Multi-task prediction method for brain invasion classification and meningioma classification

The invention discloses a multi-task prediction method for brain invasion classification and meningioma classification, which comprises three stages: in a multi-modal image feature fusion stage, image features of different modals are respectively extracted and fused; in the feature decoupling stage, a contrast learning technology is introduced, and the task specific features and the aligned task common features for the same task are more similar than the task specific features and the aligned task common features for different tasks by comparing a loss function, so that the prediction capability of each task specific feature is enhanced. In the multi-task prediction stage, the main prediction branch fuses the task specific features corresponding to the same task with the aligned task common features, and predicts the two tasks at the same time. The method is a multi-task prediction method simultaneously considering brain invasion classification and meningioma classification, preoperative prediction of brain i