Self-supervised learning to predict intrahepatic cholangiocarcinoma transcriptomic classes on routine histology.

Area of Science:

• Digital Pathology
• Computational Pathology
• Oncology

Background:

• Intrahepatic cholangiocarcinoma (iCCA) transcriptomic classification offers prognostic and therapeutic insights but faces clinical adoption barriers due to tissue and cost limitations.
• Recent advancements refined iCCA classification into five distinct subclasses.

Purpose of the Study:

• To develop and validate a self-supervised learning (SSL) model for predicting iCCA transcriptomic classes directly from digitized whole-slide images (WSIs).
• To assess the feasibility of using routine histological slides for transcriptomic classification, bypassing traditional molecular analyses.

Main Methods:

• A Giga-SSL model was trained on 766 WSIs from 246 patients (discovery set).
• The model was validated on The Cancer Genome Atlas (TCGA) cohort (n=29) and a French external validation set (n=32).
• Performance was evaluated using Area Under the Curve (AUC) for predicting transcriptomic classes.

Main Results:

• The model demonstrated good to very good performance in predicting the four most frequent iCCA transcriptomic classes (AUC 0.63-0.84) in the discovery set.
• The hepatic stem-like class was predicted with high accuracy (AUC 0.84).
• Consistent performance was observed in validation sets, with AUCs ranging from 0.76 to 0.92.

Conclusions:

• An SSL-based model can accurately predict iCCA transcriptomic classes from routine biopsy and surgical histological slides.
• This computational approach facilitates clinical implementation of transcriptomic classification for improved iCCA prognostication and treatment guidance.
• The model offers a scalable and practical solution, potentially accelerating personalized medicine in iCCA management.