Advanced Science: CRCFound, a self-supervised CT foundation model for colorectal cancer

Our team introduced CRCFound in Advanced Science, a self-supervised colorectal cancer CT foundation model pretrained on large-scale unlabeled 3D CT data.

CRCFound was systematically evaluated on multiple downstream tasks, including TNM staging, MSI/CMS-related prediction, and survival prognosis. The model significantly improved transfer performance versus training from scratch.

Key points:

• Pretrained with 5,137 unlabeled CRC CT scans.
• Benchmarked on 8 clinically relevant downstream tasks.
• Delivered stronger generalization and practical performance for diagnosis and prognosis modeling.

Paper: CRCFound: A Colorectal Cancer CT Image Foundation Model Based on Self-Supervised Learning