The invention provides a multi-constraint adaptive multi-
modal fusion
inflammatory bowel disease risk prediction method, which is applied to the crossing field of medical
data processing and
machine learning and is suitable for early diagnosis and clinical
risk stratification of
inflammatory bowel diseases. According to the method, multi-
modal omics data related to the
inflammatory bowel disease is collected, and a three-level
bioinformatics filtering strategy is adopted for preprocessing; each piece of preprocessed
modal data is coded, and modal orthogonal constraint is applied to retain modal specificity; dual-stage dynamic fusion is realized through a hierarchical attention mechanism, and modal weights are adaptively distributed; supervised comparative learning and maximum mean value difference loss are introduced, and feature discrimination and distribution consistency are enhanced; and finally, outputting an
inflammatory bowel disease risk prediction result based on the fusion features. According to the method, high-dimensional redundant features are effectively compressed, the
overfitting risk is reduced, modal specificity and complementarity are considered, the model generalization ability and subtype discrimination precision are improved, and reliable
technical support is provided for early diagnosis of the inflammatory bowel
disease.