A Multi-modality Network for Cardiomyopathy Death Risk Prediction with CMR Images and Clinical Information

Dilated Cardiomyopathy (DCM) is one of the main worldwide causes of sudden cardiac death (SCD). Early diagnostics significantly increases the chances of correct treatment and survival. However, there are no efficient methods for mortality risk prediction from learning cardiac magnetic resonance (CMR) image and clinical data due to the poor image quality and extreme imbalanced datasets. To solve this problem, we proposed an effective multi-modality network (MMNet) for mortality risk prediction in DCM, and we firstly directly optimize the AUC to train the multimodal deep learning classifier by maximizing the WMW statistic. This can achieve significant improvements in AUC, especially under the imbalanced learning problem. MMNet consists of two branches: clinical data branch and T1 mapping CMR images branch, which allows the model to learn more comprehensive features and makes a more accurate prediction. We validated our approach on a DCM dataset, which contains 450 CMR images that only holds 34 positive samples. Experimental results show that our approach archived accuracy of 98.89%, AUC of 99.61%, sensitivity of 100% and specificity of 98.8%, demonstrating the effectiveness of the proposed method.