[A new paradigm for precision diagnosis and treatment of colorectal cancer in the era of artificial intelligence--From foundation models to multi-agent collaboration].

The high heterogeneity of colorectal cancer poses significant challenges to clinical precision diagnosis and treatment. In recent years, artificial intelligence (AI) technology, represented by foundation models, has achieved breakthrough progress, offering a new opportunity to address this challenge. AI models based on a single data modality have demonstrated significant application potential in processing data such as imaging, genetics, and pathology; however, their independent analysis mode struggles to meet the needs of complex clinical decision-making. Based on a review of the current application status of AI technology in the field of colorectal cancer diagnosis and treatment, this article focuses on discussing the latest advances in research on single-modality foundation models, including those for imaging, genetics, pathology, and clinical text. It also conducts an in-depth analysis of the "information silos" bottleneck they face. Meanwhile, the article proposes the construction of a "multi-agent collaboration" system that integrates multimodal data and simulates the multidisciplinary team (MDT) consultation model, which is one of the future development directions for achieving precision diagnosis and treatment of colorectal cancer. Focusing on this new paradigm, we combine the preliminary explorations of our team in this field and look forward to its application potential in future clinical practice.