从病理图像到分子预测：多模态人工智能驱动的结直肠癌MSI检测新范式

From histopathological imaging to molecular prediction: a multimodal AI-driven paradigm for MSI detection in colorectal cancer

摘要:

结直肠癌是全球发病率和病死率均居前列的消化系统恶性肿瘤。微卫星不稳定性（MSI）是结直肠癌的重要分子标志物，对患者预后评估及免疫治疗决策具有关键指导意义。传统检测方法（如免疫组化、PCR、二代测序）虽已标准化，但仍受限于成本高、操作复杂和结果不一致。近年来，人工智能（AI）在结直肠癌MSI检测中展现出显著潜力，通过整合病理图像、基因组学及影像学等多模态数据，利用深度学习实现MSI的精准预测，推动癌症诊疗从经验模式向数据驱动范式转变。本文综述了AI结合多模态医学数据构建结直肠癌MSI检测模型的最新进展，比较不同技术路径的性能特点与临床转化瓶颈，并展望多模态融合、模型泛化及可解释性优化等未来方向，为精准医疗提供新思路。

Abstract:

Colorectal cancer (CRC) ranks among the leading causes of cancer incidence and mortality worldwide. Microsatellite instability (MSI) is a key molecular biomarker with important implications for prognosis and immunotherapy selection. Although conventional detection methods such as immunohistochemistry, PCR, and next-generation sequencing have been standardized, they remain limited by high costs, technical complexity, and inconsistent results. In recent years, artificial intelligence (AI) has shown great potential in MSI detection by integrating multimodal data that includes histopathological images, genomic information, and medical imaging to achieve accurate prediction and enable a data-driven paradigm in oncology. This review summarizes the latest advances in AI-based multimodal modeling for MSI detection in CRC, compares different methodological approaches and their translational challenges, and discusses future directions such as multimodal integration, model generalizability, and interpretability enhancement, providing new insights for precision medicine.