Random features meet MIL: a deep GP approach to colorectal MSI prediction

1

Colorectal cancer is a leading cause of cancer-related deaths, making early diagnosis essential for improving patient outcomes.
2

The proposed method integrates deep Gaussian processes with multi-instance learning to enhance colorectal cancer classification from weakly labeled data.
3

Experimental results on the TCGA-CRC dataset show the model achieves an AUC of 0.895, outperforming existing models like ResNet and EfficientNet.
4

The approach includes an attention-based aggregation mechanism, improving model interpretability by highlighting key regions in whole-slide images.
5

This research addresses challenges in colorectal cancer prediction, offering a robust tool for automated detection with potential clinical applications.

Npj Digital Medicine

by Shixuan Shen, Zeyang Wang, Tianmu Liu, Kangle Ma, Zhen Tian, Fuqiang Zhang, Qingyue Zhang
December 15, 2025

1