positron Emission Mammography, emphasizing the different approaches that have been followed in the design of the scanners, the main technical outcomes and, when such information is available, the most significant results obtained in the clinical trials.

The chapter will begin with a brief description of the principles behind the tomographic method for molecular imaging of positron emission isotopes. We will then review the per‐ formance parameters and instrumentation issues more challenging in the design of dedicated PEM scanners. We will review the PEM clinical results available so far, which allow a direct comparison between PEM, PET and MRI, and which give some clues regarding the clinical

The principles underlying Positron Emission Mammography are similar to those of Positron Emission Tomography. However, since there is a requirement for high resolution and high sensitivity, specific instrumentation issues are raised during the design and development of PEM scanners. This section will begin by presenting the principles of positron emission detection for medical imaging purposes; then, the key factors regarding the technical per‐ formance of the scanner will be presented. A very brief overview on the most important issues

Positron Emission Mammography, as well as Positron Emission Tomography, uses radio‐

The most used radiopharmaceutical for imaging cancer in PET is 18F-fluorodeoxyglucose, or 18F-FDG, a glucose analogue that is labeled with the positron emitter 18F. This radiotracer is used to detect glucose consumption, which is known to be increased in cancerous cells compared to normal cells. This is due to the higher metabolism of cancerous cells. The advantage of being able to image metabolism instead of anatomy, like in CT, is that the accelerated metabolic activity of cancerous cells occurs before the changes in the anatomical

The tracking mechanism of 18FDG inside the cancerous cells is as follows. FDG is a glucose analog and, just like glucose, is transported into the cells by glucose transporters named

**2. Positron emission mammography technology**

regarding the image reconstruction will also be presented.

pharmaceuticals that are labeled with a positron emitting radionuclide.

**2.1. Positron emission and detection**

structures are detectable.

cannot be used.

40 Mammography Techniques and Review

utility of PEM.

Of main importance for the performance of a PET scanner are its photon sensitivity and its spatial, energy and temporal resolutions. The geometry, the detector crystals and the elec‐ tronics of the system all impact on the above. In this paragraph we will overview the main aspects that affect these parameters, highlighting the most common differences between traditional clinical scanners and high resolution scanners.

The **photon sensitivity** is defined as the fraction of 511 keV photon pairs emitted from the imaging subject that are detected by the imaging system [3]. It is important that a scanner has the higher photon sensitivity possible since a higher fraction of detected photons will impact on better statistics of the acquired data and, consequently, on lower noise level of the final reconstructed images. The photon sensitivity in a clinical PET system is low. In dedicated or high resolution scanners, The sensitivity is usually improved by increasing the scanner geometric efficiency, that is, the probability that the emitted photon transverses the detected material, or by increasing the intrinsic detector efficiency.

Usually, the scanner geometric efficiency is enhanced in dedicated scanners relatively to traditional scanners mainly because the detector is brought closer to the imaging subject.

The geometric efficiency can also be increased by packing the detector elements as tightly as possible and by covering the region to be imaged with as much detector as possible. The other aspect that influences the scanner photon sensitivity is the intrinsic detector efficiency. This is defined as the likelihood that photons transversing the detector material will be stopped [3]. It depends mainly on the scintillator crystals that are used as detector elements. Scintillator crystals that have high density (ρ), with high effective atomic number (Zeff) have maximum ability to stop the 511 keV photons. In fact, a high density crystal favors the photon interaction and a high effective atomic number maximizes the probability of photoelectric interactions within the crystal, with respect to Compton events. The quantity that maximizes the crystal stopping power is ρ.Zeff5 . A scintillator that has a high stopping power will have a short attenuation length.

Table 1 lists the values of Zeff, ρ, the attenuation length and some other properties of the most common scintillator crystals used in PET scanners. The decay time determines the time resolution of the scanner and the light output determines the detector energy resolution and has effects also in the image resolution. Both these parameters will be discussed later.

Sodium iodide doped with thallium (NaI(Tl)) was the detector initially used in PET scanners. It has a very high light output (38 photons/keV), resulting in good energy and spatial resolu‐ tions [4]. However, its slow decay time leads to increased detector dead time and a high random coincidence rate (see below the discussion for system time resolution). Its low density results in a low stopping power (high attenuation length) when compared to the other crystals used in PET. Sodium iodine was first replaced by BGO (bismuth germanate: Bi4Ge3O12) that, despite its high decay time and poor light output, has an excellent stopping power.

More recently, other crystals that combine better light output with high stopping power have been introduced to PET. LSO (lutetium oxyorthosilicate: Lu2SiO5) has a high stopping power and a good light yield but, due to intrinsic properties of the crystal, its overall resolution is not as good as NaI(Tl) [5]. GSO (gadolinium orthosilicate: Gd2SiO5), despite its lower stopping power and light output, has better energy resolution than LSO. Both crystals are in use in PET scanners.

The **spatial resolution** describes the ability of the system to distinguish two closely spaced point sources. In PET, the fundamental limit of spatial resolution is imposed by the nature of positron annihilation. In fact, the emitted positron describes, before annihilating, a given path of variable length and direction. Therefore, the detected LOR contains the positron annihilation


the higher photon sensitivity possible since a higher fraction of detected photons will impact on better statistics of the acquired data and, consequently, on lower noise level of the final reconstructed images. The photon sensitivity in a clinical PET system is low. In dedicated or high resolution scanners, The sensitivity is usually improved by increasing the scanner geometric efficiency, that is, the probability that the emitted photon transverses the detected

Usually, the scanner geometric efficiency is enhanced in dedicated scanners relatively to traditional scanners mainly because the detector is brought closer to the imaging subject.

The geometric efficiency can also be increased by packing the detector elements as tightly as possible and by covering the region to be imaged with as much detector as possible. The other aspect that influences the scanner photon sensitivity is the intrinsic detector efficiency. This is defined as the likelihood that photons transversing the detector material will be stopped [3]. It depends mainly on the scintillator crystals that are used as detector elements. Scintillator crystals that have high density (ρ), with high effective atomic number (Zeff) have maximum ability to stop the 511 keV photons. In fact, a high density crystal favors the photon interaction and a high effective atomic number maximizes the probability of photoelectric interactions within the crystal, with respect to Compton events. The quantity that maximizes the crystal

Table 1 lists the values of Zeff, ρ, the attenuation length and some other properties of the most common scintillator crystals used in PET scanners. The decay time determines the time resolution of the scanner and the light output determines the detector energy resolution and

Sodium iodide doped with thallium (NaI(Tl)) was the detector initially used in PET scanners. It has a very high light output (38 photons/keV), resulting in good energy and spatial resolu‐ tions [4]. However, its slow decay time leads to increased detector dead time and a high random coincidence rate (see below the discussion for system time resolution). Its low density results in a low stopping power (high attenuation length) when compared to the other crystals used in PET. Sodium iodine was first replaced by BGO (bismuth germanate: Bi4Ge3O12) that,

More recently, other crystals that combine better light output with high stopping power have been introduced to PET. LSO (lutetium oxyorthosilicate: Lu2SiO5) has a high stopping power and a good light yield but, due to intrinsic properties of the crystal, its overall resolution is not as good as NaI(Tl) [5]. GSO (gadolinium orthosilicate: Gd2SiO5), despite its lower stopping power and light output, has better energy resolution than LSO. Both crystals are in use in PET

The **spatial resolution** describes the ability of the system to distinguish two closely spaced point sources. In PET, the fundamental limit of spatial resolution is imposed by the nature of positron annihilation. In fact, the emitted positron describes, before annihilating, a given path of variable length and direction. Therefore, the detected LOR contains the positron annihilation

has effects also in the image resolution. Both these parameters will be discussed later.

despite its high decay time and poor light output, has an excellent stopping power.

. A scintillator that has a high stopping power will have a short

material, or by increasing the intrinsic detector efficiency.

stopping power is ρ.Zeff5

42 Mammography Techniques and Review

attenuation length.

scanners.

**Table 1.** Properties of common scintillator crystals used in positron emission tomography.Adapted from [3] and from [6] (LuAP and LuYAP data). \* Data not found.

point, not the positron emission point, and these two points can be several millimeters apart. This positron range effect degrades the spatial resolution of the images. (Figure 1a). The positron range depends upon the energy of the emitted positron and upon the electronic density of the medium. It has been estimated a value of 0.22 mm FWHM for the positron range emitted from 18F in water, the major component of human cells [7].

In addition to the positron range, the acollinearity effect also leads to resolution degradation in PET systems. This effect is caused by the fact that the annihilation photons are almost never anti-parallel, since usually the positron and the electron are not exactly at rest when they annihilate. As a consequence, the detected line of response does not contain the point of positron-electron annihilation (Figure 1b). The degradation of the spatial resolution due to the accolinearity effect worsens as the detector diameter increases.

Another significant factor that limits PET spatial resolution is the size of the detector element. Spatial resolution may be improved significantly by reducing the detector pixel size. This is especially important in small diameter or dedicated PET scanners, where the pixel size dominates image resolution over the non-collinearity effect, which is minor for small detector diameters. Typical clinical systems use 4.0−6.0 mm detector pixel sizes, whereas small animal use detector pixels of 1.5−2.0 mm and positron emission mammography scanners use pixel sizes of 1.8−3.0 mm.

A final important factor that affects PET spatial resolution is the parallax error. This error occurs when the crystal depth at which the photon interaction takes place, known as Depth of Interaction (DOI), is not known. In this case, the LOR that unites the two activated crystals does not necessarily contain the true LOR (Figure 1c). For photons entering the scintillators at oblique angles there will be a mismatch between the true and the measured line of response. This degrading effect has greater impact in scanners where the distance between opposed detectors is smaller, like dedicated cameras. Good spatial resolution is of crucial importance for these cameras.

To improve photon sensitivity, the scintillator crystals that compose these dedicated scanners are usually long crystals. Therefore, not surprisingly, the ability to obtain DOI information has become an important factor in the design of high resolution PET scanners [8], with different strategies being followed to perform such measurements.

**Figure 1.** Schematic drawings of some of the effects that degrade spatial resolution in PET: a) the positron range, b) the accolinearity effect and c) the Depth of Interaction effect or parallax error.

The **energy resolution** indicates the precision with which the system can measure the incoming photon energy. A good energy resolution is important because it allows the use of a narrow energy window without significantly compromising photon sensitivity.

A narrow energy window helps to prevent contamination from photons that undergo scatter before interacting with the scintillator crystal, since the scatter process implies the loss of energy. It may also help to reduce the rate of random photon events, since a part of these photons undergo scatter. Random, or accidental, coincidences, occur when photons proceed‐ ing from different annihilations are detected within a same timing window and, although spatially uncorrelated, its detection is considered as a valid coincidence.

Energy resolution may be improved by using crystals with hight scintillation light output. A typical value for clinical PET scanners energy resolution is 25% FWHM at 511 keV [3].

The **time resolution** determines how well the system can decide whether two incoming photons arrived simultaneously. Good time resolution allows the use of a narrow time window, thus reducing random event's detection without compromising photon sensitivity. The reducing of random events is also important as it helps to prevent the system from saturating in high count statistics studies. The parameter that most strongly determines the temporal coincidence timing window is the scintillator decay time: a fast scintillator allows the selection of a narrow time window. A typical value for clinical PET scanners time resolution is 3 ns FWHM at 511 keV [3].

The scintillation light from the crystals is read from **photodetectors**. In PET, are most com‐ monly either photomultiplier tubes (PMT) or semiconductor based photodiodes. Photomul‐ tiplier tubes are the oldest and most reliable technology to detect and measure low levels of scintillation light. They have a high gain in the photoelectric conversion, which leads to high signal-to-noise ratios. Besides simple PMT's, a class of PMTs has been developed that provides not only energy information but also spatial information about the detected light. These PMTs, named Position-Sensitive PMTs (PS-PMT) have been found to be useful in the design of high resolution PET scanners [9].

PMTs have two major drawbacks. They have low quantum efficiency, meaning that the ratio between the incident photons and the primary produced electrons is low. In addition, PMTs are big devices, often with a small field of view, and this may constitute a drawback, especially when they are to be used in small dedicated scanners. In recent years there has also been great progress in the development of semiconductor photodetector arrays. These can be the PIN photodiode, the avalanche photodiode (APD) and the silicon drift detector (SDD). Among these, the APDs are the most used in PET cameras. Semiconductor photodetectors have many advantages over PMTs: they are very compact; they are insensitive to magnetic fields, which makes them good candidates for PET-MRI devices; they are available with large active areas; they have a very high efficiency [10]. The main disadvantages of APDs are their sensitivity to temperature and bias voltage.

#### **2.3. Image reconstrucion**

To improve photon sensitivity, the scintillator crystals that compose these dedicated scanners are usually long crystals. Therefore, not surprisingly, the ability to obtain DOI information has become an important factor in the design of high resolution PET scanners [8], with different

**Figure 1.** Schematic drawings of some of the effects that degrade spatial resolution in PET: a) the positron range, b) the

The **energy resolution** indicates the precision with which the system can measure the incoming photon energy. A good energy resolution is important because it allows the use of a narrow

A narrow energy window helps to prevent contamination from photons that undergo scatter before interacting with the scintillator crystal, since the scatter process implies the loss of energy. It may also help to reduce the rate of random photon events, since a part of these photons undergo scatter. Random, or accidental, coincidences, occur when photons proceed‐ ing from different annihilations are detected within a same timing window and, although

Energy resolution may be improved by using crystals with hight scintillation light output. A typical value for clinical PET scanners energy resolution is 25% FWHM at 511 keV [3].

The **time resolution** determines how well the system can decide whether two incoming photons arrived simultaneously. Good time resolution allows the use of a narrow time window, thus reducing random event's detection without compromising photon sensitivity. The reducing of random events is also important as it helps to prevent the system from saturating in high count statistics studies. The parameter that most strongly determines the temporal coincidence timing window is the scintillator decay time: a fast scintillator allows the selection of a narrow time window. A typical value for clinical PET scanners time resolution

The scintillation light from the crystals is read from **photodetectors**. In PET, are most com‐ monly either photomultiplier tubes (PMT) or semiconductor based photodiodes. Photomul‐ tiplier tubes are the oldest and most reliable technology to detect and measure low levels of scintillation light. They have a high gain in the photoelectric conversion, which leads to high

strategies being followed to perform such measurements.

44 Mammography Techniques and Review

accolinearity effect and c) the Depth of Interaction effect or parallax error.

is 3 ns FWHM at 511 keV [3].

energy window without significantly compromising photon sensitivity.

spatially uncorrelated, its detection is considered as a valid coincidence.

For many years, the problem of reconstructing an image from the projection data acquired in Positron Emission Tomography was addressed with analytic approaches which were inherited from X-ray computed tomography. Analytic algorithms such as filtered backprojection (FBP) are based on the direct inversion of the Radon transform. They are fast, linear, predictable, and their properties are very well known. The inversion of the Radon transform is derived for a continuous sampling and discretized afterward for sampled data [99]. Analytic algorithms are based on an idealized mathematical model for the data, the linear integral model, according to which the number of coincidence photon pairs detected along a LOR is approximately linearly proportional to the integral of the tracer density along a LOR. This model oversimplifies the physics inherent to the emission and detection processes in PET, limiting the accuracy of the images reconstructed with analytical algorithms.

In alternative to the analytic image reconstruction algorithms, model based algorithms, which can include accurate physical and statistical models of the systems, can be used. In opposition to analytical algorithms, they incorporate the discreteness of the data from the beginning. Their use usually results in improved image accuracy. As their formulation results frequently in large sets of nonlinear equations that must be solved by iterative methods, this class of algorithms is usually referred to as iterative image reconstruction algorithms. Furthermore, if statistical functions are used to derive them, they are said to be statistical iterative reconstruc‐ tion algorithms. The use of iterative methods based on probability models for image recon‐ struction was already effective in the field of astronomy in the early 1970's (Lucy, 1974, Richardson 1972). Later, in 1976, Rockmore and Macovsky introduced the Maximum Likeli‐ hood approach in the field of medical imaging. The Ordered Subsets – Expectation Maximi‐ zation (OSEM) algorithm was proposed in 1994 by Hudson and Larkin as an accelerated version of the ML-EM method and has since gained wide acceptance as a standard recon‐ struction method in PET. Nowadays, the most used algorithms for image reconstruction in PET belong to the class of iterative statistical algorithms.

#### **3. Positron emission mammography scanners**

A design of a dedicated positron emission imaging system for breast cancer was first presented in 1994 as a feasibility study for a positron emission mammography unit. Since then, more than ten other systems have been developed, two of which have become commercially available [11, 12]. Those systems differ in the number, geometry and mobility of the detectors used, with consequences on the patient positioning; its ability or not to perform biopsy; the different radiation detection scheme used, including different scintillation crystals; on the strategies used for image reconstruction from the measured projections. In this section we will review the instrumentation issues that impact on the performance of the equipment that use positron emission mammography to image disease. We will emphasize on the most demand‐ ing aspects of these dedicated instruments, showing why they hold the promise for an improved early detection of disease.

#### **3.1. The PEM-I system**

A design of a dedicated positron emission imaging system for breast cancer was first presented in 1994 by Thompson [13] from the Montreal Neurological Institute of the McGill University, Canada, as a feasibility study for a positron emission mammography unit.

The developed scanner was designed to fit a mammographic unit, so that conventional mammograms could also be performed in the same gantry, thus allowing exact registration of the emission and of the conventional mammographic images [14]. For such purposes, the system included a co-registration tool to facilitate registration between radiographic and metabolic images [15]. A schematic diagram of the scanner is presented in Figure 2.

The developed scanner consists on two planar 2x2 detector arrays of blocks of bismuth germanate (BGO) crystals placed above and below the compressed breast. The detector blocks measure 36x36x20 mm3 and are segmented into 1.9x1.9 mm2 pixels [16]. The separation between the detector heads can be adjusted to match the thickness of the breast. The system uses position sensitive photomultiplier tubes (PS-PMT) that are optically coupled to the crystal blocks. Although the PS-PMTs cover a surface of 72x72 mm2 , their useful field-of-view (FOV) is only of 65x55 mm2 . The coordinates of the coincidence on opposing PS-PMT faces are decoded by the system electronics and corrected for spatial distortion and efficiency [17].

The images from this system are obtained by performing a limited-angle weighted-backpro‐ jection algorithm. This consists on dividing the image into several equidistant planes and backprojecting the lines or response (LOR) onto those planes. With this technique, the image plane closest to the site of the tumor has the most focused image, while all the other planes present more blurred images, as it can be seen in the schematic diagram of Figure 3. This is known as the focal plane effect. The reconstruction scheme is said to be 'weighted backpro‐ jection' because the values that are added to a given plane in the image matrix are weighted accordingly to the probability of detection of an annihilation in that plane, the crystals efficiencies and the photon attenuation along the path to the crystal [14].

**3. Positron emission mammography scanners**

improved early detection of disease.

**3.1. The PEM-I system**

46 Mammography Techniques and Review

measure 36x36x20 mm3

is only of 65x55 mm2

A design of a dedicated positron emission imaging system for breast cancer was first presented in 1994 as a feasibility study for a positron emission mammography unit. Since then, more than ten other systems have been developed, two of which have become commercially available [11, 12]. Those systems differ in the number, geometry and mobility of the detectors used, with consequences on the patient positioning; its ability or not to perform biopsy; the different radiation detection scheme used, including different scintillation crystals; on the strategies used for image reconstruction from the measured projections. In this section we will review the instrumentation issues that impact on the performance of the equipment that use positron emission mammography to image disease. We will emphasize on the most demand‐ ing aspects of these dedicated instruments, showing why they hold the promise for an

A design of a dedicated positron emission imaging system for breast cancer was first presented in 1994 by Thompson [13] from the Montreal Neurological Institute of the McGill University,

The developed scanner was designed to fit a mammographic unit, so that conventional mammograms could also be performed in the same gantry, thus allowing exact registration of the emission and of the conventional mammographic images [14]. For such purposes, the system included a co-registration tool to facilitate registration between radiographic and

The developed scanner consists on two planar 2x2 detector arrays of blocks of bismuth germanate (BGO) crystals placed above and below the compressed breast. The detector blocks

between the detector heads can be adjusted to match the thickness of the breast. The system uses position sensitive photomultiplier tubes (PS-PMT) that are optically coupled to the crystal

decoded by the system electronics and corrected for spatial distortion and efficiency [17].

The images from this system are obtained by performing a limited-angle weighted-backpro‐ jection algorithm. This consists on dividing the image into several equidistant planes and backprojecting the lines or response (LOR) onto those planes. With this technique, the image plane closest to the site of the tumor has the most focused image, while all the other planes present more blurred images, as it can be seen in the schematic diagram of Figure 3. This is known as the focal plane effect. The reconstruction scheme is said to be 'weighted backpro‐ jection' because the values that are added to a given plane in the image matrix are weighted accordingly to the probability of detection of an annihilation in that plane, the crystals

and are segmented into 1.9x1.9 mm2 pixels [16]. The separation

. The coordinates of the coincidence on opposing PS-PMT faces are

, their useful field-of-view (FOV)

metabolic images [15]. A schematic diagram of the scanner is presented in Figure 2.

Canada, as a feasibility study for a positron emission mammography unit.

blocks. Although the PS-PMTs cover a surface of 72x72 mm2

efficiencies and the photon attenuation along the path to the crystal [14].

**Figure 2.** Schematic drawing of the PEM-I detector plates (white areas) mounted on a conventional mammographic unit (gray areas). From [15].

**Figure 3.** Weighted backprojection used in the PEM-I scanner. From [18].

The complete system has a spatial resolution of 2.8 mm FWHM, a time resolution of 12 ns and an efficiency of 3% at a detector separation of 55 mm [19]. It is estimated that the system is not able to detect tumors with a tumor-to-background ratio lower than 6:1 [20].

The preliminary clinical trials, performed with 16 subjects, reported 80% sensitivity and 100% specificity. The accuracy of the exam, computed as the ratio of the sum of the true findings (positive and negative) to the total number of lesions, was 86% [18]. For mammography exams performed on the same subjects, those values were, respectively, 90%, 50% and 81%. The smallest cancerous lesion detected with PEM-I was 1.1x1.1x0.9 cm3 . The lower value of sensitivity with PEM, with respect to mammography, was due, according to the authors, to the small FOV of the PEM device and to the impossibility of imaging tumors localized close to the chest wall (less than 2 cm). These limitations are related to the PMTs used, whose useful field of view is significantly smaller than its area, preventing imaging near their edges. Figure 4 shows a typical set of images obtained with PEM-I, each image corresponding to a plane of the sample, with a visible site of FDG uptake in a region close to the chest wall [19].

**Figure 4.** A typical set of images obtained with the PEM-I scanner. Each image corresponds to a plane, with the left‐ most image corresponding to the image plane closer to the upper detector. A visible site of FDG uptake can be seen in a region close to the chest wall. From [19].

#### **3.2. The Naviscan PEM system**

The original idea of Thompson for a PEM system was further developed by Weinberg and colleagues, for the Naviscan PET System. The Naviscan PEM Flex consists of two 5.6×17.3 cm2 opposed detector heads [21] that can be fit on a steriotactic mammography unit [22]. This way, emission and transmission scans can be obtained. Data acquisition is performed by moving the detectors along a linear path, in order to image as much breast as possible. The PEM detectors translation allows to image an area equal to the entire X-ray field of view [23]. The system can also work separately from the mammography unit, allowing closer chest wall access. Figure 5 shows the PEM Flex system mounted in a stereotactic X-ray mammography unit.

Each detector head contains twelve 13×13 crystal blocks, each coupled to a compact PS-PMT. The crystals are 2×2×10 mm3 of a mixed-lutetium silicate [21].

For each segment of the scan, list mode data are acquired, histogrammed and reconstructed by backprojection. This allows the operator to view partial images during the scan acquisition. At the end of the entire scan, the complete list mode data are reconstructed using a maximumlikelihood expectation-maximization algorithm.

The intrinsic spatial resolution of the system is 1.5 mm FHWM [23]. The image resolution is 2.5 mm FWHM in the plane perpendicular to the displacement and 6 mm between planes. Energy resolution was measured as 14% for 511 keV. The timing window used was 9 ns.

**Figure 5.** The PEM Flex system mounted in a stereotactic X-ray mammography unit. From [22].

The preliminary clinical trials, performed with 16 subjects, reported 80% sensitivity and 100% specificity. The accuracy of the exam, computed as the ratio of the sum of the true findings (positive and negative) to the total number of lesions, was 86% [18]. For mammography exams performed on the same subjects, those values were, respectively, 90%, 50% and 81%. The

sensitivity with PEM, with respect to mammography, was due, according to the authors, to the small FOV of the PEM device and to the impossibility of imaging tumors localized close to the chest wall (less than 2 cm). These limitations are related to the PMTs used, whose useful field of view is significantly smaller than its area, preventing imaging near their edges. Figure 4 shows a typical set of images obtained with PEM-I, each image corresponding to a plane of

**Figure 4.** A typical set of images obtained with the PEM-I scanner. Each image corresponds to a plane, with the left‐ most image corresponding to the image plane closer to the upper detector. A visible site of FDG uptake can be seen in

The original idea of Thompson for a PEM system was further developed by Weinberg and colleagues, for the Naviscan PET System. The Naviscan PEM Flex consists of two 5.6×17.3 cm2 opposed detector heads [21] that can be fit on a steriotactic mammography unit [22]. This way, emission and transmission scans can be obtained. Data acquisition is performed by moving the detectors along a linear path, in order to image as much breast as possible. The PEM detectors translation allows to image an area equal to the entire X-ray field of view [23]. The system can also work separately from the mammography unit, allowing closer chest wall access. Figure 5

Each detector head contains twelve 13×13 crystal blocks, each coupled to a compact PS-PMT.

For each segment of the scan, list mode data are acquired, histogrammed and reconstructed by backprojection. This allows the operator to view partial images during the scan acquisition. At the end of the entire scan, the complete list mode data are reconstructed using a maximum-

The intrinsic spatial resolution of the system is 1.5 mm FHWM [23]. The image resolution is 2.5 mm FWHM in the plane perpendicular to the displacement and 6 mm between planes. Energy resolution was measured as 14% for 511 keV. The timing window used was 9 ns.

of a mixed-lutetium silicate [21].

shows the PEM Flex system mounted in a stereotactic X-ray mammography unit.

the sample, with a visible site of FDG uptake in a region close to the chest wall [19].

. The lower value of

smallest cancerous lesion detected with PEM-I was 1.1x1.1x0.9 cm3

a region close to the chest wall. From [19].

48 Mammography Techniques and Review

**3.2. The Naviscan PEM system**

The crystals are 2×2×10 mm3

likelihood expectation-maximization algorithm.

The clinical trials performed so far [24, 25] were all performed on patients with known breast cancer or suspected lesions. Hence, they provide little information on the specificity of the technique. In one of these studies [25], PEM was able to visualize 39 out of 44 lesions. The non visualized lesions ranged in size from a 1 mm ductal carcinoma in situ (DCIS) to a 1 cm infiltrating ductal carcinoma. Some lesions could not be visualized due to limitations on how posterior the breast tissue is observable by the device. Others, as interpreted by the authors of the study, due to the variability in the metabolic activity of breast cancer cells, similarly to what happens with whole body PET.

**Figure 6.** Image of DCIS obtained with the PEM Flex scanner a), with MRI b) and with mammography c). Neither MRI nor mammography could detect the DCIS lesion seen in the PEM image. From [25].

The most encouraging finding in this trial was the fact that PEM was able to visualize DCIS not visualized by mammography, breast ecography or MRI. An example of such a case can be seen in Figure 6. The smallest lesion detected by PEM in this study was a 2 mm duct of DCIS. This preliminary clinical trial seems to indicate that the technology is promising and worthy of further investigation.

#### **3.3. The West Virginia University — Jefferson Lab PEM system**

Another PEM system was developed and tested at the West Virginia University and at the Jefferson Laboratory by Raylman and colleagues [26]. This PEM system, which is mounted on a stereotactic biopsy table, consists of two square 10×10 cm2 detector arrays of discrete 3×3×10 mm3 GSO crystals. The scintillation light is collected by arrays of PS-PMTs. An image of the scanner mounted on the biopsy table, together with a torso phantom can be seen in Figure 7.

**Figure 7.** Image of the West Virginia University - Jefferson Laboratory PEM system. The PEM detector heads, mounted in a biopsy table, are highlighted by the black arrows. A torso phantom can be seen in the table. From [27].

Since one of the goals of the system is to perform PEM guided biopsies, a trigonometric algorithm was developed to determine the lesion stereotactic coordinates. This algorithm uses two PEM images that are acquired at two symmetric angles (±15◦).

PEM images acquired in a single detector position were initially reconstructed using a weighted backprojection algorithm similar to the used for the PEM-I system described above, or by a limited angle tomography scheme [23]. Later, the use of acquired data at two detector positions (±15◦) [28] to guide stereotactic biopsy motivated the use of an adapted Maximum Likelihood - Expectation Maximization algorithm.

The described acquisition scenario was compared with multiple acquisitions between the same limiting angles, at small uniform increments [29]. The results were somehow mixed, with no clear evidence of significant advantage of one acquisition scenario over the other, although less artifacts were observed with the multiple angle acquisition.

This lead to a study of the complete angular sampling around the breast [30] through step and shoot acquisitions. Not surprisingly, this study showed that the complete angular sampling provided better image quality with respect to a single acquisition with stationary detectors. The study also revealed some of the weaknesses of the system, such as the low rate acquisition capability and the lack of DOI information.

Posterior work reports a new design of the scanner, now named PEM-PET [31], as it means to be a tomographic system. This system has four planar detector heads that can rotate around the breast. The detector crystal used is now LYSO, with 2×2×15 mm3 individual detector elements. The PEM-PET system has 2 mm FHWM resolution, possessing, as its anterior version, the ability to guide biopsy. The initial clinical studies have shown the ability of the system to detect lesions also detected by standard methods [32].

#### **3.4. The Duke University — Jefferson Lab PEM system**

Another system was developed at the Jefferson Laboratory and Duke University to image the compressed breast [33]. This PEM system has two opposed planar 15×20 cm2 detectors that

**Figure 8.** Image of the Duke University - Jefferson Laboratory PEMsystem positioned in a mammography unit. The PEM detectors are highlighted by black arrows.From [34].

acquire data without rotational or translational movements. The detector arrays are composed of 3×3×10 mm3 of lutetium gadolinium oxyorthosilicate, LGSO. The scintillation light is collected by arrays of PS-PMTs. This system is used mounted on an X-ray mammography unit, although the PEM detector heads must be removed to acquire the X-ray image. The distance between the detector heads can be adjusted to match the size of the breast. Image reconstruction is performed by means of the backprojection scheme. The image spatial resolution varied from 4.8 mm to 6 mm, depending on the acceptance angles of the lines of response. An image of the system can be seen in Figure 8.

A pilot clinical trial was performed using this system [34]. This trial included 23 patients with suspected breast malignancies. Therefore, it does not provide meaningful information concerning the specificity of the technique. In this study, where the majority of the evaluated lesions had diameters smaller than 2.5 cm, PEM presented a sensitivity of 86%. The size of the three malignant lesions that PEM was unable to detect varied from 8 mm to 15 mm. The system was able to detect a 4 mm DCIS that was not detected by mammography.

#### **3.5. The maxPET system**

**Figure 7.** Image of the West Virginia University - Jefferson Laboratory PEM system. The PEM detector heads, mounted

Since one of the goals of the system is to perform PEM guided biopsies, a trigonometric algorithm was developed to determine the lesion stereotactic coordinates. This algorithm uses

PEM images acquired in a single detector position were initially reconstructed using a weighted backprojection algorithm similar to the used for the PEM-I system described above, or by a limited angle tomography scheme [23]. Later, the use of acquired data at two detector positions (±15◦) [28] to guide stereotactic biopsy motivated the use of an adapted Maximum

The described acquisition scenario was compared with multiple acquisitions between the same limiting angles, at small uniform increments [29]. The results were somehow mixed, with no clear evidence of significant advantage of one acquisition scenario over the other, although

This lead to a study of the complete angular sampling around the breast [30] through step and shoot acquisitions. Not surprisingly, this study showed that the complete angular sampling provided better image quality with respect to a single acquisition with stationary detectors. The study also revealed some of the weaknesses of the system, such as the low rate acquisition

Posterior work reports a new design of the scanner, now named PEM-PET [31], as it means to be a tomographic system. This system has four planar detector heads that can rotate around

elements. The PEM-PET system has 2 mm FHWM resolution, possessing, as its anterior version, the ability to guide biopsy. The initial clinical studies have shown the ability of the

Another system was developed at the Jefferson Laboratory and Duke University to image the

individual detector

detectors that

in a biopsy table, are highlighted by the black arrows. A torso phantom can be seen in the table. From [27].

two PEM images that are acquired at two symmetric angles (±15◦).

less artifacts were observed with the multiple angle acquisition.

the breast. The detector crystal used is now LYSO, with 2×2×15 mm3

compressed breast [33]. This PEM system has two opposed planar 15×20 cm2

system to detect lesions also detected by standard methods [32].

**3.4. The Duke University — Jefferson Lab PEM system**

Likelihood - Expectation Maximization algorithm.

50 Mammography Techniques and Review

capability and the lack of DOI information.

A dedicated PET camera for mammary and axillary region imaging, maxPET, was designed and constructed at the Crump Institute for Biological Imaging [35]. This group used an alternative scheme to couple the crystal arrays to the PMT's, in order to avoid the problems associated with the inactive area near the PMT edges. The maxPET system consisted of two 15×15 cm2 planar scintillation detector plates, each composed of several modular detectors. The detectors are composed of arrays of 3×3×20 mm3 LSO crystals, each crystal array being coupled to an optical fiber which in turn is coupled to a PS-PMT. The use of the optical fiber allows the exact match between the crystal area and the active PMT dimensions, thus avoiding gaps between detector modules. It also provides better imaging close to the chest wall, since the plates are active out to the edge of the field-of-view. The main disadvantage of the fiberoptic coupling is the loss of scintillation light.

The two detector plates can be mounted in a gantry allowing variable plates separation, detector plates rotation and angular motion. Based on Monte Carlo simulation, the expected intrinsic spatial resolution of the scanner was about 2.3 mm [36]. A prototype of this system was assembled but, to our knowledge, no clinical test were ever performed.

A second prototype of this detector was build, with modified geometry and electronics. The integration of this system with a dedicated CT system was exploited [37].

#### **3.6. The LBNL PEM system**

The PEM scanner developed at the Lawrence Berkeley National Laboratory (LBNL) has two major differences from the PEM scanners described here: it has a rectangular geometry, with four detector plates surrounding the breast and it has Depth of Interaction measurements capabilities [38].

The system uses a 6 ns time window and has 5% sensitivity at the center of the FOV. The measured spatial resolution of the scanner is almost uniform in the entire field-of-view, ranging from 1.9 mm FWHM at the FOV centre to 2.1 mm at the FOV corner [38]. Images of a mini-Derenzo phantom show that the smallest lesion resolved by the system is 2.4 mm in diameter.

In the context of the development of this scanner, a simulation study was done to compare the presented rectangular detector configuration with a dual stationary detector system, such as some of the systems presented above. For such purpose it was used the Fisher information matrix, an analytical computation that allows to characterize how easily a change of one parameter in the source distribution can be identified from the measured data [40]. This study has shown that the rectangular system with Depth of Interaction capability has a higher signalto-noise ratio for detection tasks and a lower bias at a given noise level for quantitation tasks. It is worth stressing that this study did not include the case of a rotating dual head scanner [39]. The LBNL PEM system consists of four detector plates that cover a rectangular 8.2×6.0×5.0 cm3 field of view. The detector modules contain arrays of 3×3×30 mm3 LSO crystals that are coupled to a single photomultiplier tube (PMT) in one end and to a photodiode array (PD) on the other end. The ratio between the signals of the PMT and the PD allow the estimation of the Depth of Interaction of the photon [41]. The achieved DOI resolution ranges from 6 mm FWHM at the PD end to 11 mm FWHM at the PMT end [42].

The image reconstruction task for this scanner has been subject of an intense work. In an initial stage, image reconstruction was performed with a filtered backprojection based reconstruction algorithm that took into account the existence of DOI information and the irregular angular sampling of the scanner [43, 44]. Later, a list mode penalized maximum likelihood algorithm using Gaussian priors was developed [40, 45, 46]. A Monte Carlo based scatter correction algorithm was also developed [46]. To our knowledge, this scanner has never been tested with clinical data.

#### **3.7. The YAP-PEM system**

The YAP-PEM prototype was developed within a collaboration of the Italian Universities of Pisa, Ferrara, Bologna and Roma [47]. The technology of this device derives from a small animal scanner previously developed by the group. The YAP-PEM scanner has been designed with the aim of detecting 5 mm breast lesions in diameter and an activity ratio of 10:1 between the cancer and the breast tissue. The device is composed of two stationary detector heads made of yttrium aluminium perovskite (YAlO3) scintillators doped with cerium (YAP:Ce). This is a scintillator crystal that produces a light output of about 20 photons/keV, has a decay constant of 30 ns and a density of 5.4 g/cm3 [48]. It has, however, a low Z number. Each detector head has a detection area of 6×6 cm2 that comprises 30×30 detection elements with 2×2×30 mm3 each. The system uses PS-PMTs to collect the scintillation light. The distance between the detectors can range from 5 to 10 cm, depending on the breast compression used.

For image reconstruction purposes, the ML-EM algorithm has been adapted to the planar nature of the acquired data, in order to obtain a pseudo-tomographic imaging method [49]. This method works on data that is converted into histograms that are indicated for planar data. These are known as planograms [50]. Geometrical symmetries are used to speed up the computations.

Monte Carlo simulation and image reconstruction studies performed for the YAP-PEM scanner indicate that the scanner is expected to have capability of discriminating 5 mm tumors in a target-to-background ratio of 10:1. However, due to the planar nature of the data, if two sources lie on the same axial plane, the system cannot discriminate them, as it can be seen in Figure 10.

**Figure 9.** Phantom images obtained with the YAP-PEM prototype. Due to the planar nature of the scanner's data, if two sources lie on a same axial plane, the system cannot discriminate them. From [49].

#### **3.8. The Clear-PEM system**

The two detector plates can be mounted in a gantry allowing variable plates separation, detector plates rotation and angular motion. Based on Monte Carlo simulation, the expected intrinsic spatial resolution of the scanner was about 2.3 mm [36]. A prototype of this system

A second prototype of this detector was build, with modified geometry and electronics. The

The PEM scanner developed at the Lawrence Berkeley National Laboratory (LBNL) has two major differences from the PEM scanners described here: it has a rectangular geometry, with four detector plates surrounding the breast and it has Depth of Interaction measurements

The system uses a 6 ns time window and has 5% sensitivity at the center of the FOV. The measured spatial resolution of the scanner is almost uniform in the entire field-of-view, ranging from 1.9 mm FWHM at the FOV centre to 2.1 mm at the FOV corner [38]. Images of a mini-Derenzo phantom show that the smallest lesion resolved by the system is 2.4 mm in

In the context of the development of this scanner, a simulation study was done to compare the presented rectangular detector configuration with a dual stationary detector system, such as some of the systems presented above. For such purpose it was used the Fisher information matrix, an analytical computation that allows to characterize how easily a change of one parameter in the source distribution can be identified from the measured data [40]. This study has shown that the rectangular system with Depth of Interaction capability has a higher signalto-noise ratio for detection tasks and a lower bias at a given noise level for quantitation tasks. It is worth stressing that this study did not include the case of a rotating dual head scanner [39]. The LBNL PEM system consists of four detector plates that cover a rectangular 8.2×6.0×5.0

 field of view. The detector modules contain arrays of 3×3×30 mm3 LSO crystals that are coupled to a single photomultiplier tube (PMT) in one end and to a photodiode array (PD) on the other end. The ratio between the signals of the PMT and the PD allow the estimation of the Depth of Interaction of the photon [41]. The achieved DOI resolution ranges from 6 mm FWHM

The image reconstruction task for this scanner has been subject of an intense work. In an initial stage, image reconstruction was performed with a filtered backprojection based reconstruction algorithm that took into account the existence of DOI information and the irregular angular sampling of the scanner [43, 44]. Later, a list mode penalized maximum likelihood algorithm using Gaussian priors was developed [40, 45, 46]. A Monte Carlo based scatter correction algorithm was also developed [46]. To our knowledge, this scanner has never been tested with

The YAP-PEM prototype was developed within a collaboration of the Italian Universities of Pisa, Ferrara, Bologna and Roma [47]. The technology of this device derives from a small animal

at the PD end to 11 mm FWHM at the PMT end [42].

was assembled but, to our knowledge, no clinical test were ever performed.

integration of this system with a dedicated CT system was exploited [37].

**3.6. The LBNL PEM system**

52 Mammography Techniques and Review

capabilities [38].

diameter.

cm3

clinical data.

**3.7. The YAP-PEM system**

The Clear-PEM system was developed within the framework of the CrystalClear collaboration, in CERN. The system, which was designed to allow the examination of the breast and the axilla regions, is composed of a dual-plate detector head that is housed in a robotic mechanical gantry, as represented in Figure 11. For the breast examination, the patient lays in the prone position with the breast hanging through an aperture in the patient. The two detector heads are positioned in each side of the breast, as represented in Figure 11, and the projection data are acquired at several angular positions. The detector heads can be positioned at different separation distances, allowing for the accommodation of different breast sizes.

**Figure 10.** Representation of the Clear-PEM system.From [51].

The detector heads cover a 16.2×14.1 cm2 FoV. Each detector head holds 96 detector modules, is constituted of a total of 3072 LYSO:Ce crystals, each crystal having 2×2×20 mm3 . A scheme of a detector head is shown in Figure 12a. The distribution of the crystals within, the detector plates is as follows. Each detector plate is constituted of a set of four structures named supermodules, each one with 14×4 cm2 , placed side by side. Each supermodule is composed of 12×2 modules that, in turn, are composed of an array of 4×8 LYSO:Ce crystals. Therefore, each detector plate is constituted of 48×64 LYSO:Ce crystals. Figure 11: Representation of the Clear-PEM system.From (51). The detector heads cover a 16.2×14.1 cm<sup>2</sup> FoV. Each detector head holds 96 detector modules, is constituted of a total of 3072 LYSO:Ce crystals, each crystal having 2×2×20 mm<sup>3</sup> . A scheme of a detector head is shown in Figure 12a). The distribution of the crystals within, the detector plates is as follows. Each detector plate is constituted of a set of four

, placed side by side. Each

structures named supermodules, each one with 14×4 cm2

supermodule is composed of 12×2 modules that, in turn, are composed of an array of 4×8 LYSO:Ce crystals. Therefore, each detector plate is constituted of 48×64 LYSO:Ce crystals.

Figure 12: Clear-PEM detectors. a) Representation of a Clear-PEM detector head, with an highlighted detector supermodule. From (51). b) Representation of a Clear-PEM detector module with the double readout scheme. From (52). The readout of each module is performed by two 32-pixel avalanche photodiodes that are optically coupled to each side of the module, as shown in Figure 12b). This double readout **Figure 11.** Clear-PEM detectors. a) Representation of a Clear-PEM detector head, with an highlighted detector super‐ module. From [51]. b) Representation of a Clear-PEM detector module with the double readout scheme. From [52].

scheme allows the DOI measure. The DOI coordinate within the crystal is estimated from the asymmetry of the collected light at the top and bottom APD pixels. Experimental results have shown that, with this scheme, it is possible to obtain a 2 mm FWHM DOI resolution (53). This measurement is important since it increases the uniformity of measure all over the The readout of each module is performed by two 32-pixel avalanche photodiodes that are optically coupled to each side of the module, as shown in Figure 12b. This double readout scheme allows the DOI measurement. The DOI coordinate within the crystal is estimated from the asymmetry of the collected light at the top and bottom APD pixels. Experimental results have shown that, with this scheme, it is possible to obtain a 2 mm FWHM DOI resolution [53]. This measurement is important since it increases the uniformity all over the field-of-view of the scanner. This feature is not common in the universe of the positron emission mammogra‐ phy dedicated scanners and therefore it is perhaps one of the most important characteristics of this scanner. The processing of the detector analogical signals, including the readout, the low noise amplification, the sampling and the storage are implemented in dedicated ASICs (Application Specific Integrated Circuits) integrated in the detection plates. The output analogue sampled pulses are digitized by Analogue Digital Converters (ADC) and transmitted to the data acquisitions system (DAQ). This system, based on FPGA (Field Programmable Gate Arrays), is responsible for the data reduction and storage. Data is send from here to a trigger system that selects two-photon events in coincidence within a programmable timing window and, at each trigger, the relevant data frames are send to the acquisition PC where energy and time information are analyzed [54]. For events with more than one active crystal in a detector head, due to Compton scattering, an event reconstruction position algorithms is used to assign the coordinates of the interaction. Those events that are within the selected energy window are validated, their final coordinates being used to define the LORs that are stored in a listmode file [53]. Images are reconstructed using 3D statistical iterative algorithms or 2D algebraic techniques [55–57]. Phantom studies showed that the scanner is suitable for milimetric tumor detection; Derenzo phantom studies showed that lesions up to 2 mm in diameter can be clearly seen; a gelatin breast phantom indicated that lesions up to 3 mm in diameter can be detected. The first prototype, installed at the Institute of Nuclear Sciences Applied to Health, Coimbra, Portugal, was used for the first tests in a preclinical environment [58].

More recently, the ClearPEM scanner has been used to create a multimodal PEM and ultrasound scanner [59]. The aim of this system is to provide high resolution, high sensitivity and high specificity metabolic information from PEM matched with 3D high resolution anatomic information. The new ClearPEM-Sonic scanner also provides elastographic information from ultrasound, further improving the specificity of the multimodality scanner. This second prototype is installed at the Hôpital Nord, in Marseille, France, and has confirmed a spatial resolution of about 1.5 mm in phantom studies. The system allows the co-registration of US/ClearPEM and elastography/ClearPEM with an alignment precision of about 2 mm. The first clinical trial has confirm the ability to detect very small lesions only seen by MRI, although lesions very close to the chest wall are left undetect‐ ed [60].

#### **3.9. The MAMMI system**

**Figure 10.** Representation of the Clear-PEM system.From [51].

mm<sup>3</sup>

each detector plate is constituted of 48×64 LYSO:Ce crystals.

The detector heads cover a 16.2×14.1 cm<sup>2</sup>

supermodules, each one with 14×4 cm2

54 Mammography Techniques and Review

The detector heads cover a 16.2×14.1 cm2 FoV. Each detector head holds 96 detector modules,

of a detector head is shown in Figure 12a. The distribution of the crystals within, the detector plates is as follows. Each detector plate is constituted of a set of four structures named

of 12×2 modules that, in turn, are composed of an array of 4×8 LYSO:Ce crystals. Therefore,

modules, is constituted of a total of 3072 LYSO:Ce crystals, each crystal having 2×2×20

supermodule is composed of 12×2 modules that, in turn, are composed of an array of 4×8 LYSO:Ce crystals. Therefore, each detector plate is constituted of 48×64 LYSO:Ce crystals.

Figure 12: Clear-PEM detectors. a) Representation of a Clear-PEM detector head, with an highlighted detector supermodule. From (51). b) Representation of a Clear-PEM detector

**Figure 11.** Clear-PEM detectors. a) Representation of a Clear-PEM detector head, with an highlighted detector super‐ module. From [51]. b) Representation of a Clear-PEM detector module with the double readout scheme. From [52].

The readout of each module is performed by two 32-pixel avalanche photodiodes that are optically coupled to each side of the module, as shown in Figure 12b. This double readout scheme allows the DOI measurement. The DOI coordinate within the crystal is estimated from the asymmetry of the collected light at the top and bottom APD pixels. Experimental results have shown that, with this scheme, it is possible to obtain a 2 mm FWHM DOI resolution [53]. This measurement is important since it increases the uniformity all over the field-of-view of the scanner. This feature is not common in the universe of the positron emission mammogra‐ phy dedicated scanners and therefore it is perhaps one of the most important characteristics of this scanner. The processing of the detector analogical signals, including the readout, the low noise amplification, the sampling and the storage are implemented in dedicated ASICs (Application Specific Integrated Circuits) integrated in the detection plates. The output

The readout of each module is performed by two 32-pixel avalanche photodiodes that are optically coupled to each side of the module, as shown in Figure 12b). This double readout scheme allows the DOI measure. The DOI coordinate within the crystal is estimated from the asymmetry of the collected light at the top and bottom APD pixels. Experimental results have shown that, with this scheme, it is possible to obtain a 2 mm FWHM DOI resolution (53). This measurement is important since it increases the uniformity of measure all over the

. A scheme of a detector head is shown in Figure 12a). The distribution of the crystals within, the detector plates is as follows. Each detector plate is constituted of a set of four

a) b)

, placed side by side. Each supermodule is composed

, placed side by side. Each

FoV. Each detector head holds 96 detector

. A scheme

is constituted of a total of 3072 LYSO:Ce crystals, each crystal having 2×2×20 mm3

Figure 11: Representation of the Clear-PEM system.From (51).

structures named supermodules, each one with 14×4 cm2

module with the double readout scheme. From (52).

MAMMI (MAMmography with Molecular Imaging) is a dedicated breast PET, commercial‐ ized by Oncovision, which has certification in Europe and has recently received FDA approval [61].

Instead of the more often used small pixelated crystal arrays, this scanner uses continuous LYSO crystals coupled to PSPMTs.

The MAMMI prototype is a full ring PET consisting on 12 detector modules forming a dodecagon with a scanner aperture of 186 mm [12]. Each detector module uses 10 mm thick scintillation crystals with 40x40mm2 face. The back face of the crystals are coupled to the PS-PMTs. The detector has a 18% energy resolution at 511 keV, an intrinsic resolution of 1.6 mm and a DOI resolution of 4 mm.

The system aperture provides a 170 mm diameter FoV. The coincidences are allowed between a detector module and its seven opposite modules, providing a transaxial FoV aperture of 170 mm diameter. The axial FoV length is 40 mm per frame; an elevator allows the sequential move of the ring detector in step and shoot mode to increase the axial FoV to 170 mm. A timing window of 5 ns is used.

During image reconstruction the crystal faces are discretized into 2x2 mm2 pixels. Images are reconstructed with standard 3D ML-EM with a voxel size of 1x1x1 mm3 . The system matrix was obtained through calculation of the solid angle of every voxel with respect to the detection surface. Image reconstruction takes into account dead time, scatter and random events correction. A method for attenuation correction without a transmission scan has been devel‐ oped and tested with the system [62].

Patients are imaged in the prone position, which is possible due to the existence of a breast aperture in the patient table. This allows to image regions close to the chest wall without breast compression.

The preliminary results using a 1 mm in diameter 22Na point like source results of 1.6 mm FWHM axial and 1.9 transaxial in the center of the FoV and less than 3 mm FWHM in most other parts of the FoV (axial) [12, 63]. The system sensitivity at the FoV center and with a wide energy window of 250-759 keV is 1 %.

A pilot study was recently conducted with 32 invasive breast cancer patients to assess the feasibility of the system for tumor detection and characterization [64]. The system sensitivity for primary tumor visualization was 97%, which equals whole body PET-CT in the same study. This included lesions close to the chest wall.

#### **4. Conclusions**

With two commercially available PEM scanners, there is now data from several clinical studies that confirm the high sensitivity and specificity values for PEM in different clinical situations [65, 66]. These data also allow a direct comparison with other techniques such as PET and MRI and seem to indicate that the promise of molecular imaging with dedicated instruments as a valuable adjunct technique for mammography holds true.

#### **Author details**

Mónica Vieira Martins1,2

Address all correspondence to: mvmartins@estgp.pt

1 Polytechnic Institute of Portalegre, Portalegre, Portugal

2 Instituto de Biofísica e Engenharia Biomédica,Faculdade de Ciências, Universidade de Lisboa, Lisboa, Portugal

#### **References**

of the ring detector in step and shoot mode to increase the axial FoV to 170 mm. A timing

was obtained through calculation of the solid angle of every voxel with respect to the detection surface. Image reconstruction takes into account dead time, scatter and random events correction. A method for attenuation correction without a transmission scan has been devel‐

Patients are imaged in the prone position, which is possible due to the existence of a breast aperture in the patient table. This allows to image regions close to the chest wall without breast

The preliminary results using a 1 mm in diameter 22Na point like source results of 1.6 mm FWHM axial and 1.9 transaxial in the center of the FoV and less than 3 mm FWHM in most other parts of the FoV (axial) [12, 63]. The system sensitivity at the FoV center and with a wide

A pilot study was recently conducted with 32 invasive breast cancer patients to assess the feasibility of the system for tumor detection and characterization [64]. The system sensitivity for primary tumor visualization was 97%, which equals whole body PET-CT in the same study.

With two commercially available PEM scanners, there is now data from several clinical studies that confirm the high sensitivity and specificity values for PEM in different clinical situations [65, 66]. These data also allow a direct comparison with other techniques such as PET and MRI and seem to indicate that the promise of molecular imaging with dedicated instruments as a

2 Instituto de Biofísica e Engenharia Biomédica,Faculdade de Ciências, Universidade de

pixels. Images are

. The system matrix

During image reconstruction the crystal faces are discretized into 2x2 mm2

reconstructed with standard 3D ML-EM with a voxel size of 1x1x1 mm3

window of 5 ns is used.

56 Mammography Techniques and Review

compression.

**4. Conclusions**

**Author details**

Mónica Vieira Martins1,2

Lisboa, Lisboa, Portugal

oped and tested with the system [62].

energy window of 250-759 keV is 1 %.

This included lesions close to the chest wall.

valuable adjunct technique for mammography holds true.

Address all correspondence to: mvmartins@estgp.pt

1 Polytechnic Institute of Portalegre, Portalegre, Portugal


[25] Tafra L, Cheng Z, Uddo J, Lobrano MB, Stein W, Berg W a, et al. Pilot clinical trial of 18F-fluorodeoxyglucose positron-emission mammography in the surgical manage‐ ment of breast cancer. Am J Surg. 2005 Oct;190(4):628–32.

[14] Thompson C., Murthy K, Picard Y, Weinberg IN, Mako F. Positron Emission Mam‐ mography (PEM) - a promising technique for detecting breast cancer. IEEE Trans

[15] Bergman AM, Thompson C., Murthy K, Robar J., Clancy RL. Co-registration of posi‐ tron emission mammography (pem) images and x-ray mammograms. IEEE Nuclear

[16] Robar JL, Thompson CJ, Murthy IS, Clancy R, Bergman AM. Construction and cali‐ bration of detectors for high-resolution metabolic breast cancer imaging. Nucl Instru‐ ments Methods Phys Res - Sect A Accel Spectrometers Detect Assoc Equip.

[17] Robar J., Thompson CJ, Murthy K, Clancy RL, Bergman AM. Correction of spatial distortion, gain nonuniformity and efficiency variation in positron emission mam‐ mography. IEEE Nuclear Science Symposium - Conference Record. 1996. p. 1206–10.

[18] Murthy K, Aznar M, Thompson CJ, Loutfi A, Lisbona R, Gagnon JH. Results of Pre‐ liminary Clinical Trials of the Positron Emission Mammography System PEM-I: A Dedicated Breast Imaging System Producing Glucose Metabolic Images Using FDG. J

[19] Murthy K, Aznar M, Bergman AM, Thompson CJ, Robar JL, Lisbona R, et al. Positron Emission Mammographic Instrument: Initial Results. Radiology. 2000;215(1):280–5.

[20] Murthy K, Jolly D, Aznar M, Thompson CJ, Sciascia P, Loutfi A, et al. Quantification in positron emission mammography (PEM) with planar detectores: contrast resolu‐ tion measurements using a custom breast phantom and novel spherical hot spots.

[21] Wollenweber SD, Williams RC, Beylin D, Dolinsky S, Weinberg IN. Investigation of the Quantitative Capabilities of a Positron Emission Mammography System. IEEE

[22] Weinberg IN, Stepanov PY, Beylin D, Anashkin E, Lauckner K, Yarnall S, et al. PEM-2400 – a Biopsy-Ready PEM Scanner with Real-Time X-Ray Correlation Capa‐ bility. IEEE Nuclear Science Symposium - Conference Record. 2003. p. 1128–30. [23] Weinberg I, David B, Steve Y, Anashkin E, Dolinsky S, Zavarzin V, et al. Applica‐ tions of a PET Device with 1.5 mm FWHM Intrinsic Spatial Resolution to Breast Can‐ cer Imaging. IEEE International Symposium on Biomedical Imaging: from Macro to

[24] Levine EA, Freimanis RI, Perrier ND, Morton K, Lesko NM, Bergman S, et al. Posi‐ tron Emission Mammography: Initial Clinical Results. Ann Surg Oncol. 2003 Jan;

Nuclear Science Symposium - Conference Record. 2004. p. 8700–2.

Science Symposium - Conference Record. 1997. p. 1812–6.

Nucl Sci. 1995;42(4):1012–7.

58 Mammography Techniques and Review

1997;392(1-3):402–6.

Nucl Med. 1994;41(11):1851–8.

Nano. 2004. p. 1396–9.

10(1):86–91.

IEEE Trans Nucl Sci. 1999;46(6):2192–6.


[50] Brasse D, Kinahan PE, Clackdoyle R, Defrise M, Comtat C, Townsend DW. List-Mode Maximum-Likelihood Reconstruction Applied to Positron Emission Mammog‐ raphy (PEM) with Irregular Sampling. IEEE Trans Med Imaging. 2004;23(4):413–25.

[38] Wang G, Huber JS, Moses WW, Qi J, Member S, Choong W. Characterization of the

[39] Qi J, Kuo C, Huesman RH, Klein GJ, Moses WW, Member S, et al. Comparison of Rectangular and Dual-Planar Positron Emission Mammography Scanners. IEEE Nu‐

[40] Qi J, Klein GJ, Huesman RH, Member S. Image Properties of List-Mode Likelihood Reconstruction for a Rectangular Positron Emission Mammograph With DOI Meas‐

[41] Moses WW, Derenzo SE, Melcher CL, Manente RA. Room temperature LSO pin pho‐ todiode PET detector module that measures depth of interaction. IEEE Trans Nucl

[42] Wang G, Member S, Huber JS, Moses WW, Member S, Choong W, et al. Calibration of a PEM Detector With Depth of Interaction Measurement. IEEE Trans Nucl Sci.

[43] Virador PRG, Moses WW, Huesman RH, Zatsiorsky VM. Reconstruction in PET Cameras with Irregular Sampling and Depth of Interaction Capability \*. IEEE Trans

[44] Virador PRG, Moses WW, Huesman RH, Zatsiorsky VM. 3D Reconstruction in PET Cameras with Irregular Sampling and Depth of Interaction. IEEE Trans Med Imag‐

[45] Huesman RH, Klein GJ, Moses WW, Qi J, Reutter BW, Virador PRG. List-Mode Max‐ imum-Likelihood Reconstruction Applied to Positron Emission Mammography

[46] Qi J, Huesman RH. Lesion Detection and Quantitation of Positron Emission Mam‐ mography. IEEE Nuclear Science Symposium - Conference Record. 2002. p. 2248–52.

[47] Guerra A Del, Belcari N, Bencivelli W, Motta A, Righi S, Vaiano A, et al. Monte CAr‐ lo Study and Experimental Measurements of Breast Tumor Detectability with the YAP-PEMM prototype. IEEE Nuclear Science Symposium - Conference Record. 2003.

[48] Belcari N, Camarda M, Guerra A Del, Herbert D, Motta A, Vaiano A, et al. Develop‐ ment of a planar head PEM system based on an array of PSPMT and YAP crystals.

[49] Motta A, Guerra A del, Belcari N, Moehrs S, Panetta D, Righi S, et al. Fast 3D-EM reconstruction using Planograms for stationary planar positron emission mammog‐

IEEE Nuclear Science Symposium - Conference Record. 2004. p. 2179–82.

raphy camera. Comput Med imaging Graph. 2005 Dec;29(8):587–96.

(PEM) with Irregular Sampling. IEEE Trans Med Imaging. 2000;19(5):532–7.

LBNL PEM Camera. IEEE Trans Nucl Sci. 2006;53(3):1129–35.

clear Science Symposium - Conference Record. 2002. p. 1246–50.

urements. IEEE Trans Nucl Sci. 2001;48(4):1343–9.

Sci. 1995;42(4):1085–9.

60 Mammography Techniques and Review

2004;51(3):775–81.

ing. 2001;48(4):1524.

p. 1887–91.

Nucl Sci. 1998;45(3):1225–30.


Methods Phys Res Sect A Accel Spectrometers, Detect Assoc Equip. 2011 Aug; 648:S75–S78.


#### **Chapter 4 Provisional chapter**

#### **Digital Mammogram Enhancement Digital Mammogram Enhancement**

Michal Haindl and Václav Remeš Michal Haindl and Václav Remeš

Additional information is available at the end of the chapter Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/60988

#### **1. Introduction**

Methods Phys Res Sect A Accel Spectrometers, Detect Assoc Equip. 2011 Aug;

[63] Moliner L, Benlloch JM, Carles M, Correcher C, González AJ, Orero A, et al. Perform‐ ance Characteristics of the MAMMI PEMT Scanner Based on NEMA NU 2-2007. 2010 IEEE Nuclear Science Symposium Conference Record (NSS/MIC), . 2010. p. 2591–4.

[64] Koolen BB, Aukema TS, González AJ, Vogel W V., Caballero L, Peeters MJTFDV, et al. First clinical experience with a dedicated PET for hanging breast molecular imag‐

[65] Berg WA, Madsen KS, Schilling K, Pisano ED, Larsen LH, Ozonoff A, et al. Breast Cancer: Comparative Effectiveness of Positron Emission Mammography and MR

[66] Schilling K, Narayanan D, Kalinyak JE, The J, Velasquez MV, Kahn S, et al. Positron emission mammography in breast cancer presurgical planning: comparisons with magnetic resonance imaging. Eur J Nucl Med Mol Imaging. 2011 Jan;38(1):23–36.

ing. Q J Nucl Med Mol Imaging. 2013;57(1):92–100.

Imaging in Presurgical Planning. Radiology. 2011;258(1):59.

648:S75–S78.

62 Mammography Techniques and Review

Three fully automatic methods for X-ray digital mammogram enhancement based on a fast analytical textural model are presented. These efficient single and double view enhancement methods are based on the underlying two-dimensional adaptive causal autoregressive texture model. The methods locally predict breast tissue texture from single or double view mammograms and enhance breast tissue abnormalities, such as the sign of a developing cancer, using the estimated model prediction statistics. The double-view mammogram enhancement is based on the cross-prediction of two mutually registered left and right breasts' mammograms or alternatively a temporal sequence of mammograms. The single-view mammogram enhancement is based on modeling prediction error in case of not the both breasts' mammograms being available.

Breast cancer is the most common type of cancer among middle-aged women in most developed countries [1, 2]. Almost one woman in ten grows a breast cancer in her life. According to the American Cancer Society [3] about 232 670 new cases of invasive breast cancer will be diagnosed in women and about 40 000 women will die from breast cancer in US alone. US mortality rate is 30% and European mortality rate is 45% [4].

To lower the mortality rate, women in the developed countries usually regularly attend a preventive mammography screening. However, around 25% of radiologically visible cancers are missed by the radiologists at screening [5]. This means that millions of cancer cases are missed and therefore even a slightest improvement in the detection methods could have a huge impact and save many lives.

The biggest problem with current Computer-Aided Diagnosis (CAD) systems is their large false negative rate and an even larger false positive rate. Most CAD systems (e.g., [1, 6]) point out 2-3 regions of interest (ROIs) per mammogram on average. Taking into account that there are about 8 malignant mammograms in 1000 [5], the radiologists consider the current CAD systems as misleading.

©2012 prezimena autora, kod vise prvi et al., licensee InTech. This is an open access chapter distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. © 2015 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An alternative way is to automatically enhance mammograms to support radiologists with their visual mammogram evaluation. Several mammogram enhancement methods have been published [7–14]. Salvado and Roque [10] use wavelet analysis to detect microcalcifications, Dippel et al. [8] compare the merits of using either Laplacian pyramids or wavelet analysis for whole mammogram enhancement, Sakellaropoulos et al. [9] designed an adaptive wavelet based method for enhancing the contrast of the whole mammograms. Mencattini et al. [13] selectively enhance segmented mammograms regions using wavelet transformation.

An approach to diagnostic evaluation of screening mammograms based on local statistical Gaussian mixture textural models was proposed in [14]. The local evaluation tool has the form of a multivariate probability density of gray levels in a suitably chosen search window. First, the density function in the form of a Gaussian mixture is estimated from data obtained by scanning the mammogram with the search window. The estimated mixture is evaluated at each position and displays the corresponding log-likelihood value as a gray level at the window center. The resulting log-likelihood image closely correlates with the structural details of the original mammogram and emphasizes unusual places, but the method is very computationally demanding.

Radiologists regularly compare the bilateral mammogram pairs during mammogram screening in search for breast abnormalities. The mutual mammograms enhancement requires accurate registration of both breast X-ray images, which is difficult due to their elasticity. Marias et al. [15, 16] use thin-plate spline transformation [17] to align the breasts and then use wavelet based feature detection to find internal landmarks. Thin-plate spline based approach is also used by Wirth et al. in [18]. Hachama [19] deals only with the comparison of temporal mammograms based on a general method for registering images with the presence of abnormalities. However, it needs the prior abnormalities distribution knowledge. The registration and transformation are based on the Bayesian maximum a posteriori probability approach and minimization of the registration and deformation energy.

The novelty of our presented method is that whereas other alternative methods usually use simple pixel difference or trivial statistics like cross-correlation to compare the left and right images, we use the mammograms of one breast as a learning sample for the 2DCAR breast texture model [20, 21] and then try to analyze the other mammogram based on this acquired information. Using the 2DCAR model for bilateral comparison, we achieve a result which is robust to inaccurate registration, very fast, and which gives improved enhancement results compare to just a single-view analysis even using similar local texture modeling.

#### **2. Public mammogram databases**

There are not many publicly available mammogram databases [22–26], older databases like DDSM, MIAS are digitized from the X-ray films, while newer databases like INbreast are already digitaly acquired.

The Digital Database for Screening Mammography (DDSM) [24] http://marathon.csee.usf. edu/Mammography/Database.html is a database of digitized from original X-ray filmscreen in different resolutions and with associated ground truth and other information. This database was completed in 1999 and contains mammograms from four different sources using four different digitizers (DBA M2100 ImageClear, Howtek 960, Lumisys 200 Laser, Howtek MultiRad850) and 12 or 16 bits quantization. The database contains normal, benign, and histologically proven cancerous mammograms in four different views (left and right cranio-caudal (CC) and medio-lateral oblique (MLO)). It contains breast imaging reporting and data system (BI-RADS) keywords and the American College of Radiology (ACR) tissue codes (Table 1).


**Table 1.** ACR and BI-RADS codes.

2 ime knjige

computationally demanding.

**2. Public mammogram databases**

already digitaly acquired.

An alternative way is to automatically enhance mammograms to support radiologists with their visual mammogram evaluation. Several mammogram enhancement methods have been published [7–14]. Salvado and Roque [10] use wavelet analysis to detect microcalcifications, Dippel et al. [8] compare the merits of using either Laplacian pyramids or wavelet analysis for whole mammogram enhancement, Sakellaropoulos et al. [9] designed an adaptive wavelet based method for enhancing the contrast of the whole mammograms. Mencattini et al. [13]

An approach to diagnostic evaluation of screening mammograms based on local statistical Gaussian mixture textural models was proposed in [14]. The local evaluation tool has the form of a multivariate probability density of gray levels in a suitably chosen search window. First, the density function in the form of a Gaussian mixture is estimated from data obtained by scanning the mammogram with the search window. The estimated mixture is evaluated at each position and displays the corresponding log-likelihood value as a gray level at the window center. The resulting log-likelihood image closely correlates with the structural details of the original mammogram and emphasizes unusual places, but the method is very

Radiologists regularly compare the bilateral mammogram pairs during mammogram screening in search for breast abnormalities. The mutual mammograms enhancement requires accurate registration of both breast X-ray images, which is difficult due to their elasticity. Marias et al. [15, 16] use thin-plate spline transformation [17] to align the breasts and then use wavelet based feature detection to find internal landmarks. Thin-plate spline based approach is also used by Wirth et al. in [18]. Hachama [19] deals only with the comparison of temporal mammograms based on a general method for registering images with the presence of abnormalities. However, it needs the prior abnormalities distribution knowledge. The registration and transformation are based on the Bayesian maximum a posteriori probability approach and minimization of the registration and deformation energy. The novelty of our presented method is that whereas other alternative methods usually use simple pixel difference or trivial statistics like cross-correlation to compare the left and right images, we use the mammograms of one breast as a learning sample for the 2DCAR breast texture model [20, 21] and then try to analyze the other mammogram based on this acquired information. Using the 2DCAR model for bilateral comparison, we achieve a result which is robust to inaccurate registration, very fast, and which gives improved enhancement results

compare to just a single-view analysis even using similar local texture modeling.

There are not many publicly available mammogram databases [22–26], older databases like DDSM, MIAS are digitized from the X-ray films, while newer databases like INbreast are

The Digital Database for Screening Mammography (DDSM) [24] http://marathon.csee.usf. edu/Mammography/Database.html is a database of digitized from original X-ray filmscreen in different resolutions and with associated ground truth and other information. This database was completed in 1999 and contains mammograms from four different sources using four different digitizers (DBA M2100 ImageClear, Howtek 960, Lumisys 200 Laser, Howtek MultiRad850) and 12 or 16 bits quantization. The database contains normal, benign,

selectively enhance segmented mammograms regions using wavelet transformation.

The Mammographic Image Analysis Society Digital Mammogram Database (miniMIAS) [22] is also digitized to 50 microns per pixel from the original X-ray filmscreen mammograms by the scanning microdensitometer SCANDIG3. MIAS mammographic images are available via the Pilot European Image Processing Archive (PEIPA) at the University of Essex http: //peipa.essex.ac.uk/ipa/info/mias.html.

The LLNL/UCSF database ftp://gdo-biomed.ucllnl.org/pub/mammo-db/ [23] contains 198 digitized films from 50 patients with 4 views per patient (but only 2 views from one mastectomy case).

The INbreast database [26] is a mammographic database, with images acquired at a Breast Centre, located in a University Hospital (Hospital de São João, Breast Centre, Porto, Portugal). INbreast has a total of 115 cases (410 images) of which 90 cases are from women with both breasts (4 images per case) and 25 cases are from mastectomy patients (2 images per case). Several types of lesions (masses, calcifications, asymmetries, and distortions) are included. Accurate contours made by specialists are also provided in the XML format.

The recent BancoWeb LAPIMO Database http://lapimo.sel.eesc.usp.br/bancoweb/ [27] was acquired in two hospitals using Senographe 500t and Senographe 600t mammographs and digitized by using two laser scanners Lumiscan 50 and Lumiscan 75.

The overview of major features of the public mammographic databases are listed in the following Table 2.

#### **3. Mammogram enhancement methods**

The mammogram enhancement methods can be roughly categorized into frequency based and spatial based methods. The frequency based methods [7, 9, 28, 29] use mostly some wavelet multiscale decomposition with modified wavelet coefficients to enhance mammogram contrast. The spatial methods [14, 30] use some nonlinear or adaptive linear filters.

We have implemented four representative mammogram enhancement methods from several published alternatives [7–10, 10–14] to compare with our novel adaptive probabilistic mammogram enhancement method.


**Table 2.** Public Mammogram Databases: where (*nmam*) is the number of mammograms, (*nviews*) number of views, (*ngl*) number of gray levels in bits, and ↓is benign without callback.

#### **3.1. Histogram equalization**

The well known gray scale image enhancement technique is histogram equalization [31], which is based on the idea of forcing the enhanced image histogram to be uniform. This is a popular technique for contrast enhancement because because of its simplicity and effectivity. However, it may overenhance the noises and sharp regions in the original images.

#### **3.2. Matting-based enhancement**

The enhancement method based on the idea of image matting was published in [32]. It works based on the idea that mammographic images (*Y*) are a superposition of some background adipose tissue (*B*) and the interesting part, which would be the mammary glands and other breast structures (*G*).

$$Y = \mathbf{G}\mathbf{c} + B(\mathbf{1} - \mathbf{c}) \tag{1}$$

The enhancement method then selectively subtracts the background tissues from the superposition, thus creating the enhanced image.

To enable this, the authors had to estimate the background (*B*) and the opacity alpha value for each pixel by which it is blended with the rest of the image (*c*). In this method the background is set as a constant value for the whole image represented by the 85% percentile of grey values of the breast part of the image.

#### **3.3. Nonlinear unsharp masking**

A nonlinear unsharp masking (NLUM) combined with nonlinear filtering for mammogram enhancement was introduced in [33]. The method embeds different types of filters into the nonlinear filtering operator within the 3 × 3 window which fuses the enhanced and original mammogram data. The unsharp masking emphasizes high-frequencies of the signal either by subtracting a low-pass filtered signal from its original or adding a scaled high-frequency factor to the measured original. NLUM eight parameters are optimized using the proposed second-derivative-like measure of enhancement (SDME) [33].

#### **3.4. Direct contrast enhancement**

4 ime knjige

DDSM ([24])

INbreast ([26])

x *resolution* 1411 − 5641 2560 − 3328 1024 y *resolution* 3256 − 7111 3328 − 4084 1024

*benign* 870 44 66

(*ngl*) number of gray levels in bits, and ↓is benign without callback.

**3.1. Histogram equalization**

**3.2. Matting-based enhancement**

superposition, thus creating the enhanced image.

of grey values of the breast part of the image.

second-derivative-like measure of enhancement (SDME) [33].

**3.3. Nonlinear unsharp masking**

breast structures (*G*).

miniMIAS ([22])

*nmam* 10480 410 322 198 1473 *nviews* 4 4 2 4 4 *ngl* 16/12 14 8 12 12

*normal* 695 70 204 38 294 ↓ *benign* 141 116 128 994

*malignant* 914 180 52 32 112 *density ACR ACR own scale no no BI-RADS yes yes no no yes*

**Table 2.** Public Mammogram Databases: where (*nmam*) is the number of mammograms, (*nviews*) number of views,

The well known gray scale image enhancement technique is histogram equalization [31], which is based on the idea of forcing the enhanced image histogram to be uniform. This is a popular technique for contrast enhancement because because of its simplicity and effectivity.

The enhancement method based on the idea of image matting was published in [32]. It works based on the idea that mammographic images (*Y*) are a superposition of some background adipose tissue (*B*) and the interesting part, which would be the mammary glands and other

The enhancement method then selectively subtracts the background tissues from the

To enable this, the authors had to estimate the background (*B*) and the opacity alpha value for each pixel by which it is blended with the rest of the image (*c*). In this method the background is set as a constant value for the whole image represented by the 85% percentile

A nonlinear unsharp masking (NLUM) combined with nonlinear filtering for mammogram enhancement was introduced in [33]. The method embeds different types of filters into the nonlinear filtering operator within the 3 × 3 window which fuses the enhanced and original mammogram data. The unsharp masking emphasizes high-frequencies of the signal either by subtracting a low-pass filtered signal from its original or adding a scaled high-frequency factor to the measured original. NLUM eight parameters are optimized using the proposed

However, it may overenhance the noises and sharp regions in the original images.

LLNL ([23])

*Y* = *Gc* + *B*(1 − *c*) (1)

LAPIMO ([27] )

An enhancement method based on wavelet transformation was described in [29]. The method performs a multi-level 2D wavelet transfomation and at each level the 3 highpass components are divided by the lowpass-lowpass component, getting a directional contrast estimate, which is further multiplied by a constant contrast enhancement factor *λ*. Starting at the deepest level of the transform, the inverse transform is performed one scale at a time. For each scale level, before the inverse transform step, the 3 modified components are multiplied by the newly computed lowpass component. This way the authors achieve a contrast enhancement without the introduction of too much of additional noise.

#### **4. Probabilistic mammogram enhancement**

These our methods use Markovian texture models for the analysis of local texture characteristics and enhancing breast tissue abnormalities such as microcalcifications and masses which could be the sign of a developing cancer. We make the presumption that left and right breasts are architecturally symmetrical. This presumption is indeed reasonable, since radiologists frequently compare double-view mammograms to find asymmetrical parts, which could indicate a developing cancer. The texture based symmetry detection neither needs to assume the pixel-wise correspondence of the both breast images, nor their ideal sub-pixel registration inside the breast area.

The double-view methods consist of three major steps: registration, model parameters adaptive estimation, and the cross-prediction based analysis.

#### **4.1. Mammogram registration**

The registration process is described for mammographic MLO views, but it can be easily adapted also for CC views. Since we compare the images based on textural features rather than pixel-wise, we do not require as precise registration as other methods, and can use a simple registration based on the affine transformation.

Three reference points are needed for the affine transformation (Figure 1). We chose the nipple and one point above and one below that are closest to the pectoral muscle.

**Figure 1.** Registered mammograms with visible reference points.

The nipple is located using the heuristic method described in [34]. It works on the idea of the nipple being a point on the skin-line of the breast which is the most distant from the line of the pectoral muscle. After the candidates for the nipple reference points have been found in both the mammograms, the position of the reference point can still slightly differ in both images. Therefore, we adjust their position by searching the neighborhood on the skin line of the breast for the most correlated window.

The remaining reference point candidates have to be further adjusted as well. Since the bilateral mammograms usually do not cover the same area of the breast, some anatomical parts of the breast can be seen only in one of the images and therefore the reference points wouldn't match. To make up for this problem, we measure the distance of the points to the nipple, weighted by the nipples distance to the pectoral muscle. The weighting compensates for the differences of positioning of the breast in the mammogram which could result in one image displaying the breast bigger than the other. We then adjust the corresponding reference points, so that they are on the skin line with the most similar weighted distance to the nipple possible.

Having found the reference points, the affine transformation is performed. Figure 1 in the leftmost images shows the images of right and left breast with marked line of the pectoral muscle (colored in red) and the distance from the pectoral muscle to the nipple. The rightmost images show the registered breasts with the reference points painted as white squares with the right breast (shown on the left side) transformed to match the left breast.

#### **4.2. Adaptive textural model**

The X-ray mammographic tissue is locally modeled by its dedicated independent Gaussian noise-driven autoregressive random field two-dimensional texture model (2DCAR), which is a rare exception among Markovian random field model family that can be completely analytically solved [35, 36]. Apart from that, this descriptive model has good modeling performance, all statistics can be evaluated recursively, and the model is very fast to evaluate.

The 2DCAR random field is a Markovian family of random variables with a joint probability density on the set of all possible realizations *Y* of the *M* × *N* lattice *I*, subject to the following condition:

$$\begin{split} p(Y \mid \gamma, \sigma^{-2}) &= (2\pi\sigma^2)^{-\frac{(MN-1)}{2}} \\ &\exp\left\{ \frac{-1}{2} \text{tr}\left\{ \sigma^{-2} \begin{pmatrix} -a \\ \gamma^T \end{pmatrix}^T \tilde{V}\_{MN-1} \begin{pmatrix} -a \\ \gamma^T \end{pmatrix} \right\} \right\} \end{split} \tag{2}$$

where *α* is a unit vector, *tr*() is a trace of the corresponding matrix, and the following notation is used

$$\begin{aligned} \tilde{V}\_{r-1} &= \sum\_{k=1}^{r-1} \begin{pmatrix} Y\_k \ Y\_k^T & Y\_k X\_k^T \\ X\_k Y\_k^T & X\_k X\_k^T \end{pmatrix} \\ &= \begin{pmatrix} \tilde{V}\_{y(r-1)} & \tilde{V}\_{xy(r-1)}^T \\ \tilde{V}\_{xy(r-1)} & \tilde{V}\_{x(r-1)} \end{pmatrix} \end{aligned}$$

Here, *r* = [*r*1,*r*2, *φ*] is spatial multiindex denoting history of movements on the rectangular lattice *I*, where *r*1,*r*<sup>2</sup> are row and column indices, and the direction of the model development is *φ* ∈ {0◦, 45◦, 90◦, 135◦, 180◦, 225◦, 270◦, 315◦} . The 2DCAR model can be expressed as a stationary causal uncorrelated noise-driven 2D autoregressive process:

$$Y\_r = \gamma\_\phi X\_r + \varepsilon\_r \tag{3}$$

where *γφ* = [*a*1,..., *aη*] is the parameter vector, *η* = *cardinality*(*I<sup>c</sup> <sup>r</sup>* ), *I<sup>c</sup> <sup>r</sup>* denotes a causal (or alternatively unilateral) contextual neighborhood (i.e., all support pixels were previously visited and thus they are known). Elements in *I<sup>c</sup> <sup>r</sup>* do not need to be topological neighbours of each other, i.e., if *s* is a neighbour of *r* then ∃*t*, *t* ∈ *I* located between *r* and *s* at a distance *<sup>δ</sup>*(*r*, *<sup>t</sup>*) *<sup>&</sup>lt; <sup>δ</sup>*(*r*,*s*) such as *<sup>t</sup>* <sup>∈</sup>/ *<sup>I</sup><sup>c</sup> <sup>r</sup>* . This type of a neighbourhood system is also called a functional neighbourhood system and its application is illustrated in Figure 2. Its optimal configuration can be found analytically using the Bayesian statistics see [36] for details. Furthermore, *er* denotes white Gaussian noise with zero mean and a constant but unknown variance *<sup>σ</sup>*2, and *Xr* is a support vector of *Yr*−*<sup>s</sup>* where *<sup>s</sup>* <sup>∈</sup> *<sup>I</sup><sup>c</sup> <sup>r</sup>* . The method uses a locally adaptive version of this 2DCAR model [36], where its recursive statistics are modified by an exponential forgetting factor, i.e., a constant smaller than 1 which is used to weight the older data.

#### *4.2.1. Parameter estimation*

Parameter estimation of the 2DCAR model using either the maximum likelihood, the least square or Bayesian methods can be found analytically. The Bayesian parameter estimates of the 2DCAR model using the normal-gamma parameter prior are:

$$
\gamma\_{r-1}^T = V\_{x(r-1)}^{-1} V\_{xy(r-1)} \quad \text{ } \tag{4}
$$

$$
\theta\_{r-1}^2 = \frac{\lambda\_{(r-1)}}{\beta(r)} \; \; \; \tag{5}
$$

where

6 ime knjige

the nipple possible.

condition:

is used

**4.2. Adaptive textural model**

*<sup>p</sup>*(*<sup>Y</sup>* <sup>|</sup> *<sup>γ</sup>*, *<sup>σ</sup>*−2)=(2*πσ*2)<sup>−</sup> (*MN*−1)

exp −1 <sup>2</sup> *tr*

*V*˜ *<sup>r</sup>*−<sup>1</sup> =

of the breast for the most correlated window.

The nipple is located using the heuristic method described in [34]. It works on the idea of the nipple being a point on the skin-line of the breast which is the most distant from the line of the pectoral muscle. After the candidates for the nipple reference points have been found in both the mammograms, the position of the reference point can still slightly differ in both images. Therefore, we adjust their position by searching the neighborhood on the skin line

The remaining reference point candidates have to be further adjusted as well. Since the bilateral mammograms usually do not cover the same area of the breast, some anatomical parts of the breast can be seen only in one of the images and therefore the reference points wouldn't match. To make up for this problem, we measure the distance of the points to the nipple, weighted by the nipples distance to the pectoral muscle. The weighting compensates for the differences of positioning of the breast in the mammogram which could result in one image displaying the breast bigger than the other. We then adjust the corresponding reference points, so that they are on the skin line with the most similar weighted distance to

Having found the reference points, the affine transformation is performed. Figure 1 in the leftmost images shows the images of right and left breast with marked line of the pectoral muscle (colored in red) and the distance from the pectoral muscle to the nipple. The rightmost images show the registered breasts with the reference points painted as white squares with the right breast (shown on the left side) transformed to match the left breast.

The X-ray mammographic tissue is locally modeled by its dedicated independent Gaussian noise-driven autoregressive random field two-dimensional texture model (2DCAR), which is a rare exception among Markovian random field model family that can be completely analytically solved [35, 36]. Apart from that, this descriptive model has good modeling performance, all statistics can be evaluated recursively, and the model is very fast to evaluate. The 2DCAR random field is a Markovian family of random variables with a joint probability density on the set of all possible realizations *Y* of the *M* × *N* lattice *I*, subject to the following

2

*r*−1 ∑ *k*=1

*V*˜

= *V*˜

 *σ*−<sup>2</sup> −*α γT <sup>T</sup> V*˜ *MN*−1

where *α* is a unit vector, *tr*() is a trace of the corresponding matrix, and the following notation

*YkY<sup>T</sup>*

*XkY<sup>T</sup>*

*<sup>y</sup>*(*r*−1) *<sup>V</sup>*˜ *<sup>T</sup>*

*xy*(*r*−1) *<sup>V</sup>*˜

*<sup>k</sup> YkX<sup>T</sup> k*

 .

*<sup>k</sup> XkX<sup>T</sup> k*

*xy*(*r*−1)

*x*(*r*−1)

 −*α γT*

, (2)

$$\begin{aligned} \lambda\_{(r-1)} &= V\_{y(r-1)} - V\_{xy(r-1)}^T V\_{x(r-1)}^{-1} V\_{xy(r-1)} \\ V\_{(r-1)} &= \tilde{V}\_{(r-1)} + V\_{(0)} \\ \beta(r) &= \beta(0) + r - 1 \end{aligned}$$

and *β*(0) is an initialization constant and submatrices in *V*(0) are from the parameter prior. The parameter estimates (4),(5) can also be evaluated recursively [36] using the proces history (Y(*r*−1)). The posterior probability density [36] of the model is:

**Figure 2.** Single-view MLO mammogram enhancement using different functional neighbourhoods consecutively rightwards - 9, 5, and 3 pixel neighbourhood distance from the enhanced pixel (blue pixels - bottom row).

$$p(Y\_r \mid Y^{(r-1)}, \hat{\gamma}\_{r-1}) = \frac{\Gamma(\frac{\beta(r) - \eta + 3}{2})}{\Gamma(\frac{\beta(r) - \eta + 2}{2}) \, \pi^{\frac{1}{2}} \, (1 + X\_r^T V\_{\mathbf{x}(r-1)}^{-1} X\_r)^{\frac{1}{2}} \, |\, \lambda\_{(r-1)}|^{\frac{1}{2}}}$$

$$\left(1 + \frac{(Y\_r - \hat{\gamma}\_{r-1} X\_r)^T \lambda\_{(r-1)}^{-1} (Y\_r - \hat{\gamma}\_{r-1} X\_r)}{1 + X\_r^T V\_{\mathbf{x}(r-1)}^{-1} X\_r}\right)^{-\frac{\beta(r) - \eta + 3}{2}}\tag{6}$$

And the conditional mean value predictor of the one-step-ahead predictive posterior density (6) for the normal-gamma parameter prior is

$$E\left\{Y\_r \mid Y^{(r-1)}\right\} = \hat{\gamma}\_{r-1} X\_r \ . \tag{7}$$

**Figure 3.** INbreast MLO mammogram enhancement comparison rightwards: the original mammogram, histogram equalization, [32], [33], [29], and the presented enhancement methods, respectively.

#### *4.2.2. Prediction*

8 ime knjige

**Figure 2.** Single-view MLO mammogram enhancement using different functional neighbourhoods consecutively

(*Yr* <sup>−</sup> *<sup>γ</sup>*ˆ*r*−1*Xr*)*Tλ*−<sup>1</sup>

And the conditional mean value predictor of the one-step-ahead predictive posterior density

<sup>2</sup> )

*<sup>r</sup> V*−<sup>1</sup> *x*(*r*−1)

(*r*−1)

*<sup>r</sup> V*−<sup>1</sup> *x*(*r*−1) *Xr*

*Xr*) 1 <sup>2</sup> |*λ*(*r*−1)|

(*Yr* − *<sup>γ</sup>*ˆ*r*−1*Xr*)

1 2

<sup>−</sup> *<sup>β</sup>*(*r*)−*η*+<sup>3</sup> 2

(6)

 

= *<sup>γ</sup>*ˆ*r*−1*Xr* . (7)

<sup>2</sup> (1 + *X<sup>T</sup>*

1 + *X<sup>T</sup>*

rightwards - 9, 5, and 3 pixel neighbourhood distance from the enhanced pixel (blue pixels - bottom row).

, *<sup>γ</sup>*ˆ*r*−1) = <sup>Γ</sup>( *<sup>β</sup>*(*r*)−*η*+<sup>3</sup>

<sup>2</sup> ) *<sup>π</sup>* <sup>1</sup>

*Yr* <sup>|</sup> *<sup>Y</sup>*(*r*−1)

Γ( *<sup>β</sup>*(*r*)−*η*+<sup>2</sup>

 1 +

> *E*

*<sup>p</sup>*(*Yr* <sup>|</sup> *<sup>Y</sup>*(*r*−1)

(6) for the normal-gamma parameter prior is

The conditional mean value of the one-step-ahead predictive posterior density for the normal-gamma parameter prior is

$$E\left\{\mathbf{Y}\_{r}\mid\mathbf{Y}^{(r-1)}\right\}=\hat{\gamma}\_{r-1}X\_{r}\ .\tag{8}$$

The predictor (8) is used only for single-view mammogram enhancement. For double-view mammograms where there are available both left and right breasts mammograms the method uses the cross-prediction (10),(11).

**Figure 4.** MIAS MLO mammogram enhancement comparison rightwards: the original mammogram, histogram equalization, [32], [33], [29], and the presented enhancement methods, respectively.

#### **4.3. Enhancement methods**

Let us denote two mutually registered (e.g., left and right breasts') mammograms *Y* and *Y*˜, the local 2DCAR model parameters estimates (4), (5) computed on the mammogram image *Y γ*ˆ *<sup>T</sup> <sup>r</sup>*−1, *<sup>σ</sup>*<sup>ˆ</sup> <sup>2</sup> *<sup>r</sup>*−1. The same parameter estimates (4), (5) computed on the other mammogram *<sup>Y</sup>*˜ are denoted *γ*˜ *<sup>T</sup> <sup>r</sup>*−1, *<sup>σ</sup>*˜ <sup>2</sup> *<sup>r</sup>*−1, and the corresponding support vector is *<sup>X</sup>*˜*r*. The directional models are computed in the following angles *φ* ∈ Φ = {0◦, 45◦, 90◦, 135◦, 180◦, 225◦, 270◦, 315◦}.

#### *4.3.1. Single-view enhancement*

The single-view enhacement method is computed from up to eight directional models, i.e.,

**Figure 5.** INbreast multiple-view MLO mammogram enhancement consecutively rightwards - ground truth, pixel difference between registered LMLO and RMLO, cross-predicted gradient, and cross-prediction probability density. The upper row contains LMLO, bottom row RMLO.

$$\mathcal{Y}\_r^{\text{enh}} = \sum\_{\forall \phi \in \Phi} (\mathcal{Y}\_{r+1} - \hat{\gamma}\_{r-1} \mathcal{X}\_r) \quad , \tag{9}$$

where <sup>Φ</sup>¯ <sup>⊆</sup> <sup>Φ</sup>. All the enhanced values are normalized into the 0 <sup>−</sup> 255 range.

#### *4.3.2. Double-view enhancement*

10 ime knjige

**Figure 4.** MIAS MLO mammogram enhancement comparison rightwards: the original mammogram, histogram

Let us denote two mutually registered (e.g., left and right breasts') mammograms *Y* and *Y*˜, the local 2DCAR model parameters estimates (4), (5) computed on the mammogram image

are computed in the following angles *φ* ∈ Φ = {0◦, 45◦, 90◦, 135◦, 180◦, 225◦, 270◦, 315◦}.

The single-view enhacement method is computed from up to eight directional models, i.e.,

*<sup>r</sup>*−1. The same parameter estimates (4), (5) computed on the other mammogram *<sup>Y</sup>*˜

*<sup>r</sup>*−1, and the corresponding support vector is *<sup>X</sup>*˜*r*. The directional models

equalization, [32], [33], [29], and the presented enhancement methods, respectively.

**4.3. Enhancement methods**

*<sup>r</sup>*−1, *<sup>σ</sup>*˜ <sup>2</sup>

*4.3.1. Single-view enhancement*

*Y γ*ˆ *<sup>T</sup>*

*<sup>r</sup>*−1, *<sup>σ</sup>*<sup>ˆ</sup> <sup>2</sup>

are denoted *γ*˜ *<sup>T</sup>*

The double-view enhancement is based on statistics computed on one breast image and applied to the complementary one. The cross-prediction between images *Y*,*Y*˜ is computed as follows:

$$E\left\{\tilde{Y}\_r \mid Y^{(r-1)}\right\} = \hat{\gamma}\_{r-1}\tilde{X}\_r\tag{10}$$

and the opposite direction cross-prediction is analogously

$$E\left\{\mathbf{Y}\_{r}\mid\tilde{\mathbf{Y}}^{(r-1)}\right\}=\tilde{\gamma}\_{r-1}\mathbf{X}\_{r}\;.\tag{11}$$

The enhanced mammograms are then the corresponding cross-prediction statistics images. The corresponding cross-prediction probability densities are *p*(*Y*˜ *<sup>r</sup>* <sup>|</sup> *<sup>Y</sup>*˜(*r*−1), *<sup>γ</sup>*ˆ*r*−1) and *<sup>p</sup>*(*Yr* <sup>|</sup> *<sup>Y</sup>*(*r*−1), *<sup>γ</sup>*˜*r*−1).

The proposed double-view enhancement methods are

$$Y\_r^{com\_1} = \sum\_{\forall \phi \in \Phi} \left( \tilde{Y}\_{r+1} - \hat{\gamma}\_{r-1} X\_r \right) \quad , \tag{12}$$

$$Y\_r^{com\_2} = \sum\_{\forall \phi \in \Phi} p(\tilde{Y}\_r \mid \tilde{Y}^{(r-1)}, \hat{\gamma}\_{r-1}) \ . \tag{13}$$

#### **5. Experimental results**

The comparative experimental results (Figures 3, 4) were tested on the miniMIAS database [22] and on the state-of-the-art public digital mammogram INbreast database [26]. Comparing the alternative methods (Section 3) with our proposed adaptive enhancement, it is clearly visible that whereas these methods enhance prevailingly contrast, our method enhances textural abnormalities in the breast tissue which is more useful for the radiologists.

Our adaptive enhancement methods were also successfully tested on the Digital Database for Screening Mammography (DDSM) from the University of South Florida [24]. These results are reported elsewhere.

The spatial textural model allows seamless and natural generalization into multiple-view mammogram enhancement (be it bilateral, as presented in Figures 5, 6, or temporal). Double-view medio-lateral oblique digital mammograms' enhancements from the INbreast database (Figure 5) and the miniMIAS database (Figure 6) show the cross-prediction based enhancement performance. Comparing the cross-prediction enhancements on Figures 5, 6 respectively with the same breast single-view enhancements on Figures 3, 4, the benefits of the cross-prediction are clearly visible.

Both our double-view enhancement methods are compared with the registered image pixel difference which is standardly used for comparison ([15, 18, 19])

$$
\Delta Y\_r = \max \{ Y\_r^R - Y\_r^L, 0 \} \ . \tag{14}
$$

This standard double-view enhancement method (Figures 5, 6 - second columns) is inferior compared to the both proposed double-view enhancement methods ((12), (13)) which simultaneously exhibit more contrast and increased details' visibility.

Finally, all three proposed enhancement methods are very fast - they can be computed on the presented mammograms with a standard PC in a matter of several seconds.

**Figure 6.** Multiple-view medio-lateral mammogram (INbreast) enhancement consecutively rightwards - ground truth, pixel difference between registered LMLO and RMLO, cross-predicted gradient, and cross-prediction probability density. The upper row contains LMLO, bottom row RMLO.

#### **6. Conclusions**

12 ime knjige

*<sup>p</sup>*(*Yr* <sup>|</sup> *<sup>Y</sup>*(*r*−1), *<sup>γ</sup>*˜*r*−1).

**5. Experimental results**

are reported elsewhere.

the cross-prediction are clearly visible.

difference which is standardly used for comparison ([15, 18, 19])

simultaneously exhibit more contrast and increased details' visibility.

and the opposite direction cross-prediction is analogously

The proposed double-view enhancement methods are

*E* 

The corresponding cross-prediction probability densities are *p*(*Y*˜

*<sup>Y</sup>com*<sup>1</sup> *<sup>r</sup>* <sup>=</sup> ∑

*<sup>Y</sup>com*<sup>2</sup> *<sup>r</sup>* <sup>=</sup> ∑

<sup>∀</sup>*φ*∈Φ¯

<sup>∀</sup>*φ*∈Φ¯

 *Y*˜

*p*(*Y*˜

The comparative experimental results (Figures 3, 4) were tested on the miniMIAS database [22] and on the state-of-the-art public digital mammogram INbreast database [26]. Comparing the alternative methods (Section 3) with our proposed adaptive enhancement, it is clearly visible that whereas these methods enhance prevailingly contrast, our method enhances textural abnormalities in the breast tissue which is more useful for the radiologists. Our adaptive enhancement methods were also successfully tested on the Digital Database for Screening Mammography (DDSM) from the University of South Florida [24]. These results

The spatial textural model allows seamless and natural generalization into multiple-view mammogram enhancement (be it bilateral, as presented in Figures 5, 6, or temporal). Double-view medio-lateral oblique digital mammograms' enhancements from the INbreast database (Figure 5) and the miniMIAS database (Figure 6) show the cross-prediction based enhancement performance. Comparing the cross-prediction enhancements on Figures 5, 6 respectively with the same breast single-view enhancements on Figures 3, 4, the benefits of

Both our double-view enhancement methods are compared with the registered image pixel

This standard double-view enhancement method (Figures 5, 6 - second columns) is inferior compared to the both proposed double-view enhancement methods ((12), (13)) which

Finally, all three proposed enhancement methods are very fast - they can be computed on

*<sup>r</sup>* <sup>−</sup> *<sup>Y</sup><sup>L</sup>*

<sup>∆</sup>*Yr* <sup>=</sup> *max*{*Y<sup>R</sup>*

the presented mammograms with a standard PC in a matter of several seconds.

*<sup>r</sup>*+<sup>1</sup> − *<sup>γ</sup>*ˆ*r*−1*Xr*

*<sup>r</sup>* <sup>|</sup> *<sup>Y</sup>*˜(*r*−1)

*Yr* <sup>|</sup> *<sup>Y</sup>*˜(*r*−1)

The enhanced mammograms are then the corresponding cross-prediction statistics images.

= *<sup>γ</sup>*˜*r*−1*Xr* . (11)

*<sup>r</sup>* <sup>|</sup> *<sup>Y</sup>*˜(*r*−1), *<sup>γ</sup>*ˆ*r*−1) and

, (12)

, *<sup>γ</sup>*ˆ*r*−1) . (13)

*<sup>r</sup>* , 0} . (14)

We proposed three novel fast methods for completely automatic mammogram enhancement which highlight regions of interest, detected as textural abnormalities. Cancerous areas typically manifest themselves in X-ray mammography as such textural defects which is advantageous for our methods in comparison with most alternative mammogram enhancement methods that primarily enhance only the image contrast. Thus the enhanced mammograms can help radiologists to decrease their false negative evaluation rate.

These methods are based on the underlying two-dimensional adaptive CAR texture model. Although the algorithms use random field type model, the model is very fast due to the efficient recursive model predictor estimation and therefore is much faster than the usual alternative Markov chain Monte Carlo estimation approach. The enhancement can be either single or double view depending on the available data. The single-view methods allow significant mammogram enhancement without the need of paired mammogram registration. The double-view methods benefit from mutual textural information in the registered bilateral breast pairs. Contrary to the simple pixel difference values or cross-correlations, the textural feature comparison brings increased robustness to registration inaccuracies inevitably encountered due to the elasticity of the breast. The double-view methods could alternatively be used for the enhancement of a temporal sequence of mammograms.

#### **Acknowledgements**

This research was supported by the Czech Science Foundation project GACR 14-10911S. ˇ

#### **Author details**

Michal Haindl and Václav Remeš

The Institute of Information Theory and Automation of the Czech Academy of Sciences, Prague, Czech Republic

#### **References**


[11] Yan, Z.; Zhang, Y.; Liu, B.; Zheng, J.; Lu, L.; Xie, Y.; Liang, Z.; Li, J. Extracting hidden visual information from mammography images using conjugate image enhancement software. In *Information Acquisition, 2005 IEEE International Conference on*; IEEE: 2005.

14 ime knjige

**Acknowledgements**

Michal Haindl and Václav Remeš

Alamitos, CA, USA, 2002.

[4] Kopans, D. B. *Radiology* 2010, *256,* 15–20.

*Technology Assessment* 2005, *9,*.

*International Workshop on*; 2005.

**Author details**

**References**

Prague, Czech Republic

society, 2012.

*3,* 32–46.

*48,* 787.

2002, *21,* 343–353.

inevitably encountered due to the elasticity of the breast. The double-view methods could

alternatively be used for the enhancement of a temporal sequence of mammograms.

This research was supported by the Czech Science Foundation project GACR 14-10911S.

The Institute of Information Theory and Automation of the Czech Academy of Sciences,

[1] Tweed, T.; Miguet, S. Automatic Detection of Regions of Interest in Mammographies Based on a Combined Analysis of Texture and Histogram. In *Pattern Recognition, 2002. Proceedings. 16th International Conference on*, Vol. 2; IEEE Computer Society: Los

[2] Qi, H.; Diakides, N. A. Thermal Infrared Imaging in Early Breast Cancer Detection - A Survey of Recent Research. In *Proceedings of the 25th Annual International Conference of the*

[3] ACS, "Breast cancer facts & figures 2011 -? 2012", Technical Report, American cancer

[5] Taylor, P. and Champness, J. and Given-Wilson, R. and Johnston, K. and Potts, H., *Health*

[6] Haindl, M.; Mikeš, S.; Scarpa, G. Unsupervised Detection of Mammogram Regions of Interest. In *Knowledge-Based Intelligent Information and Engineering Systems*, Vol. 4694;

[7] Chang, C.-M.; Laine, A. *IEEE Transactions on Information Technology in Biomedicine* 1999,

[8] Dippel, S.; Stahl, M.; Wiemker, R.; Blaffert, T. *Medical Imaging, IEEE Transactions on*

[9] Sakellaropoulos, P.; Costaridou, L.; Panayiotakis, G. *Physics in medicine and biology* 2003,

[10] Salvado, J.; Roque, B. Detection of calcifications in digital mammograms using wavelet analysis and contrast enhancement. In *Intelligent Signal Processing, 2005 IEEE*

*IEEE Engineering in Medicine and Biology Society*, Vol. 2; 2003.

Apolloni, B.; Howlett, R. J.; Jain, L., Eds.; Springer: 2007.

ˇ


**Chapter 5**
