Modeling Liver Fibrosis Progression in Patients With Viral Hepatitis Using the Machine Learning Tool Subtype and Stage Inference (SuStaIn)

A Suzuki, Katsuhiro Sano, Yuya Saito, P. A. Wijeratne, Kotaro Yamamoto, Shohei Fujita, Hyung Jun Woo, Nobuo Tomizawa, Koji Kamagata, Hiroshi Imamura, Shuichiro Shiina, Akio Saiura, Kenichi Ikejima, Ryohei Kuwatsuru, Yoshitaka Masutani, Daniel C. Alexander, Shigeki AokiCureus 2025 · DOI

Abstract

Background and aim: The natural progression of liver fibrosis and its association with biomarker changes have not been fully established. This study aimed to investigate liver fibrosis progression in patients with hepatitis B virus (HBV) and hepatitis C virus (HCV) infection using a novel machine learning tool called 'Subtype and Stage Inference (SuStaIn).'

Methods: This single-center retrospective study included 168 patients (29 HBV, 50 HCV, 89 controls). All patients underwent gadoxetic acid-enhanced MRI. Biomarkers included serum markers (FIB-4, APRI) and imaging markers (Q-LSC, CEI, RV/TV). SuStaIn was used to identify disease progression patterns from cross-sectional data.

Results: The sequence of biomarker conversion was determined to be APRI, FIB-4, CEI, Q-LSC, and finally RV/TV. This sequence was consistent across both HBV and HCV groups.

Conclusions: Abnormalities in serum biomarkers occur first, followed by functional decline in gadoxetic acid uptake, and finally morphological changes like right lobe atrophy.

Introduction

Liver fibrosis, the progressive scarring of liver tissue due to chronic injury from viruses like Hepatitis B (HBV) and C (HCV), is a major clinical concern. While liver biopsy remains the diagnostic gold standard, its invasiveness and cost limit its use for routine monitoring. Non-invasive alternatives, such as serum biomarkers and MRI-based imaging, have been developed, but the chronological order in which these markers change during disease progression has remained unclear. Understanding this sequence is critical for early intervention and patient stratification. This study utilizes machine learning to infer a long-term timeline from cross-sectional data.

Methods and Materials

The study utilized Subtype and Stage Inference (SuStaIn), an unsupervised machine learning algorithm originally developed for neurodegenerative diseases. SuStaIn reconstructs disease timelines from cross-sectional data by identifying the most likely sequence of biomarker "conversions" (moving from normal to abnormal states).

Patient Cohort

The retrospective analysis included 168 individuals selected from an initial pool of 300 patients suspected of having liver disease.

Figure 1: Patient selection flowchart illustrating the inclusion and exclusion criteria for the study cohort.

The final cohort consisted of 29 HBV patients, 50 HCV patients, and 89 controls (low-risk patients with normal ALT levels and no chronic liver disease).

Biomarkers and Imaging

Five biomarkers were analyzed:

  1. Serum: APRI (aspartate aminotransferase to platelet ratio index) and FIB-4 (fibrosis-4 index).
  2. Functional Imaging: Quantitative liver-spleen contrast ratio (Q-LSC) and contrast index of liver-muscle (CEI) derived from gadoxetic acid-enhanced MRI. These measure the liver's active uptake of contrast agents.
  3. Anatomical Imaging: Right liver lobe to total liver volume ratio (RV/TV), measuring physical atrophy.

Results

The SuStaIn algorithm identified a consistent "Serum → Function → Form" progression pipeline across the study population. While the algorithm initially looked for different "subtypes" of progression, it found that the vast majority of patients (all HBV and 47/50 HCV) fit into a single, consistent progression model.

The Sequence of Decline

The most probable order of biomarker conversion was determined to be:

  1. APRI (Serum)
  2. FIB-4 (Serum)
  3. CEI (Functional Imaging)
  4. Q-LSC (Functional Imaging)
  5. RV/TV (Morphological/Anatomical)

Key Findings

  • Universality: The progression pattern was nearly identical for both HBV and HCV, suggesting a common biological pathway for viral-induced fibrosis regardless of the specific viral trigger.
  • Early Warning: APRI was identified as the earliest marker to become abnormal, suggesting it may be the most sensitive tool for initial screening.
  • Functional vs. Structural: Functional MRI markers (CEI/Q-LSC) showed abnormalities before structural changes (RV/TV) were detectable, confirming that functional decline precedes physical atrophy.

Discussion and Conclusion

This study demonstrates that machine learning can effectively reconstruct disease timelines from one-time patient snapshots. The findings suggest that biochemical changes in the blood are the "canaries in the coal mine," followed by a decrease in the liver's functional ability to process contrast agents, and finally, physical shrinkage of the right lobe.

While the study is limited by its single-center retrospective design and focus on early-to-moderate progression, it provides a clear chronology of liver fibrosis. This "timeline" offers a roadmap for clinicians to better interpret non-invasive tests and intervene earlier in the disease process. Future work should involve longitudinal data to validate the SuStaIn-predicted sequence and explore if these patterns hold for other conditions like metabolic-associated steatohepatitis (MASH).

Summary updated

Abstract

Background and aim The natural progression of liver fibrosis and its association with biomarker changes have not been fully established in the literature. This study aimed to investigate liver fibrosis progression in patients with hepatitis B virus (HBV) and hepatitis C virus (HCV) infection using a novel machine learning tool called 'Subtype and Stage Inference (SuStaIn).' SuStaIn can identify disease progression patterns and subgroups from cross-sectional biomarker data. Methods This single-center retrospective study included 168 consecutive patients (mean age, 67.0 years ± 13.2; 91 male), comprising 29 with HBV, 50 with HCV, and 89 controls. All patients underwent gadoxetic acid-enhanced magnetic resonance imaging between January 2019 and December 2019. Imaging biomarkers measured were quantitative liver-spleen contrast ratio (Q-LSC), contrast index of liver-muscle signal intensity (CEI), and right liver lobe to the total liver volume ratio (RV/TV), while the serum biomarkers were fibrosis-4 index (FIB-4) and aspartate aminotransferase to platelet ratio index (APRI). SuStaIn was applied using z-scored biomarkers derived from each biomarker relative to the control group. Results The most likely ordering of the biomarker conversion in liver fibrosis progression was determined to be APRI, FIB-4, CEI, Q-LSC, and RV/TV. This sequence was observed to be consistent in all patients infected with either HBV or HCV, regardless of the virus type. Conclusions Using SuStaIn, new insights into liver fibrosis progression in patients infected with either HBV or HCV were achieved: abnormalities in serum biomarkers occurred first, followed by a decrease in hepatic uptake of gadoxetic acid, ultimately resulting in right lobe atrophy.
BibTeX
@inproceedings{suzuki2025modeling,
  author = {A Suzuki and Katsuhiro Sano and Yuya Saito and P. A. Wijeratne and Kotaro Yamamoto and Shohei Fujita and Hyung Jun Woo and Nobuo Tomizawa and Koji Kamagata and Hiroshi Imamura and Shuichiro Shiina and Akio Saiura and Kenichi Ikejima and Ryohei Kuwatsuru and Yoshitaka Masutani and Daniel C. Alexander and Shigeki Aoki},
  title = {{Modeling Liver Fibrosis Progression in Patients With Viral Hepatitis Using the Machine Learning Tool Subtype and Stage Inference (SuStaIn)}},
  booktitle = {Cureus},
  year = {2025},
  doi = {10.7759/cureus.78744},
}
Figure