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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate risk stratification is vital for colorectal cancer (CRC) treatment.
• Current deep learning models struggle with CRC diagnosis due to limited annotated data and poor generalizability.

Purpose of the Study:

• To introduce CRCFound, a self-supervised learning-based CT image foundation model for CRC.
• To overcome the challenge of insufficient annotated data in CRC deep learning models.

Main Methods:

• CRCFound was pretrained on 5137 unlabeled CRC CT images using self-supervised learning.
• The model's adaptability was tested on six diagnostic and two prognosis tasks.

Main Results:

• CRCFound demonstrated strong performance and generalization ability across various CRC-related tasks.
• The model effectively transferred learning to most CRC diagnostic and prognosis tasks.
• It successfully addressed the issue of limited annotated data.

Conclusions:

• CRCFound offers a promising solution for accurate CRC diagnosis and personalized treatment.
• The foundation model shows significant potential for improving clinical applications in CRC.
• It enhances feature representation learning for CRC CT images.