Detecting and staging diabetic retinopathy in retinal images using multi-branch CNN

Purpose This paper aims to propose a solution for detecting and grading diabetic retinopathy (DR) in retinal images using convolutional neural network (CNN)-based approach. It could classify input into normal class or an abnormal class, which would be further split four stages of abnormalities automatically. Design/methodology/approach The proposed is developed based on newly CNN architecture, namely, DeepRoot. consists one main branch, connected by two side branches. branch responsible the primary feature extractor both high-level low-level features images. Then, branches extract more complex detailed from outputted branch. They are designed capture details small traces DR images, modified zoom-in/zoom-out attention layers. Findings method trained, validated tested Kaggle dataset. regularization trained model evaluated unseen data samples, were self-collected real scenario hospital. achieves promising performance with sensitivity 98.18% under classes scenario. Originality/value new CNN-based architecture (i.e. DeepRoot) introduced concept multi-branch network. assist solving problem unbalanced dataset, especially when there common characteristics across different DR). Different at depths