Dual-Path and Multi-Scale Enhanced Attention Network for Retinal Diseases Classification Using Ultra-Wide-Field Images

Early computer-aided early diagnosis (CAD) based on retinal imaging is critical to the timely management and treatment planning of retina-related diseases. However, inherent characteristics images complexity their pathological patterns, such as low image contrast different lesion sizes, restrict performance CAD systems. Recently, ultra-wide-field (UWF) have become a useful tool for disease detection due capability capturing much broader view retina (i.e., up 200°), in comparison with most commonly used fundus (45°). In this paper, we propose an attention-based multi-branch network diseases classification four subject groups. The proposed method consists multi-scale feature fusion module dual attention module. Specifically, small-scale lesions are identified using features extracted from To better explore obtained features, global graph incorporated enable recognize salient objects interest. Comprehensive validations both private public datasets were carried out verify effectiveness model.