Esophageal fistula risk prediction method based on clinical data and radiomics characteristics

Abstract

The invention discloses an esophageal fistula risk prediction method based on clinical data and iconomics characteristics, relates to the technical field of medical artificial intelligence, and combines deep learning and a traditional iconomics method to extract rich iconomics characteristics through multiple image processing technologies so as to predict the risk of esophageal fistula. The generalization ability and robustness of the model are ensured through a multi-stage feature selection strategy, and a plurality of prediction models based on different feature combinations are developed; comprising a clinical feature model, a manual radiomics model, a deep learning radiomics model, a clinical-deep learning radiomics model and a clinical-manual radiomics model, through cross validation and model optimization, the high efficiency of each model in clinical prediction is ensured, the performance of the models is comprehensively compared, and the accuracy of clinical prediction is improved. According to the esophageal fistula risk prediction method, the esophageal fistula risk prediction accuracy after the esophageal cancer patient receives radiotherapy or chemoradiotherapy is improved, a powerful basis is provided for treatment decision of the esophageal cancer patient, and the esophageal fistula risk prediction method has a wide clinical application prospect.

Images