Open Access
Add to BibSonomyAbstract
In this paper we analyze a fibrosis scoring method based on measurement of the fibrillar collagen area from second harmonic generation (SHG) microscopy images of unstained histological slices from human liver biopsies. The study is conducted on a cohort of one hundred chronic hepatitis C patients with intermediate to strong Metavir and Ishak stages of liver fibrosis. We highlight a key parameter of our scoring method to discriminate between high and low fibrosis stages. Moreover, according to the intensity histograms of the SHG images and simple mathematical arguments, we show that our area-based method is equivalent to an intensity-based method, despite saturation of the images. Finally we propose an improvement of our scoring method using very simple image processing tools.
© 2015 Optical Society of America
1. Introduction
Liver fibrosis is a symptom of liver dysfunction which is characterized by the accumulation of extracellular matrix (ECM) mainly composed of type I and III collagen fibers (scar tissue). Its origin is often linked to chronic stressors: alcohol, fat accumulation, exposure to certain drugs (hormonal modulators), and chronic hepatitis B or C virus (HBV, HCV) infections. These forms of hepatitis represent more than 400 000 deaths per year in the world in relation to complications of cirrhosis [1], the most advanced stage of liver fibrosis.
Efforts have been undertaken in recent decades to better characterize this slowly progressive disease which is often asymptomatic until the occurrence of liver decompensation at the end-stage. Although noninvasive diagnosis [2] (Fibrotest, Fibroscan) provide first indication of the disease prominence, examination of ECM-stained histological sections from liver biopsies by trained pathologists is still considered as the gold standard method for fibrosis evaluation. In that frame, Metavir [3] and Ishak [4] staging systems using coloration of the ECM with Sirius Red or Masson Trichrome dyes provide semi-quantitative scores which allow assessing the severity of the disease. The optical absorbance of the dye is directly related to the amount of collagen whose structure across the biopsy gives additional information on the progress of the disease, such as the presence of collagen bridges between portal veins. However, this is a time-consuming approach which suffers from intra-and inter-observer discrepancy as well as from variation of the staining over time and from a sample to another. The development of a robust image-based method using unstained histological liver sections to quantify fibrillar collagen and possibly to implement morphometric analysis, is then of paramount interest.
In a first paper [5], we showed that Second Harmonic Generation (SHG) microscopy of type I fibrillar collagen of unstained liver sections could provide a robust method for quantifying fibrosis. Our scoring method is based on measuring the relative area covered by collagen deposits in the SHG image of histological slices (fibrosis-SHG index), the total area of the slices being determined from the corresponding TPEF (Two-Photon Excitation Fluorescence) image. SHG images were binarized according to a well chosen intensity threshold so as to remove as much as possible the basal collagen before calculating the index. A fair correlation between the SHG index and the Metavir score was observed on a cohort of 119 biopsies from patients with chronic liver disease [5].
In a second paper we demonstrated that the SHG index is robust to experimental settings like the objective’s numerical aperture (for low NA values), laser power and sample thickness provided that the microscope was previously calibrated on any non-fibrotic area of a sample series. Indeed, SHG intensity from various non fibrotic areas of a series of histological slices was found almost constant so that it can be used as a reference for calibration [6].
In this paper we go further into the implementation and analysis of our SHG scoring method. First, we compare it to both Metavir and Ishak staging systems, using a new cohort of patients suffering from chronic HCV infection. In a second step, we analyze the effect of the threshold value on the SHG index. We also analyze how our method could be equivalent to an intensity measurement despite the fact that it is based only on an area measurement of collagen deposits whose intensity is over a preset threshold and even from partially saturated images. We finally propose very simple image processing tools to account for architectural features of collagen during fibrosis progression.
2. Materials and methods
2.1 Two-photon microscopy system
Our imaging setup was based on a modified confocal microscope composed of an Olympus BX51WI upright stand and a FluoView 300 scanning head (Olympus France, Rungis). A femtosecond Ti:Sapphire laser (Verdi-V5/Mira900F, Coherent France, les Ulis) tuned at λ = 830nm was used as the light source for all experiments. This delivered an average power of ~100mW at the sample in the form of ~200fs FWHM pulses at a repetition rate of 76MHz. The laser power was adjusted so as to achieve a trade-off between signal-to-noise ratio and photobleaching of the sample autofluorescence. The laser beam was circularly polarized at the sample using a quarter-wave plate set above the objective lens. The TPEF light was epicollected and detected through the descanned pathway of the microscope (confocal pinhole removed) by one of the two internal photomultiplier tubes (PMT, Hamamatsu, R928) of the scanning head. The SHG light was collected in the forward direction by a 0.3NA condenser (Olympus France, IX-ULWCD) and detected through a 415nm bandpass filter (Edmund Optics, NT65-680, 10nm FWHM) by a PMT module (Hamamatsu, H5783-01), which was connected to a transimpedance amplifier (Hamamatsu, C7319) so as to match the SHG detection to the internal PMT of the microscope and to use the full range of Fluoview hardware and software (switchable gains of 105, 106 and 107V/A). The excitation wavelength was blocked by 2-mm thick BG39 Schott filters (Edmund Optics Europe, York, UK) inserted on the TPEF and SHG detection pathways.
SHG and TPEF images were acquired with 12-bit intensity resolution and recorded as TIFF files using the microscope software (Fluoview 4.3c version). The slowest scanning speed was selected, corresponding to a pixel dwell-time of about 10µs or a total acquisition time of 2.68sec per frame of 512 x 512 pixel2.
A low magnification 4X/0.16NA microscope objective (UPLSAPO, Olympus France, Rungis) was used for all experiments. The field of view of the 4X objective (3.5x3.5mm2) was not sufficient to image entirely the slices from needle biopsies (typically 20mm long) in a single shot. Series of connexe images were then automatically acquired with a XY translation stage (Marzhaüser, Marzhaüser Wetzlar GmbH & Co.) under control of Labview software. The lateral and axial extents (1/e radii) of the two-photon point-spread function (PSF2) expected for the 4X/0.16NA objective lens were respectively 2.2µm and 48.4µm [7]. Given the value of the axial extent of the PSF2, TPEF signals result from incoherent optical integration over the whole thickness of the slices (4µm). This is also valid for SHG signals, as coherence effects were not observed in such samples [6]. Note that the expected lateral extent of the PSF2 (4.4µm in diameter) is lower than the pixel width (3500/512 = 6.8µm). However, we verified that this slight undersampling did not affect the SHG indices. Indeed, the same results were obtained for the 1024x1024 image definition but at the price of a longer acquisition time, a greater image size (x4) and sometimes photobleaching (the pixel dwell-time remains constant whatever the definition of the microscope).
2.2 Liver sample preparation and histological status
The cohort included 147 patients with chronic hepatitis C infection enrolled at a single centre within the Trent Study of Patients with Hepatitis C Virus Infection (Nottingham University Hospitals NHS Trust). Needle biopsies were performed for diagnostic purposes between 1986 and 2008. Then, the piece of liver was fixed in formaldehyde and embedded in paraffin. Histological sections were then obtained with a standardize thickness (4µm) and a length over 20mm. Histological status was assessed by pathologists on serial sections of the same biopsies stained with Sirius Red, and classified according to the Metavir [3] and Ishak [4] scoring systems. Five scores are associated to the Metavir scoring system from F0 (no fibrosis) to F4 (cirrhosis). Seven scores are associated to the Ishak scoring system from I0 (no fibrosis) to I6 (cirrhosis). Definition and equivalence between the scores of these two major systems can be found in [8] for example. Several samples (42 on 147) were declared inadequate for staging, mostly because of an important pollution of the sample by paraffin. Thus, 105 unstained sections were analyzed by our SHG scoring method and compared to Metavir and Ishak scores.
2.3 SHG scoring
SHG scoring of fibrillar collagen that we implemented in this study was based on the method mentioned in introduction [5]. According to the Metavir staging system, this method was shown to be able to discriminate patients with low (or moderate) and advanced fibrosis (or cirrhosis) stages [5]. For the present study we first performed SHG scoring on the new cohort described above using the same settings as previously used. In particular, the supplied voltage of the PMT and the transimpedance gain (106V/A) were first adjusted, at a fixed laser excitation (~100mW at the sample), on non-fibrotic areas of the slice series, so as to obtain a mean reference intensity of 4% of the full intensity range of the SHG image. This baseline was selected for convenience in order to display both low and high density areas of collagen on the images. However, this choice leads to a partial saturation of the SHG images for the highest density collagen areas, because of the remarkably high dynamic of SHG intensity from fibrous tissues. Fortunately this saturation is not detrimental to the SHG scoring as shown in the present paper. Once the device was calibrated, SHG and TPEF images of the 105 unstained sections were acquired in parallel. SHG images were then binarized using a threshold of 25% of the full intensity range of the image, and the number of pixels over this threshold was calculated. The threshold value was chosen empirically in previous studies according to the intensity histograms of the SHG images, in order to get rid off as much as possible of the basal collagen in non-fibrotic areas. In section 3.3 we will show that the key parameter of our scoring method is the threshold-to-baseline ratio, regardless of the value of each of these quantities and of the saturation level of the SHG images.
Figure 1 shows raw and binary (after thresholding) SHG images for two representative samples of low (F1-I2) and high (F4-I6) fibrosis groups.
Fig. 1 Representative SHG images of low (a,c) and high (b,d) fibrosis groups. (a) and (b) are raw SHG images. The PMT supplied voltage was adjusted so as to obtain a mean reference intensity of 4% of the full intensity range in a non-fibrotic area of (a). The corresponding raw TPEF images are shown in inset and are used to define a mask that delimits the contour of the biopsy and is applied to SHG images. (c) and (d) are binarized SHG images after 25% thresholding.
One important issue for determining the SHG index from needle liver biopsies is measuring the total area of the slices, that is evaluating the overall amount of pixels of the slices. This is achieved by taking advantage of the TPEF produced by autofluorescence of hepatocytes, which are present almost everywhere in the biopsies, thereby allowing to delimit the contour of the slices and thus to generate a mask for the SHG images.
Calculation of the SHG index from raw TPEF and SHG images was implemented inside a macro written for Image J software. No operator intervention is needed for the calculation of the SHG index which strongly reduces the variability of the results obtained by our scoring method. Moreover, the time between images acquisition and obtention of the SHG index is of the order of a few seconds which is well adapted to clinical diagnosis.
3. Results
The SHG index was calculated for the 105 biopsies with the previous method and compared to the Metavir and Ishak scores obtained by trained anatomo-pathologists.
3.1 SHG index versus Metavir and Ishak scores
Results are presented in Fig. 2 as statistical box charts for Metavir in Fig. 2(a) and Ishak in Fig. 2(b).
Fig. 2 SHG index versus Metavir (a) and Ishak (b) scores. The box chart graph is defined as follow: the whiskers display extreme values whereas the boxes represent the median, 25th and 75th percentiles. The mean values are represented by the small box (▫). The numbers between brackets give the number of samples in each histological stage.
The nonlinear correlation between the median value of SHG index and the histological status was evidenced by the high values of Spearman’s coefficients of rank correlation: ρ = 0.90 (p = 0.037) for Metavir and ρ = 0.89 (p = 0.007) for Ishak. However, the major argument for assessing the ability of the SHG scoring method to discriminate between different stages of fibrosis is the importance of statistical differences of SHG index according to Metavir and Ishak scores. The p-values associated to the Mann-Whitney statistical test are given in Table 1. Due to the lack of sample with F0 and I0/I1 scores, the statistics associated to these stages are not represented.
Table 1. Statistical differences of SHG index between (a) Metavir scores and (b) Ishak scores. The p-value is the two-tailed probability associated to the Mann-Whitney test. Statistical significance was considered to be reached for p < 0.05.
According to the Metavir system, our method is significant for the differentiation of F2/F3 (p = 0.001) and even more F3/F4 (p = 0.0002) stages. However, it fails to discriminate between early stages F1/F2 (p = 0.84). According to the Ishak system, our method is significant for the differentiation of I3/I4 (p = 0.0036) and I5/I6 (p = 0.0011) stages. However, it fails to discriminate between early stages I2/I3 (p = 0.83) and more advanced stages I4/I5 (0.47). This is not surprising for the latter because the main difference between I4 and I5 stages is the presence of nodules (morphometric criterion) but not the amount of collagen.
3.2 ROC curves for detection of advanced fibrosis
The present results aim at demonstrating the ability of our SHG scoring method to discriminate between early and advanced stages of fibrosis. For Metavir scale, early stages correspond to F0 to F2 scores. For Ishak scale, early stages correspond to I0 to I3 scores. The Receiver Operating Characteristic (ROC) curve determines whether a parameter (here the SHG index) is able to discriminate two populations (here early and advanced stages of fibrosis). The area under ROC curve (AUROC) is representative of the efficiency of this parameter.
It is remarkable that the values of AUROC obtained in this study are very close for Metavir (Fig. 3(a)) and Ishak scales (Fig. 3(b)). This study confirmed that SHG index is able to discriminate between early (F012) and advanced (F34) fibrosis stages according to Metavir scoring system on this new cohort of patient and demonstrated that SHG index can discriminate between early (I0123) and advanced (I456) fibrosis stages according to Ishak scoring system.
Fig. 3 ROC curves of SHG index for the detection of advanced fibrosis for (a) Metavir (F012 vs F34) and (b) Ishak (I0123 vs I456). The area under the curve is 0.84 (95% confidence interval: 0.77-0.92) for Metavir and 0.82 (95% confidence interval: 0.74-0.91) for Ishak.
3.3 Threshold-to-baseline ratio
To demonstrate that the threshold-to-baseline ratio defined in section 2.3 is the key parameter of our scoring method, we first compared on Fig. 4(a) the SHG index measured using the previous settings (4% baseline and 25% threshold) to the one measured by using a 0.4% baseline (detection gain divided by a factor of 10) and a 2.5% threshold for 35 best quality samples of the cohort. The baseline or reference intensity was reduced by switching the gain of the transimpedance amplifier of the SHG detection path from 106 to 105V/A.
Fig. 4 (a) Comparison of the SHG index of 35 samples obtained for two different detection gains (106 versus 105V/A) of the microscope but identical threshold-to-baseline ratios. The straight line is the best linear fit (slope = 0,98, R2 = 0,995). (b) Comparison of the percentages of the saturated pixels of SHG images of the same 35 samples for the two different detection gains (106 versus 105V/A).
For this latter value SHG images were almost unsaturated and will be then used in section 3.4 for comparison with intensity scoring. It is clear from Fig. 4(a) that the SHG index is almost independent of the detection gain provided that the threshold-to-baseline ratio remains unchanged, and this despite the fact that a great proportion of pixels are saturated in intensity in case of the gain 106 (more than 40% for the highest SHG scores or fibrosis stages) whereas this proportion is much lower for the gain 105, as shown in Fig. 4(b).
To illustrate the effect of the threshold-to-baseline ratio on the ability of our method to discriminate low and high fibrosis stages, we plotted in Fig. 5 the ratio of the SHG index for 3 pairs of samples with extreme fibrosis status (F1-F2 and F4) versus the value of the threshold used for calculating the SHG index. The calculation was performed from the almost unsaturated SHG images (baseline 0.4%, gain 105) in order to be able to scan a large range of threshold values. The curve in Fig. 5 brings up a sharp maximum around a value of 100 for the threshold, or around 2.5% of the full intensity range of the image (4095). This value maximizes the dynamic of the SHG index between high and low fibrosis status. An optimal and robust threshold-to-baseline ratio can thus be established around the value 25/4 that we effectively used in our scoring method.
Fig. 5 Normalized SHG index ratio averaged on 3 pairs of samples with extreme fibrosis status (F4 to F1-F2 ratio) versus the value of the threshold used for calculating the SHG index.
3.4 Area versus intensity SHG scoring methods
To show that our SHG index is well representative of the amount of fibrillar collagen within the biopsies, we compare it in Fig. 6(a) with the total intensity (from pixels over the threshold) per unit area of the almost unsaturated SHG images (gain 105) of the slices previously selected. The two scores appear well correlated as shown by linear fitting of the data (R2 = 0.962). This demonstrates the equivalence of the two methods, but with the paramount difference that only the SHG index is robust to image saturation (see Fig. 4(a)). The reason of such an equivalence is that the mean SHG intensity per selected pixel above the threshold is almost constant whatever the score, as shown in Fig. 6(b).
Fig. 6 (a) Total SHG intensity of the thresholded pixels per unit area of the samples and (b) mean SHG intensity per thresholded pixel as a function of the SHG index.
To account for this quite surprising result, we plotted in Fig. 7 the intensity histograms H(I) of a couple of low and high fibrosis samples. The two histograms are almost decreasing exponential, as evidenced by the linear behavior of the curves plotted in logartihmic scale (see inset of Fig. 6).
Fig. 7 SHG intensity histograms of two biopsies with low (box) and high (triangle) fibrosis status. The inset displays the relevant part of these two histograms in semi-logarithmic scale in order to evidence the almost exponential behavior. The bin width of the histograms was set to 10 in order to reduce fluctuations.
A remarkable feature of the function H(I) = A.exp[-α.(I-I0)] (A, α > 0, I > I0 > 0) is that the mean value <I> of I (here the intensity of a SHG image pixel) over a given interval (here from the threshold intensity It to the maximum intensity Im), which is written:
is independent of the values of A and I0. In other words, translating H(I) or equivalently multiplying it by a constant, does not change <I> over a given interval. As the intensity histograms of the SHG images have almost the same slope α (see Fig. 7 inset) whatever the score, the mean intensity <I> per thresholded pixel of the SHG images is almost independent of the fibrosis stage (see Fig. 6(b)). This explains why our SHG index (or the number of pixels per unit area over a preset threshold) is almost proportional to the total intensity of these pixels (see Fig. 6(a)), which is expected to account for the collagen amount within the biopsy.It now remains to account for the quasi exponential behavior and the evolution of the SHG intensity histograms, in connection with both the SHG process from fibrillar collagen and the fiber growth during liver fibrosis.
3.5 Image processing
In this study the SHG index was used to quantify collagen amount of several biopsies. This index considers SHG pixels as independant and does not take into account any morphological features of collagen expansion. Nevertheless, it is well known that fibrosis is assessed by combining both collagen amount and architectural disorganization [8], as it is the case for semiquantitative scoring systems like Metavir and Ishak. Thus, a simple modification of the algorithm was tested including an additional step before the calculation of the SHG index. During this step a median filter is applied to the binary SHG image. The effect of a median filter on a 2D image is to replace every pixel by the median value of its closest neighbors. There are many advantages of applying such a filter. Indeed, when applied to a binary image, the resulting image after median filtering is still binary and the SHG index can be calculated directly without adding any further binarization step. Moreover, this step can be easily included in the previous macro under Image J without adding significant computation time and without any operator intervention. The main effect of such a filtering is to suppress small isolated collagen deposits (collagen distributed in sinusoidal spaces) while keeping unchanged large collagen structures (aggregated collagen), that are more significative of fibrosis progression [9]. Figure 8 shows binary SHG images (after thresholding and TPEF masking) before and after median filtering, for a patient with F4-I6 histological status.
Fig. 8 SHG binary image before (a) and after (b) median filtering for a patient with F4-I6 histological status. The inset is a zoom on a non fibrotic area and demonstrates that isolated collagen deposits is removed by median filtering.
SHG index was calulated for each biopsy after median filtering of their binary SHG image and the corresponding ROC curves similar to those of Fig. 3 without median filtering, are presented in Fig. 9.
Fig. 9 ROC curves of SHG index for the detection of advanced fibrosis for (a) Metavir (F012 vs F34) and (b) Ishak (I0123 vs I456), after median filtering of the binarized SHG images. The area under the curve is 0.87 (95% confidence interval: 0.79-0.94) for Metavir and 0.85 (95% confidence interval: 0.77-0.93) for Ishak.
An increase of AUROC values was observed for both Metavir (0.87 vs 0.84) and Ishak (0.85 vs 0.82) scores. These results are in agreement with the ones of a recent paper [10], in which the authors implement a fully automated method based on SHG imaging to built the so-called q-Fibrosis index incorporating both collagen quantification and morphological features (information about spatial organization) of the fibrous tissue (portal collagen, septal collagen, sinusoidal collagen). They show that q-Fibrosis index is better than simple collagen proportion area (CPA) for discriminating between the five Metavir stages and claim that their method is less sensitive to sampling error. However, their index is based on a complex algorithm that takes into account 87 parameters of collagen architectural features and should not be suitable for all diseases [8]. All of these results show that nonlinear optical methods can be considered for staging fibrosis by combining robust quantification of collagen amount and morphometric aspects of collagen deposits.
4. Conclusion
To conclude, this study demonstrated that measuring the area of collagen deposits from low magnification SHG images of histological slices from human liver biopsies using nonlinear microscopy provides a reliable and robust method to assess fibrosis in terms of collagen amount, according to the Metavir and Ishak scoring systems. The threshold-to-baseline ratio used for calibration and processing of the SHG images was shown to be the key parameter of our scoring method. The study revealed the existence of an optimal value for this ratio, which optimizes discrimination between advanced and non-advanced fibrosis. Moreover the area-based method was shown to be equivalent to an intensity-based method despite image saturation. This was explained by the quasi-exponential shape of the SHG intensity histograms associated to fibrillar collagen. A next step for this study will be to understand the very peculiar behavior of these histograms which is probably characteristic of second harmonic generation by multi-scale fibrillar structures. Further experiments on various collagenous tissues are currently underway in order to test this assumption.
References and links
1. J. F. Perz, G. L. Armstrong, L. A. Farrington, Y. J. F. Hutin, and B. P. Bell, “The contributions of hepatitis B virus and hepatitis C virus infections to cirrhosis and primary liver cancer worldwide,” J. Hepatol. 45(4), 529–538 (2006). [CrossRef] [PubMed]
2. L. Castéra, J. Vergniol, J. Foucher, B. Le Bail, E. Chanteloup, M. Haaser, M. Darriet, P. Couzigou, and V. De Lédinghen, “Prospective comparison of transient elastography, Fibrotest, APRI, and liver biopsy for the assessment of fibrosis in chronic hepatitis C,” Gastroenterology 128(2), 343–350 (2005). [CrossRef] [PubMed]
3. P. Bedossa, and The METAVIR cooperative group, “Presentation of a grid for computer analysis for compilation of histopathologic lesions in chronic viral hepatitis C. Cooperative study of the METAVIR group,” Ann. Pathol. 13(4), 260–265 (1993). [PubMed]
4. K. Ishak, A. Baptista, L. Bianchi, F. Callea, J. De Groote, F. Gudat, H. Denk, V. Desmet, G. Korb, R. N. MacSween, M. J. Phillips, B. G. Portmann, H. Poulsen, P. J. Scheuer, M. Schmid, and H. Thaler, “Histological grading and staging of chronic hepatitis,” J. Hepatol. 22(6), 696–699 (1995). [CrossRef] [PubMed]
5. L. Gailhouste, Y. Le Grand, C. Odin, D. Guyader, B. Turlin, F. Ezan, Y. Désille, T. Guilbert, A. Bessard, C. Frémin, N. Theret, and G. Baffet, “Fibrillar collagen scoring by second harmonic microscopy: A new tool in the assessment of liver fibrosis,” J. Hepatol. 52(3), 398–406 (2010). [CrossRef] [PubMed]
6. T. Guilbert, C. Odin, Y. Le Grand, L. Gailhouste, B. Turlin, F. Ezan, Y. Désille, G. Baffet, and D. Guyader, “A robust collagen scoring method for human liver fibrosis by second harmonic microscopy,” Opt. Express 18(25), 25794–25807 (2010). [CrossRef] [PubMed]
7. W. R. Zipfel, R. M. Williams, and W. W. Webb, “Nonlinear magic: multiphoton microscopy in the biosciences,” Nat. Biotechnol. 21(11), 1369–1377 (2003). [CrossRef] [PubMed]
8. T. Asselah, P. Marcellin, and P. Bedossa, “Improving performance of liver biopsy in fibrosis assessment,” J. Hepatol. 61(2), 193–195 (2014). [CrossRef] [PubMed]
9. D. C. Tai, N. Tan, S. Xu, C. H. Kang, S. M. Chia, C. L. Cheng, A. Wee, C. L. Wei, A. M. Raja, G. Xiao, S. Chang, J. C. Rajapakse, P. T. So, H. H. Tang, C. S. Chen, and H. Yu, “Fibro-C-Index: comprehensive, morphology-based quantification of liver fibrosis using second harmonic generation and two-photon microscopy,” J. Biomed. Opt. 14(4), 044013 (2009). [CrossRef] [PubMed]
10. S. Xu, Y. Wang, D. C. S. Tai, S. Wang, C. L. Cheng, Q. Peng, J. Yan, Y. Chen, J. Sun, X. Liang, Y. Zhu, J. C. Rajapakse, R. E. Welsch, P. T. So, A. Wee, J. Hou, and H. Yu, “qFibrosis: A fully-quantitative innovative method incorporating histological features to facilitate accurate fibrosis scoring in animal model and chronic hepatitis B patients,” J. Hepatol. 61(2), 260–269 (2014). [CrossRef] [PubMed]
