Auxiliary discrimination method and device for nasopharynx cancer radiation brain injury image and medium

The invention discloses an auxiliary discrimination method and device for a nasopharyngeal carcinoma radiation brain injury image and a medium. The method comprises the steps that an MRI image is collected and preprocessed, a temporal lobe area is segmented, a level prediction model is trained through the temporal lobe area of the MRI image and corresponding focus marks, and the probability that a radiation brain injury focus exists in a single temporal lobe image area is predicted; training a patient level prediction model by taking the probability that all the layers of each subject contain the focus output by the layer level prediction model as input and taking whether the subject has radiation brain injury as output; and finally, obtaining the radiation brain injury disease probability of a single nasopharynx cancer subject by using the trained level-level prediction model and the patient-level prediction model, and when a plurality of modals coexist, outputting a mean value of prediction results of the p