A Machine Learning Approach to Liver Histological Evaluation Predicts Clinically Significant Portal Hypertension in
Jaime Bosch1,2, Chuhan Chung3, Oscar M Carrasco-Zevallos4
1Department of Biomedical Research, University of Bern, Bern, Switzerland.
Hepatology (Baltimore, Md.)
|August 1, 2021
Summary
A machine learning model can predict clinically significant portal hypertension (CSPH) from liver histology in NASH patients. This AI tool shows promise for non-invasively assessing portal pressure, aiding in patient management.
Area of Science:
- Hepatology
- Medical Artificial Intelligence
- Digital Pathology
Background:
- Hepatic venous pressure gradient (HVPG) measurement is the gold standard for assessing portal hypertension but requires specialized expertise.
- Non-invasive methods for estimating portal pressure are needed, especially for patients with non-alcoholic steatohepatitis (NASH) and cirrhosis.
Purpose of the Study:
- To develop and validate a machine learning (ML) algorithm to extrapolate HVPG from liver histology in patients with NASH and compensated cirrhosis.
- To assess the ML model's ability to predict clinically significant portal hypertension (CSPH) and its association with clinical outcomes.
Main Methods:
- A convolutional neural network was trained on trichrome-stained liver biopsy slides from NASH patients (n=130).
- The ML model generated an ML HVPG score, which was validated in a separate test set (n=88).
- The ML HVPG score's correlation with measured HVPG, prediction of CSPH (HVPG ≥ 10 mm Hg), and association with clinical events were analyzed.
Main Results:
- The ML HVPG score showed a stronger correlation with measured HVPG (ρ=0.47) than hepatic collagen morphometry (ρ=0.28).
- The ML HVPG score effectively differentiated normal, elevated, and CSPH levels and demonstrated good performance for CSPH prediction (AUROC: 0.85 in training, 0.76 in test set).
- Improved CSPH discrimination was achieved by integrating ML parameters for nodularity, Enhanced Liver Fibrosis, platelets, AST, and bilirubin (AUROC: 0.85 in test set).
- Changes in ML HVPG score, but not baseline score, predicted clinical events and were associated with hemodynamic response and fibrosis improvement.
Conclusions:
- A machine learning model utilizing liver biopsy images can accurately predict CSPH in NASH patients with cirrhosis.
- This AI-driven approach offers a potential non-invasive method for assessing portal hypertension, complementing traditional HVPG measurements.
