Retinal Artery/Vein Classification Based on Multi-Scale Category Fusion

Retinal blood vessel morphological abnormalities are generally associated with cardiovascular, cerebrovascular, and systemic diseases, automatic artery/vein classification is particularly important for medical image analysis and clinical decision making. This work proposes a retinal artery/vein classification model based on multi-scale category fusion in order to improve the accuracy of arteriovenous classification. Aiming at the high similarity of arteriovenous features, a multi-scale feature extraction module is proposed to enhance feature extraction by aggregating multi-scale features of hierarchical residuals in Res2Net single residuals. Furthermore, a multi-layer semantic supervision structure is designed to supervise and fuse the arteriovenous features at different layers to obtain more semantic details to enhance the distinguishing ability of features. Finally, a category-weighted fusion module is introduced to concatenate the feature maps of the same category together to refine the overall segmentation results. The proposed method is verified on two public available fundus image datasets with different scales, namely, DRIVE and LES-AV. The experimental results show that the proposed method performs well in the task of arteriovenous classification and outperforms most of the existing methods.