Fig. 6

Study overview. This study incorporates both NCCT and clinical information for HT prediction. The WSDL model includes a pipeline of preprocessing, ImageNet pretrained dynamic convolution neural network (DCNN) and AS loss. The baseline DL was built without AS loss to output the prediction probability. For the ML models, both DL-based features and clinical information were combined with feature engineering to give the predictions. The system produces seven outputs, including predictions of five ML models, the WSDL model and the baseline of the WSDL model