Machine Learning for Predicting Colorectal Precancerous Lesions: A Comparative Study

Abstract

Early screening and identification of precancerous lesions in colorectal cancer (CRC) are crucial for improving prognosis, and the application of machine learning in this field has achieved significant progress. This article systematically reviews the multimodal features of CRC precancerous lesions, covering multiple dimensions such as endoscopic images, molecular markers, genomics, gut microbiota, and Traditional Chinese Medicine (TCM) syndrome patterns. It further compares the performance and applicability of three types of prediction methods: traditional machine learning, deep learning, and fusion learning. On this basis, the article proposes that by constructing standardized multimodal datasets, developing data fusion and causal inference methods, and enhancing model interpretability, it is possible to improve the accuracy and clinical translation potential of prediction models, providing theoretical and technical references for early warning and precise prevention of CRC.