Histological AI-Assisted Analysis Outperforms Pathologist Assessment in Predicting Disease Outcomes in PSC Patients

Area of Science:

• Hepatology and digital pathology.
• Application of artificial intelligence in medical diagnostics.

Background:

• Liver biopsy is crucial for prognostication in primary sclerosing cholangitis (PSC).
• Manual histological assessment in PSC is subjective and lacks precise quantification.
• Key histological features include portal tracts, fibrosis, biliary epithelium, portal inflammation, and vasculature.

Purpose of the Study:

• To develop and validate a neural network model for precise quantification of liver histology in PSC.
• To assess the prognostic significance of AI-derived histological features for clinical endpoints in PSC.

Main Methods:

• A neural network model was trained on 603 liver sections from PSC patients, other liver diseases, and controls.
• The model quantified key histological features: portal tracts, fibrosis, biliary epithelium, portal inflammation, and vasculature.
• Model performance was evaluated for predicting liver transplantation (LT) due to end-stage liver disease (ESLD) and a composite endpoint (LT, cholangiocarcinoma, liver-related death).

Main Results:

• AI-derived portal inflammation and fibrosis outperformed manual scoring in predicting LT due to ESLD (AUC 0.86-0.87 vs. 0.81-0.82).
• AI-derived vascular area and biliary epithelium also showed prognostic value for LT due to ESLD (AUC 0.80, 0.83).
• AI-derived portal inflammation was independently associated with the composite endpoint (HR 2.51).

Conclusions:

• AI-measured portal inflammation is a robust predictor of clinical outcomes in PSC.
• AI-derived measures of fibrosis, vascular area, and biliary epithelium also hold prognostic value.
• AI-based histological assessment surpasses manual scoring, highlighting its potential as a surrogate endpoint in PSC.