An attentional mechanism model for segmenting multiple lesion regions in the diabetic retina

Abstract Diabetic retinopathy, a leading cause of blindness in diabetic patients, necessitates precise lesion segmentation for effective grading. To address the challenges of low pixel occupancy and capturing multi-scale features in fundus images, we introduce a novel deep learning method: the Multi-Scale Spatial Attention Gate (MSAG) mechanism network. The model inputs images of various scales to extract diverse semantic information. Our innovative Spatial Attention Gate merges low-level spatial details with high-level semantic content, assigning hierarchical attention weights for accurate segmentation. Using the Modified Spatial Attention Gate in the inference stage enhances precision by combining prediction scales hierarchically, thus improving segmentation accuracy without increasing training costs. Testing on the IDRiD-S Dataset demonstrates that our fusion method surpasses others, with our model achieving balanced accuracy across all lesion types. Notably, our approach shows a 1\%-3\% higher accuracy for microaneurysms and overall accuracy compared to similar models, marking a significant advancement in diabetic retinopathy lesion segmentation.