Highlight
- Development of an interpretable AI-based framework to characterise collagen deposition phenotypes (CDPs) from picrosirius red-stained liver biopsies in MASLD.
- CDPs capture distinct collagen morphologies, demonstrating superior sensitivity and biological specificity compared to traditional fibrosis staging and collagen proportionate area (CPA) metrics.
- Transcriptomic and proteomic pathway analyses link specific CDPs to active extracellular matrix remodelling and liver functional status.
- Selected CDPs show prognostic relevance for clinical outcomes in MASLD, with external validation supporting model generalizability.
- All computational tools and models are openly accessible, facilitating transparent, reproducible multi-omics pathology research.
Study Background
Liver fibrosis is a pivotal pathological hallmark in metabolic dysfunction-associated steatotic liver disease (MASLD), formerly known as non-alcoholic fatty liver disease. Fibrosis progression portends increased risk for cirrhosis, liver failure, and hepatocellular carcinoma, representing a significant clinical challenge globally due to the rising burden of obesity and metabolic syndrome. Accurate fibrosis assessment is essential for prognosis, treatment tailoring, and understanding disease mechanisms. However, conventional histological fibrosis staging systems provide only ordinal categorical assessments that lack granularity. Quantitative methods such as collagen proportionate area (CPA) improve measurement precision but neglect spatial architecture critical for biological context. Advanced methods leveraging artificial intelligence (AI) have shown potential for enhanced fibrosis characterization but remain largely proprietary and inaccessible to academic research, limiting transparency and reproducibility. This study addresses this gap by proposing an interpretable, AI-driven framework that identifies detailed collagen deposition phenotypes (CDPs) from routine picrosirius red (PSR)-stained liver biopsy slides in MASLD patients, with integrated molecular and clinical correlations.
Study Design
This investigation utilized a discovery cohort of MASLD patients with available PSR-stained liver biopsy slides, matched with transcriptomic and proteomic datasets. An external validation cohort was also incorporated to test model robustness. The study’s approach involved:
- Development of an AI-based image analysis pipeline employing machine learning algorithms to segment and quantify collagen deposition patterns from digitized PSR stains.
- Definition of distinct collagen deposition phenotypes (CDPs) based on unsupervised clustering of morphologic collagen features extracted from the images.
- Comparative analysis of CDPs with established histological fibrosis staging and CPA quantification metrics.
- Integration of matched liver tissue transcriptomics and proteomics for pathway enrichment analysis linked to each CDP, elucidating underlying biological processes.
- Evaluation of clinical correlates and prognostic performance of CDPs in predicting disease progression or adverse outcomes within the cohorts.
All computational models and tools were shared in an open-access manner to ensure reproducibility and foster further research.
Key Findings
The AI framework identified multiple discrete collagen deposition phenotypes (CDP 1 to 5), each characterized by distinct architectural and morphometric collagen patterns beyond what is captured by traditional fibrosis scores or CPA.
Enhanced Sensitivity and Biological Specificity
Compared to CPA and ordinal fibrosis scores, CDPs provided significantly increased sensitivity in detecting subtle differences in collagen morphology. This improved resolution uncovered distinct molecular signatures in transcriptomic and proteomic profiles co-localizing with specific CDPs.
Pathway and Functional Associations
CDPs 4 and 5 strongly correlated with pathways indicative of active extracellular matrix remodeling, including upregulation of matrix metalloproteinases, collagen crosslinking enzymes, and fibrogenic signaling cascades such as TGF-β and integrin pathways. Conversely, other CDPs aligned with pathways related to immune regulation or hepatocyte function, underscoring the heterogeneity of fibrosis biology.
Clinical and Prognostic Correlations
Certain CDPs demonstrated significant associations with liver functional status markers and clinical outcomes within the discovery cohort, including progression to advanced fibrosis and liver-related events. Prognostic discrimination, although somewhat attenuated, remained evident in the external validation cohort, supporting model generalizability.
Transparency and Reproducibility
All models, algorithms, and analytic tools were released openly, enabling other research groups to independently verify, refine, and extend the findings. This commitment addresses significant challenges faced in the field due to proprietary AI approaches limiting cross-study comparisons.
Expert Commentary
The current study exemplifies the power of AI-driven quantitative pathology to refine our understanding of liver fibrosis in MASLD. By moving beyond coarse categorical staging and simple collagen quantitation, the identification of discrete collagen deposition phenotypes provides a biologically and clinically meaningful framework. The close integration with multi-omics datasets unveils mechanistic pathways underlying fibrosis heterogeneity, which might inform targeted therapeutic strategies. One important limitation is the attenuation of prognostic accuracy in the external cohort, highlighting the need for further validation in larger, more diverse populations. Additionally, clinical translation will require integrating this framework with non-invasive fibrosis assessment tools and exploring longitudinal phenotypic evolution.
This open-access approach sets an important precedent for transparency in computational pathology, mitigating a critical barrier impeding progress in liver fibrosis research. Future studies should expand incorporation of spatial transcriptomics and single-cell analyses to further disentangle cellular contributors to the distinct collagen phenotypes.
Conclusion
This pioneering AI-based framework significantly advances fibrosis assessment in MASLD by capturing nuanced collagen deposition patterns that traditional methods overlook. The resulting collagen deposition phenotypes provide richer biological insights into extracellular matrix remodeling and functional liver status. Their clinical relevance is underscored by prognostic associations validated across cohorts. Open dissemination of the models promotes reproducible multi-omics pathology research and paves the way toward precision liver fibrosis phenotyping.
This approach has the potential to refine patient risk stratification, accelerate biomarker discovery, and ultimately inform the development of antifibrotic therapies tailored to distinct fibrotic microenvironments in MASLD.
Funding and ClinicalTrials.gov
The study details do not indicate specific funding sources or clinical trial registration numbers. Acknowledgement of supporting institutions and funding agencies would be typically found in the original publication.
References
1. Wojciechowska M, et al. Decoding fibrosis: Transcriptomic and clinical insights via AI-derived collagen deposition phenotypes in MASLD. Hepatology (Baltimore, Md.). 2026 Jun 24. PMID: 42341328.
2. Friedman SL. Liver fibrosis — from bench to bedside. J Hepatol. 2003;38 Suppl 1:S38-53.
3. Bedossa P, Carrat F. Liver fibrosis: clinical epidemiology, risk factors and natural history. Dig Dis. 2009;27(5):518-524.
4. Salgado-Somoza A, et al. Advanced AI techniques for histological assessment of liver fibrosis. Liver Int. 2023;43(4):745-756.
5. Kisseleva T, Brenner DA. Mechanisms of fibrogenesis. Exp Biol Med. 2008;233:109-122.