**4. Digital assessment of liver steatosis**

ered as an auxiliary method or technique, yet in the age of high-tech medicine, processes ongoing on the level of organelles are the ones which by characteristic ultrastructural changes frequently refer to or indicate a particular pathology. The following must be strictly observed in

ly squeezing them and immediate immersion in fixing solution; 2) chemical composition of fix‐ ing solution, temperature, sample fixation and rinsing time; 3) embedding of liver tissue samples in mixture of epoxy resins in accordance with polymerization time of these resins; 4) quality of sample cutting with ultramicrotome and contrasting with uranyl acetate and lead cit‐ rate; 5) all cells and their organelles visible in the ultra-thin slices under the electron microscope have to be examined. It should be noted that resolution of transmission electron microscope (TEM) is within the range of 0.2 to 2 nm and resolution of scanning electron microscope is 4 nm.

To reduce the potential inaccuracies in the processing and evaluation of biopsy specimens it is important to look for modern solutions in order to maximize the efficiency of use of biop‐ sy specimens. One of the solutions could be application of virtual microscopy having exten‐ sive mathematical basis with fractal and entropy considerations as well as technological support by appropriate software and hardware. Implementation of innovations into practice

The digital image analysis [12-14] and computed morphometry in general is considered an important tool in pathology. It can decrease the workload of voluminous repeated measure‐ ments and increase the accuracy and objectiveness of the results. In several fields, e.g., im‐ munohistochemical and molecular typing of breast cancer, the application of digital image

In virtual microscopy, the demands for mathematical basis are higher than in routine histol‐ ogy. This is illustrated by examples of entropy considerations, Delaunay's triangulation or fractal geometry and general non-Euclidean geometry for irregularly shaped biological ob‐ jects [14-16]. Sophisticated software must be elaborated as well. Additional technical re‐ quirements exist for image resolution and size, fast wide-band data transfer as well as digital data storing [12, 13]. The slide scanners and visualisation software are available and

Computed morphometry becomes more practical in association with virtual microscopy and digital image analysis as well. As postulated in reference [16] the natural development of science occurs from the ability to recognize, name and classify the object (corresponding to the diagnosis, e.g., chronic viral hepatitis C) to semiquantitative, ordering measurements (e.g., the activity assessment by Knodell or any analogous scale), finally reaching quantita‐ tive characteristics. Descriptive diagnoses and semiquantitative estimates are widely used in the "classic" liver pathology. In order to gain sufficient reliability and fastness, scalar meas‐ urements would require digital assessment [16]. Computed morphometry on the basis of

pieces without mechanical‐

electron microscopy: 1) liver tissue sampling and slicing into 1 mm3

260 Liver Biopsy – Indications, Procedures, Results

**3. Virtual microscopy: The general principles**

virtual microscopy is a way towards scalar measurements.

analysis is already highly practical [13].

improve continuously [12].

could significantly increase the effectiveness of liver biopsy specimens.

Among Western population, liver steatosis is a frequent finding [28-29] as it is associated with such common factors as chronic viral hepatitis [19], alcohol drinking, diabetes mellitus or obesity [17]. It has been considered a risk factor for liver fibrosis [18, 19]. Steatosis, includ‐ ing non-alcoholic steatohepatitis [19] has become an important target in diagnostics and sci‐ entific research therefore highly reproducible measurements are necessary to evaluate the course of disease, outcome and effect of treatment. The biopsy is still considered a gold standard in the diagnosis and assessment of steatosis as the imaging including ultrasonogra‐ phy, computed tomography and magnetic resonance imaging can be affected by lower sen‐ sitivity [17, 30]. The severity of steatosis in liver biopsies can be graded by several semiquantitative systems (Table 2) assessing the eyeballed proportion of affected cells [30-35].

The present semiquantitative estimates are subjective and limit the possibilities of statistic analysis [18]. Numerical value, expressing the exact percentage of affected cells would be more reliable if an adequate biopsy is analysed. Such measurement is possible, especially in computer-assisted way, but it would require architecturally arranged count of nuclei and fat vacuoles per biopsy. Thus, the measurement would be time-consuming and accordingly ex‐ pensive. On the other hand, steatosis is relatively easy target for digital quantification of the general fat amount due to the regular shape and distinct colour of fat vacuoles [18, 19]. The digital quantification of steatosis shows high reproducibility exceeding the quality of man‐ ual estimate [19]. Commercial software for image analysis has been recently employed and novel automated procedures are under development [18]. The estimate is more reliable if both morphological and chromatic operators are used in order to characterise lipid particles [18]. The fat vacuole is optically and geometrically simple object – optically empty after rou‐ tine processing and deparaffinisation, thus white and rounded. If colour only is used for identification, however, the sinusoids, empty portal vessels and bile ducts [30] as well as glycogen nuclei in hepatocytes might be undertaken as false positives (Figure 1). The round‐ ed shape of fat vacuole helps to exclude longitudinal or tangential sections of sinusoids, blood vessels and portal bile ducts. In haematoxylin-eosin stained sections, the colour con‐ trast can be used to identify glycogen nuclei as in this case the optically empty space is sur‐ rounded by basophilic nuclear membrane in contrast to fat vacuole located in eosinophilic cytoplasm. Thus, the conclusion at present is to include both chromatic, size and shape as‐ sessment [18, 30]. Manual check can improve the accuracy in case of perpendicular sections of small vessels and fat cysts [30]. However, such control would increase the workload. The benefits of objectiveness and numerical value of continuous variable still remain. More stud‐ ies would be necessary to determine how accurate the control must be for practical means; theoretically the significant vascular changes in cirrhosis point towards the idea that accu‐ rate identification of fat vacuoles is a must to avoid non-random errors.

**Figure 1.** Liver steatosis. Note the macrovesicular steatosis (stars) characterised by size of fat vacuole exceeding the diameter of hepatocyte nucleus, and the microvesicular steatosis (small arrows) caused by fat vacuoles smaller than hepatocyte nucleus. The optically empty fat vacuoles must be promptly distinguished from glycogen nuclei (large ar‐

Future Aspects of Liver Biopsy: From Reality to Mathematical Basis of Virtual Microscopy

http://dx.doi.org/10.5772/52753

263

The fat stains as Sudan IV are well-known [4]. However, several researchers have reported tech‐ nical problems. The artifacts can include deformation of lipid vacuoles as well as sinusoidal and background staining [17, 36-38]. The non-lipid positivity would limit the possibilities of colour analysis, and the deformation – of shape analysis. The practicality of osmium tetroxide stain is negatively affected by the necessity to use frozen tissue and by the toxicity of reagents [4].

Several research groups have reported that manual assessment of steatosis leads to signifi‐ cantly higher estimates than computer-obtained data [17, 19] regardless if area measurement or stereological point counting is used [30]. The coefficient can be as high as 3.78 [19]. Practi‐ cising physicians should remember that association between degree of steatosis and risk of cirrhosis is proved using manual assessments and thus the scales are adjusted for manual use. Consequently, interpretation of digital data cannot involve the use of unadjusted previ‐

It should be noted that the principal meaning of diagnosing steatosis is not affected by the evaluation method. Increasing steatosis percent is associated with advancing fibrosis stage both manually and digitally [19]. The data obtained by pathologist and automated software show close correlation [17]. After liver transplantation, aspartate aminotransferase, alanine aminotransferase and prothrombin time have shown better correlation with automated measurements in 4 of 5 posttransplant time points but the total bilirubin level correlated bet‐ ter with manual assessment in 3 of 5 time points. The graft survival showed a significant as‐ sociation with macrovesicular steatosis both in automated and manual measurements

row) and sinusoids (arrowhead). Haematoxylin-eosin stain, original magnification 400x

although the p value was less for automated measurement [17].

ous scales as risk classes.


**Table 2.** The different grading systems of liver steatosis

general fat amount due to the regular shape and distinct colour of fat vacuoles [18, 19]. The digital quantification of steatosis shows high reproducibility exceeding the quality of man‐ ual estimate [19]. Commercial software for image analysis has been recently employed and novel automated procedures are under development [18]. The estimate is more reliable if both morphological and chromatic operators are used in order to characterise lipid particles [18]. The fat vacuole is optically and geometrically simple object – optically empty after rou‐ tine processing and deparaffinisation, thus white and rounded. If colour only is used for identification, however, the sinusoids, empty portal vessels and bile ducts [30] as well as glycogen nuclei in hepatocytes might be undertaken as false positives (Figure 1). The round‐ ed shape of fat vacuole helps to exclude longitudinal or tangential sections of sinusoids, blood vessels and portal bile ducts. In haematoxylin-eosin stained sections, the colour con‐ trast can be used to identify glycogen nuclei as in this case the optically empty space is sur‐ rounded by basophilic nuclear membrane in contrast to fat vacuole located in eosinophilic cytoplasm. Thus, the conclusion at present is to include both chromatic, size and shape as‐ sessment [18, 30]. Manual check can improve the accuracy in case of perpendicular sections of small vessels and fat cysts [30]. However, such control would increase the workload. The benefits of objectiveness and numerical value of continuous variable still remain. More stud‐ ies would be necessary to determine how accurate the control must be for practical means; theoretically the significant vascular changes in cirrhosis point towards the idea that accu‐

rate identification of fat vacuoles is a must to avoid non-random errors.

Estimating the percentage of affected hepatocytes in 5% bands [30]

Mild: less than 30% hepatocytes involved Moderate: 30-60% hepatocytes involved Severe: more than 60% hepatocytes involved

262 Liver Biopsy – Indications, Procedures, Results

Grade 4: at least 75% hepatocytes involved

Grade 4: at least 67% hepatocytes involved

Grade 3: more than 66% hepatocytes involved

Grade 0: less than 1%

Grade 1: less than 33%

Grade 0 (no or minimal steatosis): less than 5% hepatocytes involved Grade 1: at least 5% but less than 25% hepatocytes involved Grade 2: at least 25% but less than 50% hepatocytes involved Grade 3: at least 50% but less than 75% hepatocytes involved

Grade 1: at least 1% but less than 6% hepatocytes involved Grade 2: at least 6% but less than 34% hepatocytes involved Grade 3: at least 34% but less than 67% hepatocytes involved

Grade 2: at least 33% but less than 66% hepatocytes involved

**Table 2.** The different grading systems of liver steatosis

**Grading Reference**

[31]

[32]

[34] [35]

[33]

**Figure 1.** Liver steatosis. Note the macrovesicular steatosis (stars) characterised by size of fat vacuole exceeding the diameter of hepatocyte nucleus, and the microvesicular steatosis (small arrows) caused by fat vacuoles smaller than hepatocyte nucleus. The optically empty fat vacuoles must be promptly distinguished from glycogen nuclei (large ar‐ row) and sinusoids (arrowhead). Haematoxylin-eosin stain, original magnification 400x

The fat stains as Sudan IV are well-known [4]. However, several researchers have reported tech‐ nical problems. The artifacts can include deformation of lipid vacuoles as well as sinusoidal and background staining [17, 36-38]. The non-lipid positivity would limit the possibilities of colour analysis, and the deformation – of shape analysis. The practicality of osmium tetroxide stain is negatively affected by the necessity to use frozen tissue and by the toxicity of reagents [4].

Several research groups have reported that manual assessment of steatosis leads to signifi‐ cantly higher estimates than computer-obtained data [17, 19] regardless if area measurement or stereological point counting is used [30]. The coefficient can be as high as 3.78 [19]. Practi‐ cising physicians should remember that association between degree of steatosis and risk of cirrhosis is proved using manual assessments and thus the scales are adjusted for manual use. Consequently, interpretation of digital data cannot involve the use of unadjusted previ‐ ous scales as risk classes.

It should be noted that the principal meaning of diagnosing steatosis is not affected by the evaluation method. Increasing steatosis percent is associated with advancing fibrosis stage both manually and digitally [19]. The data obtained by pathologist and automated software show close correlation [17]. After liver transplantation, aspartate aminotransferase, alanine aminotransferase and prothrombin time have shown better correlation with automated measurements in 4 of 5 posttransplant time points but the total bilirubin level correlated bet‐ ter with manual assessment in 3 of 5 time points. The graft survival showed a significant as‐ sociation with macrovesicular steatosis both in automated and manual measurements although the p value was less for automated measurement [17].

When analysing liver steatosis, the observations of higher accuracy in resin-embedded sam‐ ples [18] request more technological progress in order to create methodology for easy use in routine samples.

hepatic functional reserve was demonstrated [24]. The problem was insufficient accuracy of computer-assisted morphometry [21] manifesting as inter-observer differences. Poor correla‐ tion of the fibrosis area with Ishak staging score has been observed as well [21]. Other scientists have also found that analysis of early fibrosis necessities qualitative assessment despite the general correlation between amount of connective tissue and Ishak grade of fibrosis [20]. Tis‐ sue geometry differences in subsequent sections also can be more accurately classified by hu‐

Future Aspects of Liver Biopsy: From Reality to Mathematical Basis of Virtual Microscopy

http://dx.doi.org/10.5772/52753

265

Digital image analysis for the evaluation of fibrosis in chronic viral hepatitis C has been studied also as mentioned in references [41-42]. Automatic quantification of liver fibrosis in‐ cluding the validation of the method has been performed as described in reference [43]. Oth‐ er investigators have employed computerised image analysis for the evaluation of fibrosis as well [44-47]. In most investigations, correlation between digital and manual semiquantita‐ tive score has been shown [20, 44-47]. However, the digital data do not allow to differentiate

**6. Digital biopsy analysis for inflammatory liver lesion: Future begins**

The incorporation of Mandelbrot's fractal geometry [48] into the digital evaluation of liver

The short description of fractal is provided in Table 3; detailed characteristics can be found

Beds of rivers, irregularity of coastline, profiles of mountain chain, clouds

The fractal is a mathematical object characterised by self-similar patterns. At every scale, fractal shows (infinitely) either the same structure or is at least similar to other scales. The complexity is retained independently of magnification. Thus, although fractal curve is one dimensional similarly to regular line, the fractal dimension is greater than topological dimension. Due to the infinite similarity, fractals cannot be measured in traditional ways. Although fractals have got significant popularity due to their beauty, the importance of fractal theory is in the mathematical basis and the ability to describe, among other

Branching of blood vessels or bronchi, the invasive edge of tumour, neurons. See also Figure 2-6

Biological fractal-like objects have limited range of self-similarity upon magnification thus behaving as random fractals, in contrast to mathematical/geomaterical constructs with unlimited

man eye [22]. Full section digital analysis seems to be important [20].

biopsy for chronic hepatitis has brought revolutionary changes [40].

processes, the biological phenomena.

level of complexity (self-exact fractals)

between low stages of fibrosis [20, 45, 47].

**today**

in recent reviews [49].

Definition and essential features of fractal

Fractals in nature: selected

Fractals in biology: selected

Peculiarities of fractals in

**Table 3.** The characteristics of fractals

examples

examples

biology

Digital stereological point counting has been employed in liver steatosis evaluation as well [33]. The researchers have observed the same fact that manual semiquantitative assessment tends to be significantly higher. The lack of precision in manual evaluation can be related to the physiology of vision and processing of the visual information [19, 39].

Some researchers have also come to the conclusion that automated assessment of liver stea‐ tosis is more time-consuming than manual [30]. The time input for digital measurement is found to be threefold greater than for manual evaluation [19]. Although this opinion is based on trustable experience, half of the problem is solved already as the whole slide imag‐ ing eliminates the need to choose appropriate number of representative fields submitted for analysis and the necessity for human participation in the obtaining and archiving of digital images. Besides the whole slide imaging, the degree of automatisation must be further in‐ creased: optimal software abolishes the manual correction of object inclusion into measure‐ ments. However, this deserves morphologically correct mathematical model. Other groups have considered computer-aided morphometry to be fast and objective [16].
