Abstract <p>In the context of chronic liver diseases, where variability in progression necessitates early and precise diagnosis, this study addresses the limitations of traditional histological analysis and the shortcomings of existing deep learning approaches. A novel patch-level classification model employing multi-scale feature extraction and fusion was developed to enhance the grading accuracy and interpretability of liver biopsies, analyzing 1322 cases across various staining methods. The study also introduces a slide-level aggregation framework, comparing different diagnostic models, to efficiently integrate local histological information. Results from extensive validation show that the slide-level model consistently achieved high F1 scores, notably 0.9 for inflammatory activity and steatosis, and demonstrated rapid diagnostic capabilities with less than one minute per slide on average. The patch-level model also performed well, with an F1 score of 0.64 for ballooning and 0.99 for other indicators, and proved transferable to public datasets. The conclusion drawn is that the proposed analytical framework offers a reliable basis for the diagnosis and treatment of chronic liver diseases, with the added benefit of robust interpretability, suggesting its practical utility in clinical settings.</p> Graphical abstract <p>The study presents an analytical framework for efficient slide-level grading of liver biopsy, aiming to enhance the accuracy and interpretability of chronic liver disease diagnosis. The process begins with the extraction of a patient’s liver biopsy sample, which is then stained for analysis. The training phase of the framework, as depicted in the model diagram, involves using pathological images at 10<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11517_2024_3266_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\times $</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> and 20<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11517_2024_3266_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\times $</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> magnifications from the stained biopsy sample to obtain corresponding patches. These patches are subsequently fed into two identical pretrained networks for feature extraction. The model parameters are updated by calculating the classification loss of the network for the 20<InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11517_2024_3266_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\times $</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> magnification patches and the similarity loss after feature fusion of the corresponding patches, constituting the core of the patch-level work. The inference phase of the framework utilizes pathological images at 20<InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11517_2024_3266_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\times $</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> magnification to derive foreground patches and their corresponding position information. These patches are processed through the feature extractor trained in the earlier phase to derive their features. Ultimately, these features are fused with position information, and the aggregation method is applied to obtain the diagnostic results, forming the core of the slide-level work. The framework’s effectiveness is validated by high F1 scores, rapid diagnostic capabilities, and transferability to public datasets, suggesting its potential utility in clinical settings for diagnosing and treating chronic liver diseases.</p>

Toward efficient slide-level grading of liver biopsy via explainable deep learning framework

期刊论文
发表日期：

2025年1月13日

查看全文

Bingchen Li,
Qiming He,
Jing Chang,
Bo Yang,
Xi Tang,
Yonghong He,
Tian Guan,
Guangde Zhou

摘要

Abstract

In the context of chronic liver diseases, where variability in progression necessitates early and precise diagnosis, this study addresses the limitations of traditional histological analysis and the shortcomings of existing deep learning approaches. A novel patch-level classification model employing multi-scale feature extraction and fusion was developed to enhance the grading accuracy and interpretability of liver biopsies, analyzing 1322 cases across various staining methods. The study also introduces a slide-level aggregation framework, comparing different diagnostic models, to efficiently integrate local histological information. Results from extensive validation show that the slide-level model consistently achieved high F1 scores, notably 0.9 for inflammatory activity and steatosis, and demonstrated rapid diagnostic capabilities with less than one minute per slide on average. The patch-level model also performed well, with an F1 score of 0.64 for ballooning and 0.99 for other indicators, and proved transferable to public datasets. The conclusion drawn is that the proposed analytical framework offers a reliable basis for the diagnosis and treatment of chronic liver diseases, with the added benefit of robust interpretability, suggesting its practical utility in clinical settings.

Graphical abstract

The study presents an analytical framework for efficient slide-level grading of liver biopsy, aiming to enhance the accuracy and interpretability of chronic liver disease diagnosis. The process begins with the extraction of a patient’s liver biopsy sample, which is then stained for analysis. The training phase of the framework, as depicted in the model diagram, involves using pathological images at 10 $\times $ $\times$ and 20 $\times $ $\times$ magnifications from the stained biopsy sample to obtain corresponding patches. These patches are subsequently fed into two identical pretrained networks for feature extraction. The model parameters are updated by calculating the classification loss of the network for the 20 $\times $ $\times$ magnification patches and the similarity loss after feature fusion of the corresponding patches, constituting the core of the patch-level work. The inference phase of the framework utilizes pathological images at 20 $\times $ $\times$ magnification to derive foreground patches and their corresponding position information. These patches are processed through the feature extractor trained in the earlier phase to derive their features. Ultimately, these features are fused with position information, and the aggregation method is applied to obtain the diagnostic results, forming the core of the slide-level work. The framework’s effectiveness is validated by high F1 scores, rapid diagnostic capabilities, and transferability to public datasets, suggesting its potential utility in clinical settings for diagnosing and treating chronic liver diseases.