Multi-Scale Attention Network for Diabetic Retinopathy Classification

Diabetic Retinopathy (DR) is a highly prevalent complication of diabetes mellitus, which causes lesions on the retina that affect vision may lead to blindness if it not detected and diagnosed early. Convolutional neural networks (CNN) are becoming state-of-the-art approach for automatic detection DR by using fundus images. The high-level features extracted CNN mostly utilised classification retina. This representation capable classifying different classes; however, more effective detecting damages needed. paper proposes multi-scale attention network (MSA-Net) classification. proposed applies encoder embed image in representational space, where combination mid used enrich representation. Then feature pyramid included describe retinal structure locality. Furthermore, enhance discriminative power mechanism top model trained standard way cross-entropy loss classify severity level. In parallel as an auxiliary task, weakly annotated data detect healthy non-healthy surrogate task helps its distinguishing method when implemented has achieved outstanding results two public datasets: EyePACS APTOS.