The distribution of multi-center medical datasets is different,and the generalization of the model trained by single-center datasets is often poor,resulting in great limitations in application.Mixup training can effectively improve the generalization of the model,and the model fusion method based on Dempster-Shafer evidence theory(DST)can effectively fuse the best decision of multiple mod-els.Therefore,we propose an effective model for the diagnosis of lumbar disc herniation in response to the poor generalization of medical models trained by single-center datasets.The generalization of the model is enhanced by Mixup training,and the best decision is obtained by the method of multi-model decision fusion based on DST.After testing on the external test set,the method obtains 88.22%classifi-cation accuracy,88.12%F1 score and AUC value of 87.69%.
lumbar magnetic resonance imaginglumbar disc herniationMixupmulti-model fu-sion decision
国家科技期刊平台
NSTL
万方数据
