CoADS: Cross attention based dual-space graph network for survival prediction of lung cancer using whole slide images.

•We present a novel framework for the overall survival prediction using pathological images.•The proposed method integrated both the morphological information in physical space and the contextual information in latent space.•The designed cross-attention module fused the instance information of WSIs across different branches effectively.•Extensive experiments have illustrated that our approach can exploit discriminative patterns and thus achieve competitive performance in survival analysis.