Cascaded Multi-Scale Feature Interaction for choroidal atrophy segmentation

The recent work has achieved great success in utilizing multi-scale feature ensembling for medical image segmentation. In this paper, we propose a new module called cascaded multi-scale feature interaction (CMSI) for choroidal atrophy segmentation in fundus images. The proposed CMSI module makes full use of multi-scale features, including using cascaded pooling and convolution to implement feature interactions at different scales and using strip pooling to capture long-distance features, which makes it more flexible than traditional convolution on the choroidal atrophy region with various scales in fundus image. Based on the U-shape network, we use the ResNet as the backbone to extract hierarchical feature representations. The proposed CMSI module is added at the top of the encoder path. In summary, our main contributions are summarized in two aspects as follows: (1) The CMSI module is proposed for multi-scale feature ensembling by cascading multi-scale pooling and strip pooling. (2) The Dice coefficients of our proposed network on choroidal atrophy segmentation increased by 4.15% compared to U-Net.