Abstract: Highlights•A novel dual-stream network is proposed to exploit WSI pyramids for cancer prognosis.•A square pooling layer is devised to make the network more efficient in computation.•A cross attention pooling is developed to make dual-stream feature fusion smoother.•Our network could perform better than SOTA with an average improvement of 4.6%.•Ablation study verified the effectiveness of our dual-stream and cross-attention.
Loading