Retinal multi-disease classification using the varices feature-based dual-channel network

Fundus disease is the main cause of visual defect in the cases of non-congenital visual disability, where diabetic retinopathy, ischemic optic neuropathy, optic neuritis, and glaucoma are the most common diseases. Early detection and treatment are the key to control fundus lesions. At present, manual diagnosis may lead to the problem of wasting time and misdiagnosis. On this basis, this paper proposes a dual-channel network for multi-disease diagnosis based on retinal varices features and presents a complete fundus retinal image-assisted diagnosis solution. Firstly, on the advice of ophthalmologists, the retinal varices features of various diseases are extracted. Then, combined with the varices attention mechanism, a dual-channel network retinal multi-disease classification model (VAM-DCN) is constructed. Finally, the retinal varices features are put into a dual channel for network learning and training. The proposed method is verified on the clinical data (normal retina, diabetic retinopathy, ischemic optic neuropathy, optic neuritis, and glaucoma) of Dalian NO.3 People's Hospital, and the precision, recall, F1-score, and accuracy can reach 99.44%, 99.39%, 99.41%, and 99.4%, respectively. The proposed method can help ophthalmologist realize the multi-disease classification of fundus retinal images, reduce the possibility of misdiagnosis and missed diagnosis, which has certain clinical medical value.