AI redefines fibrosis patterns in steatotic liver disease: Opportunities and challenges

Introduction

Steatotic liver disease (SLD) has emerged as a major global health burden, with rapidly rising prevalence and substantial contributions to both liver-related and cardiometabolic morbidity and mortality.^[1] It comprises a heterogeneous group of chronic liver disorders with diverse pathophysiological drivers and variable clinical outcomes, posing significant challenges for risk stratification and therapeutic development.^[2] Across its major subtypes, including metabolic dysfunction-associated steatotic liver disease (MASLD), metabolic dysfunction-associated alcoholic liver disease (MetALD), and alcoholic liver disease (ALD), patients are exposed to differing constellations of metabolic stress, alcohol-related toxicity, and inflammatory signaling. These distinct pathogenic influences may translate into variable patterns of fibrotic remodeling and organ-specific complications. However, conventional histopathology, which relies largely on semiquantitative scoring systems and interobserver interpretation, often lacks the resolution to capture such spatial and mechanistic nuances, thereby limiting deeper insight into disease biology and individualized risk assessment.

AI-based digital pathology provides a transformative platform to overcome these limitations by enabling automated, quantitative, and spatially resolved tissue analysis.^[3] This approach allows high-resolution mapping of collagen deposition, zonal fibrosis patterns, and microenvironmental remodeling, providing a framework to link histologic features with underlying mechanisms and clinically relevant outcomes. By moving beyond global fibrosis staging to “where and how” fibrosis develops, AI-based approaches can elucidate both disease biology and individualized patient risk.

The potential of AI-based digital pathology spans several interrelated areas. First, AI-based spatial analysis may help explain the heterogeneity of fibrosis across SLD subtypes. MASLD, MetALD, and ALD represent distinct constellations of metabolic stress and alcohol-related injury, which may give rise to different regional patterns of fibrotic remodeling. In a recent study employing AI-based quantitative fibrosis assessment with second harmonic generation imaging in 88 biopsy samples, investigators demonstrated that, despite comparable overall fibrosis stages, MetALD and ALD exhibited significantly greater collagen density in periportal and zone 2 regions than MASLD.^[4] Similarly, lean patients with MASLD may exhibit different zonal distributions and collagen architectures compared with non-lean patients, reflecting differences in metabolic pathways, lipid deposition, and local inflammatory responses. Mapping these spatial patterns could provide both diagnostic insight and a framework for understanding subtype-specific pathophysiology.

Second, digital pathology enables mechanistic studies of SLD by dissecting how spatially resolved microenvironmental factors drive fibrosis. Regional variations in hepatic microcirculation and blood flow along the hepatic vascular axis modulate exposure of hepatocytes and stellate cells to metabolic substrates, alcohol, and inflammatory mediators. Metabolic dysfunction, cardiometabolic risk factors, and alcohol toxicity may therefore exert spatially distinct effects, promoting localized stellate cell activation, chronic inflammation, and vascular remodeling, which shape collagen deposition and extracellular matrix architecture. Each of these risk factors can contribute to liver fibrosis progression, but the relative influence of different factors in specific hepatic zones requires further investigation.^[5,6] High-resolution spatial mapping integrated with molecular profiling can clarify whether metabolic inflammation and alcohol-induced injury act via separate or overlapping pathways. This mechanistic insight may inform the development of targeted therapies by identifying zone-specific and microenvironment-dependent fibrogenic pathways, clarifying how interventions reshape spatial fibrosis architecture, and supporting rational combination strategies that address both hepatic and cardiometabolic drivers of disease.^[7]

Third, AI-based spatial analysis supports precision classification and individualized management by linking fibrosis patterns to clinical outcomes.^[8] Patients with MASLD show heterogeneous trajectories: some primarily develop liver-related events such as cirrhosis or hepatocellular carcinoma, while others are prone to cardiometabolic complications.^[9] AI-derived spatial signatures can identify fibrosis patterns predictive of specific outcomes. For example, prior AI-derived spatial analyses have shown that fibrosis progression in central and pericentral regions correlates with subsequent renal function decline.^[10] Additionally, spatial analysis can define a cardiovascular-risk predominant MASLD subtype, characterized by fibrosis patterns reflecting systemic metabolic stress and a higher risk of cardiovascular events. Recognizing this subtype enables clinicians to prioritize cardiovascular risk management, including lifestyle modification, blood pressure and lipid control, and cardiac monitoring, while simultaneously addressing liver-directed therapy. By integrating hepatic and systemic risk stratification, AI-based digital pathology provides a pathway toward precision hepatology and heart-liver co-management.^[11]

Fourth, digital pathology may support region-specific monitoring and therapy by distinguishing areas of stable versus rapidly progressing fibrosis within the same liver.^[12] Fibrosis is rarely uniform, and different regions may progress at different rates due to local variations in blood supply, inflammation, or stellate cell activation. Identifying rapidly progressing regions enables clinicians to prioritize surveillance and interventions in areas at highest risk, while avoiding unnecessary procedures in more stable regions. Such targeted monitoring could inform decisions on biopsy sampling, imaging follow-up, or even region-specific therapeutic approaches in the future. Moreover, mapping fibrosis dynamics at high spatial resolution may help anticipate complications such as portal hypertension or segmental liver dysfunction by linking structural remodeling to functional consequences. By capturing the intrahepatic heterogeneity of disease progression, AI-based digital pathology provides a precise framework for individualized, risk-adapted patient management.

Fifth, AI-based digital pathology offers unique opportunities in clinical trials by providing objective, quantitative, and spatially resolved endpoints.^[13] Traditional histologic scoring often lacks sensitivity to detect subtle changes or regional effects of interventions. By contrast, AI-derived measures such as qFibrosis can capture both progression and regression of fibrosis on a continuous scale across multiple liver regions. For instance, strengthened lifestyle intervention in MASLD patients produced pronounced fibrosis regression in the periportal region, as detected by qFibrosis, whereas conventional histology failed to capture these regional differences.^[14] Importantly, different pharmacologic agents may act on distinct microenvironments or fibrosis patterns within the liver, and AI-based spatial analysis can reveal these region-specific effects. Such insights enable the rational design of combination therapies targeting complementary fibrogenic pathways or spatially distinct microenvironments, thereby addressing the multifactorial nature of SLD and potentially enhancing overall efficacy.^[15] Spatially resolved, quantitative endpoints, therefore, facilitate earlier and more precise evaluation of treatment effects, improve differentiation between intervention groups and placebo responses, and may reduce sample size requirements or study duration. By integrating digital pathology into clinical trials, researchers can accelerate drug development, optimize patient selection, and better understand how interventions reshape liver architecture at a mechanistic level, ultimately supporting precision and combined therapeutic strategies.^[13]

Despite its transformative potential, several challenges must be addressed before AI-based digital pathology can be widely implemented in clinical practice. First, standardization of image acquisition and algorithm training is critical, as variations in slide preparation, staining protocols, scanning resolution, and even tissue handling can introduce bias and affect the consistency of extracted features. Without rigorous standardization, results may not be comparable across laboratories or studies. Second, cross-center reproducibility remains a major hurdle. Algorithms trained in one center may perform differently in another due to differences in patient populations, equipment, or local histology practices, limiting generalizability. Third, cost considerations and infrastructure requirements may pose barriers, as high-resolution scanners, computational resources, and specialized personnel are necessary to implement AI-based workflows, potentially restricting access to larger or resource-limited centers. Fourth, validation in large and diverse cohorts is essential. Current studies are often limited by small sample sizes, homogenous populations, or single-center data, raising concerns about the robustness and reliability of identified spatial patterns when applied to broader patient populations. Finally, linking spatial histologic phenotypes to clinically meaningful outcomes requires carefully designed prospective studies. Only by correlating AI-derived tissue signatures with both hepatic endpoints (e.g., cirrhosis progression, hepatocellular carcinoma) and systemic outcomes (e.g., cardiovascular events, renal function) can we determine whether these digital biomarkers truly guide individualized management strategies and therapeutic decision-making. Addressing these challenges is crucial for translating the promise of AI-based digital pathology into actionable clinical tools.

Conclusion

In conclusion, AI-based digital pathology should be regarded not merely as an advanced diagnostic tool but as a platform for understanding the spatial logic of liver disease. By capturing heterogeneity, linking microenvironmental mechanisms to fibrosis architecture, and associating patterns with distinct clinical outcomes, digital pathology provides a roadmap for precision hepatology. In an era where metabolic liver disease and cardiovascular disease are increasingly intertwined, leveraging spatial tissue analytics to inform individualized, integrated management may represent one of the most impactful directions for both research and clinical care.

Financial Disclosure

Ming-Hua Zheng serves as a speaker for AstraZeneca, Hisky Medical Technologies, and Novo Nordisk; as a consultant for Boehringer Ingelheim and Eieling Technology; and has received consulting fees from Boehringer Ingelheim. Yusuf Yilmaz has received personal consultancy fees from Novo Nordisk and Zydus.

Conflict of Interest

The remaining authors declare no conflicts of interest.

Use of Artificial Intelligence

The authors used an AI-assisted language editing tool (ChatGPT, OpenAI) to improve the clarity and grammar of the manuscript. The authors reviewed and edited the output and take full responsibility for the content of the publication.

Author Contributions

Concept: X-DZ, M-HZ; Design: M-HZ; Supervision: M-HZ; Data Collection and/or Processing: X-DZ, M-HZ; Analysis and/or Interpretation: X-DZ, YY, M-HZ; Literature Review: M-HZ; Writing: X-DZ, YY, M-HZ; Critical Review: X-DZ, YY, M-HZ.

Peer-review

Externally peer-reviewed.