PathAI Blog

The Power of LiverExplore: Quantifying Fibrosis with Unprecedented Precision

Written by Admin | Nov 6, 2025 2:00:00 PM

Liver fibrosis is a significant complication of chronic liver diseases, including metabolic dysfunction-associated steatohepatitis (MASH). Advanced fibrosis (F3-F4) is a key indicator for MASH outcomes, with the risk of mortality and other severe morbidities (e.g., liver cancer) increasing as fibrosis progresses.

Despite advanced fibrosis being previously considered irreversible, recent evidence suggests that these fibrosis patterns can, indeed, regress. While the potential for regression of advanced fibrosis offers hope to patients suffering from this complication, there remain limited therapeutic options for patients with MASH and fibrosis. Within the last two years, resmetirom (a THR-β agonist) and semaglutide (a GLP-1 receptor agonist) have been approved for MASH patients with F2-3 fibrosis. Currently, there are no approved therapies for patients with compensated cirrhosis (F4). As the MASH therapeutic landscape continues to expand, a clear opportunity exists to better understand how the spectrum of fibrosis changes during disease progression in patients with MASH to more effectively treat patients based on their unique fibrotic state.

There are two key problems facing the study of advanced fibrosis in MASH:

  1. Variability in pathologist interpretation – Previously, evaluation of fibrosis in liver biopsies has required specialized stains (Masson trichrome, picrosirius red) or imaging (second harmonic generation), rather than routine H&E staining, making it difficult to assess fibrosis alongside other histologic features. This issue, combined with subjective, categorical fibrosis scoring criteria lead to a high degree of variability in how pathologists interpret fibrosis present within liver biopsies.
  2. Semi-quantitative fibrosis measurement – The fibrosis scoring criteria defined by the MASH CRN scoring guidelines and used by pathologists is semi-quantitative in nature. Furthermore, the underlying biology of liver fibrosis is multifaceted. The ability to quantify the total amount of liver fibrosis in a continuous manner, as well as the amounts of each fibrosis subtype, would vastly improve our understanding of the changes in fibrosis that accompany disease progression and, potentially, regression, capturing changes that are missed through ordinal scoring alone. A more granular, nuanced understanding of advanced fibrosis could lead to better therapeutics and clinical trial designs.

These challenges highlight the need for a deeper understanding of the changes in liver microarchitecture that accompany the spectrum of fibrosis in MASH and related liver diseases. As pathology workflows become increasingly digital in nature, an opportunity exists to leverage AI-powered algorithms to overcome these challenges.

What if investigators could gain comprehensive, quantitative insights into the composition of the liver parenchyma – including fibrosis – from a routine, H&E-stained liver biopsy? This is the transformative potential of LiverExplore*. Developed by PathAI, LiverExplore is a suite of deep-learning models trained to exhaustively predict and quantify tissue regions, cell types, lobular zones, and fibrotic subtypes directly from H&E-stained WSIs of liver biopsies [1].

Quantifying Fibrosis from H&E with LiverExplore

LiverExplore generates interpretable overlays that visualize model-derived segmentations of input H&E-stained WSI into biologically relevant regions: artifact, tissue, cell, lobular zones, and fibrosis subtype (Figure 1). This blog will focus on the fibrosis subtyping model within LiverExplore.

LiverExplore’s fibrosis model leverages Inferred Quantitative Multimodal Anisotropy Imaging (iQMAI) [2], a machine learning approach that allows the identification and quantification of collagen directly from an H&E-stained WSI at scale, bypassing the need for specialized stains or complex imaging to evaluate fibrosis in histologic specimens. Powerfully, this  model detects and exhaustively segments all collagen across entire input H&E-stained WSIs into non-pathological collagen (structural collagen, portal collagen, perivenular fibrosis) as well as categories of fibrosis associated with different degrees of liver disease severity (perisinusoidal fibrosis, periportal fibrosis, incomplete septal fibrosis, complete septal fibrosis, and nodular fibrosis) (Figure 2). These predictions are visible as heatmap overlays on the input WSIs, providing visual interpretability to model predictions.

Importantly, LiverExplore fibrosis overlays enable the calculation of over 500 human-interpretable features (HIFs) quantifying the areas of liver fibrosis subtypes and the cellular composition in the vicinity of these subtypes. The HIFs reveal the granularity and complexity of this aspect of the liver microarchitecture. The simplest fibrosis-related HIFs quantify the area of predicted fibrosis subtypes within usable tissue (i.e., free of artifact) across the entire WSI. Combining LiverExplore-derived predictions of different regions yields novel spatial information that is difficult to evaluate by manual review (for example, the proximity of ballooned hepatocytes to perisinusoidal fibrosis). Overall, these HIFs provide a quantitative and comprehensive characterization of fibrosis in the liver microarchitecture.

Figure 1. LiverExplore overlays and detected classes

Figure 2. Representative biopsies showing LiverExplore fibrosis subtyping predictions spanning fibrosis stages F0 through F4, Scale Bars: 200 µm

How does the liver fibrotic composition change at a granular level as MASH progresses?

The MASH CRN staging guidelines are commonly used by pathologists for manual fibrosis scoring. Based on these criteria, patients are classified into five fibrosis stages, depending on the fibrosis subtypes present in a biopsy (summarized in Figure 2). Stage F0 reflects no fibrosis. Stage F1 is assigned to patients with perisinusoidal or periportal fibrosis, while patients with both perisinusoidal and periportal fibrosis are classified as stage F2. Patients with biopsies containing bridging fibrosis (i.e., complete septal fibrosis) are considered stage F3, and patients with cirrhosis (i.e., nodular fibrosis) are staged as F4. Apart from these broad, semi-quantitative criteria, the natural history of fibrosis within MASH (i.e., the changes in fibrosis subtypes as MASH progresses) remains unclear.

Given its ability to uniquely characterize fibrosis subtypes from H&E images, LiverExplore enables direct quantification of fibrotic composition beyond the categorical CRN stages, providing an unprecedented level of granularity through which fibrosis phenotypes can be studied.

We hypothesized that fibrosis subtypes progressively change along a continuum with increasing disease severity, rather than as a stepwise process as reflected in the CRN scoring criteria. Using a cohort of patients enrolled in the STELLAR-3 and STELLAR-4 clinical trials (patients with F3-F4 MASH at trial enrollment) [3], we compared LiverExplore-calculated metrics of the amount of fibrosis subtypes across AI-derived continuous fibrosis stages obtained using AIM-MASH+ [4].

As fibrosis severity increased, we quantified changes in the breakdown of fibrosis subtypes within MASH biopsies (Figure 3):

  • Nodular fibrosis: Beginning at a continuous fibrosis score of 3.0, the amount of nodular fibrosis within total tissue began to progressively increase alongside fibrosis severity. At continuous scores of 4.0 and higher, the proportion of nodular fibrosis expanded at a much greater rate, ultimately accounting for the vast majority of fibrosis within samples with extremely high continuous fibrosis scores.
  • Complete septal (bridging) fibrosis: This advanced fibrosis subtype was detected in biopsies with continuous fibrosis scores slightly above 2.0 and became progressively more abundant as a percent of total tissue in parallel with fibrosis severity. A maximum percentage of complete septal fibrosis was measured in specimens with continuous scores near 4.0. At continuous scores greater than 4.0, complete septal fibrosis decreased in abundance as nodular fibrosis expanded.
  • Perisinusoidal fibrosis: This fibrosis subtype was identified in cases with continuous fibrosis scores between 1.0 and 2.0, with its amount progressively increasing with worsening fibrosis severity. Similar to complete septal fibrosis, perisinusoidal fibrosis decreased slightly in abundance in continuous fibrosis scores greater than 4.0.

These results demonstrate the spectrum of fibrosis present within the categorical fibrosis stages. Furthermore, the quantification of fibrosis progression at the sub-ordinal level underlines the variability present within a single fibrosis category. These results highlight the value of quantifying the fibrosis continuum in MASH, which may lead to a more granular understanding of how fibrosis may progress and, by extension, potentially regress.

Figure 3. Quantitative changes in MASH fibrosis composition revealed at sub-ordinal  resolution with increasing disease severity. Continuous fibrosis scores were derived from AIM-MASH+. Bars show feature values averaged across all samples falling within each continuous fibrosis score interval for predicted fibrosis subtypes.

Can prognostic features be identified by LiverExplore?

Given the changes in LiverExplore features accompanying fibrosis progression, we hypothesized that these features may be prognostic of outcome. Using the STELLAR-3 and STELLAR-4 cohorts, we tested individual LiverExplore HIFs from baseline biopsies to identify their association with progression to cirrhosis or other liver-related events.

In both STELLAR-3 and STELLAR-4, LiverExplore features quantifying nodular or advanced fibrosis were associated with higher risk of liver-related events. Conversely, features quantifying incomplete septal fibrosis in STELLAR-3, and features quantifying complete septal fibrosis and densities of hepatocytes in proximity to fibrosis in STELLAR-4, were associated with lower liver-related event risk. These results are likely driven by the distinct histologies within each study due to their individual enrollment criteria (F3 and F4 disease, respectively).

The ability to more thoroughly detect and quantify histologic changes in fibrosis may lead to a better understanding of the features associated with disease progression and regression and inform future biomarker development for precision medicine or combination therapy approaches in MASH.

Figure 4. Association between LiverExplore features in baseline samples and clinical outcomes from A) STELLAR-3 and B) STELLAR-4. Each point represents a single feature’s nominal p-value and effect size from Cox regression. Large colored markers indicate features exceeding FDR-adjusted significance.

Conclusions and a Look Ahead

LiverExplore is a suite of digital pathology models that provide comprehensive, granular quantifications of histologic features – including fibrosis – from routine, H&E-stained liver biopsies. LiverExplore quantifies the areas of disease relevant fibrosis subtypes within a liver biopsy in an automated fashion, directly addressing the major challenges plaguing the study of liver fibrosis: variable and semi-quantitative assessment.

Importantly, LiverExplore’s collagen predictions utilize an iQMAI framework, which allows the investigation of additional features relating to collagen morphology, such as fiber width and tortuosity [2]. Future work will continue to capitalize on the power of iQMAI, allowing the quantification of collagen morphology in conjunction with the fibrosis subtypes described in this blog. The ability to study fibrosis at this level of detail from a routine H&E-stained biopsy has the potential to unlock key insights into the liver fibrotic microenvironment.

The spectrum of therapeutics being investigated in MASH clinical trials is expanding, and combination trials may be on the horizon. LiverExplore has the unique potential to dissect therapeutic-induced changes in fibrosis using digitized images of preexisting H&E-stained trial biopsies, thus yielding critical evidence of fibrotic regression in these trials. Future blogs will demonstrate examples of such insights gained from LiverExplore in post hoc analyses of MASH trials.

* LiverExplore and AIM-MASH+ are For Research Use Only. Not for use in diagnostic procedures.

References

  1. Stanford-Moore, A., et al. (2025) Comprehensive characterization of granular fibrotic and cellular features in liver tissue enabled by deep learning models." medRxiv 2025.06.12.25328580.
  2. Nguyen, T.H. and Zhang, J., et al. (2025) Quantification of collagen and associated features from H&E-stained whole slide pathology images across cancer types using a physics-based deep learning model. bioRxiv 2025.03.17.643273.
  3. Harrison, S.A., et al. (2020) Selonsertib for patients with bridging fibrosis or compensated cirrhosis due to NASH: Results from randomized phase III STELLAR trials. J Hepatol 73:26–39.
  4. Iyer, J.S. (2024) AI-based automation of enrollment criteria and endpoint assessment in clinical trials in liver diseases. Nature Med. 30:2914–2923