Fig. 4From: Assessment of emerging pretraining strategies in interpretable multimodal deep learning for cancer prognosticationMultimodal modeling; Pretrained encoders are used to extract features from gene expression, DNAm, and WSI; embeddings are fused using Trilinear Fusion; prognosis is predicted using a feedforward neural network. Predicted hazards are dichotomized to generate a Kaplan Meier plot to compare low and high risk groupsBack to article page