**4. Image quality in mammography**

Image quality is a fundamental concept for the control and optimisation of mammography; it aims to improve the early detection of cancer and other pathological lesions in the breasts. The image quality can be quantified by measuring the contrast noise ratio, the signal noise ratio, the modulation transfer function (MTF) (spatial resolution), the noise, the uniformity and various artefacts such as the ghost image. In order to assess these physical parameters, phantoms and more specific devices are needed to perform quality control in mammography. The current trend is to utilise a contrast-detail phantom (CDMAM) which permits the assessment of the image quality as a function of the contrast threshold associated with the circular objects of different diameters and thickness (SEFM, 2008). The "clinical "assessment of the image quality is better performed by means of the receiver operating curve (ROC) methodology. Unfortunately, this type of analysis is far too complex and is time consuming. In addition, it is required a database with too many images, which makes its application difficult in the clinical practice of routine image quality control (NCRP, 2005).

#### **4.1 MTF**

The modulation transfer function (MTF) is a quantitative and objective measurement of the image quality that can be provided by a system. MTF gives information about the magnitude of the object contrast which is transferred to the image as a function of the spatial frequency. The low spatial frequencies correspond to rough details, whereas the high ones define the fine details or the edges of the structures. For example, an MTF with a value of 0.5 for a determined frequency means that the inherent modulation (contrast) of the object will diminish at 50% given the limitations of the image system (ICRU, 2009) (Fig. 7).

In practice, the MTF is determined through the Fourier transform of the line spread function (LSF), which is obtained through the differentiation of the edge response (ESF).

Spatial resolution is expressed in terms of spatial frequency, which in turn is associated with a MTF value.

Fig. 7. Modulation transfer function of mammography systems (ICRU, 2009).

#### **4.2 Noise**

122 Imaging of the Breast – Technical Aspects and Clinical Implication

anode/filter combinations, such as Mo/Mo or Mo/Rh, operating in the 25-35 kVp range. The spectra of several anode/filter combinations (Fig. 6) show the important proportion of of X-rays characteristics (17.5 and 19.7 keV) from the molybdenum target and the strong suppression of the spectrum at energies >20 keV because of the k-shell absorption edge of the molybdenum filter (Fig. 6a) or else higher than 23 keV because of the Rh filter (Fig. 6b). The characteristics of the new digital detectors make it possible to use other anode/filter combinations such as Rh/Rh (Fig. 6c), W/Rh (Fig. 6d), W/Ag and W/Al which have some advantages for imaging dense or thick breasts. In addition, the breast doses associated with

The adequate selection of the spectrum (beam) may reduce the dose values above 20% (Young, 2006; Dance, 2000; Riabi, 2010). The threshold value for breast thickness where the spectrum is changed depends on the AEC calibration which is performed by technical services who install the equipment (which should be done together with those medical staff who use the equipment). The correct selection of the X-ray beam will strongly influence the

Image quality is a fundamental concept for the control and optimisation of mammography; it aims to improve the early detection of cancer and other pathological lesions in the breasts. The image quality can be quantified by measuring the contrast noise ratio, the signal noise ratio, the modulation transfer function (MTF) (spatial resolution), the noise, the uniformity and various artefacts such as the ghost image. In order to assess these physical parameters, phantoms and more specific devices are needed to perform quality control in mammography. The current trend is to utilise a contrast-detail phantom (CDMAM) which permits the assessment of the image quality as a function of the contrast threshold associated with the circular objects of different diameters and thickness (SEFM, 2008). The "clinical "assessment of the image quality is better performed by means of the receiver operating curve (ROC) methodology. Unfortunately, this type of analysis is far too complex and is time consuming. In addition, it is required a database with too many images, which makes its application difficult

The modulation transfer function (MTF) is a quantitative and objective measurement of the image quality that can be provided by a system. MTF gives information about the magnitude of the object contrast which is transferred to the image as a function of the spatial frequency. The low spatial frequencies correspond to rough details, whereas the high ones define the fine details or the edges of the structures. For example, an MTF with a value of 0.5 for a determined frequency means that the inherent modulation (contrast) of the object will

In practice, the MTF is determined through the Fourier transform of the line spread function

Spatial resolution is expressed in terms of spatial frequency, which in turn is associated with

diminish at 50% given the limitations of the image system (ICRU, 2009) (Fig. 7).

(LSF), which is obtained through the differentiation of the edge response (ESF).

these combinations are lower than those delivered with Mo/Mo or Mo/Rh.

in the clinical practice of routine image quality control (NCRP, 2005).

dose and image quality.

**4.1 MTF** 

a MTF value.

**4. Image quality in mammography** 

The main component of noise in radiographic images is the quantum noise, which is associated with the statistical fluctuations in the photons' fluence on the detector and with the random variations in the absorption. The simplest way to define the noise is through a measurement of the standard deviation (sd) of the number of photons absorbed (N) in a region of the detector. This figure complies with Poisson statistics, and therefore sd = N0.5 (i.e., the sd is related to the square root of the Kerma) (Chevalier, 2010).

The structural noise in the digital detectors emerges mainly from the lack of homogeneity in the sensitivity of the elementary detectors (i.e., from the fixed spatial variation of the image detecting structure), which means that it is also proportional to the dose. Moreover, this noise causes the appearance of a structured background in the image, which is usually removed through the flat field techniques. These techniques include the creation of a corrective mask from a direct and uniform image of the X-ray beams (Chevalier, 2010).

In digital systems, what has to be added to these two types of noise is the electronic one, which emerges from the electronic readout outside the pixels and in the amplification of the signal and which does not depend on the dose. The electronic noise owes essentially to the dark noise in the detectors and decreases if the temperature of the surroundings is lowered or reduced. Therefore, the digital equipment has to operate in temperatures at intervals between 20-30 ºC.

#### **4.3 Uniformity**

The initial operation which usually occurs is a "flat-field", a correction of the uniformity of gain. The non-uniformity of the sensitivity of the detector is corrected through a gain map and is also used to correct all the images acquired. Moreover, if an element of a single (pixel) detector is defective its signal can be replaced with a reasonable combination of adjacent detector signals. This is acceptable if the defective detectors are isolated and only few of them are faulty. Detectors of the CR type presents a lack of uniformity due to the heal effect that is very depending on the X-ray unit.

Image Quality Requirements for Digital Mammography in Breast Cancer Screening 125

ray spectral data, photon interaction cross-sections and Monte Carlo codes. Other important factors affecting the calculations are the mammography system's characteristics and the imaging system components and geometry. The differences in cG values quoted by different

The breast model most commonly adopted (Fig. 8) has a central region consisting of a homogeneous mixture of adipose and glandular tissue surrounded by a layer at all sides, except for the one corresponding to the chest wall representing the skin. It is assumed that the breast is firmly compressed by a polycarbonate compression paddle. The percentage of breast glandularity is defined as the fraction by weight of glandular tissue at the central region (without skin). Most authors employ the elemental tissue composition published by

Initially, it was assumed that a 50:50 mixture of adipose and glandular tissues was representative of a typical breast (Hammerstein, 1979). On this basis, phantoms of several thicknesses were constructed assuming this "standard" composition with the aim of facilitating DG estimates in the practice. This assumption implied that the fraction of glandular tissue was independent of compressed breast thickness. On the basis of this data, the standard "phantom" was defined as a 4.5 cm thickness of PMMA, representing the "standard breast" (4 cm thick and 50%/50% glandular/adipose tissue) (IPEM, 1989; CEC, 1996). Data indicating that the composition of the average compressed breast deviates from the 50:50 composition has been published (Geise, 1996; Klein et al., 1997; Young et al., 1998; Chevalier, 1998; Beckett, 2000; Zoetelief et al., 2006). In addition, it was found that breast glandularity decreases when the compressed breast thickness increases. In some of these works it is also determined that the equivalent thickness of PMMA gives the same incident air kerma at its upper surface as for that of a breast of a specified thickness and composition

Fig. 8. Breast model geometry. The rectangular section represents a vertical cross-section through the breast coplanar with the focal spot of the X-ray tube. The D-shape section represents the breast in craniocaudal projection. The shaded and outer regions represent, respectively, the breast parenchyma and the skin (0.5 cm of adipose tissue). In the work of

DG is generally calculated through the following relationship (ICRU, 2005):

authors were as high as 15-16% (Klein, 1997; ICRU, 2005; Dance 2000).

(Geise, 1996; Dance et al., 2000, 2009; Kruger, 2001; Argo, 2004).

Dance (2009) it is used as a voxelised breast model.

**5.1 Practical issues** 

Hammerstein et al. (1979).

## **4.4 Artefacts**

Artefacts are undesirable characteristics which are not related to the mammary anatomic structures of a radiographic image. They can hinder the image by hiding or simulate a lesion on detection.

Artefacts can be caused by the source of X-rays, the beam filter, the compression device, breast support, grid, and flaws in processing, amongst others. In digital mammography, besides the sources just cited, the non-uniformity in the response of the elemental detectors may also generate artefacts, owing to the results of an inadequate flat-fielding. Another drawback in the digital system is the presence of reminiscent images (ghost images), resulting from previous exposures (ICRU, 2009). The latest appears more often with CR systems or aSe based flat panel detectors.
