Uncertainty-aware and causal test-time adaptive foundation model for robust colorectal cancer pathology diagnosis

1

Colorectal cancer (CRC) is a leading malignancy globally, with over 1.9 million new cases and 935,000 deaths estimated in 2020.
2

Current computational pathology models face challenges like domain shifts, unreliable uncertainty estimation, and spurious correlations.
3

The Uncertainty-Aware and Causal Test-Time Adaptive Foundation Model (UAD-FM) integrates uncertainty decomposition and causal adaptation.
4

UAD-FM demonstrates superior accuracy, calibration, and domain robustness across five public CRC datasets compared to existing models.
5

The model produces interpretable uncertainty maps, enhancing collaboration between AI systems and human pathologists for reliable diagnosis.

Npj Digital Medicine

by Shenghan Lou, Genshen Mo, Xiao Zhang, Hao Wang, Hao Li, Keru Ma, Huiying Li, Xinyue Zhang, Meihong Yan, Haonan Xie, Yuze Huang, Chuangqi Li, Siyuan Ma, Hongxue Meng, Lei Cao, Peng Han
December 6, 2025

1