A novel DAG network based on multi-feature fusion of fundus images for multi-classification of diabetic retinopathy

Diabetic retinopathy is one of the most serious causes of blindness, clinically, doctors' judgment of lesion grade is time-consuming and laborious, which may lead to early misdiagnosis. Here, the features of diabetic retinopathy are of great significance to the effective diagnosis and prognosis of ophthalmologists. Accordingly, this paper proposes a novel directed acyclic graph (DAG) network for the multi-classification of diabetic retinopathy based on multi-feature fusion of fundus images. Firstly, according to the prior knowledge of doctors, different algorithms are used to extract three features of microaneurysms, neovascularization, and cotton wool spots from diabetic retinas of different grades. Then, these features are fed into the proposed novel DAG network. Besides, diabetic retinas of different grades are learned by combining the multi-feature fusion mechanism. Finally, the optimized classification model is applied to the clinical multi-classification of diabetic retina. IDRiD dataset and clinical dataset from Dalian NO.3 People's Hospital are used to evaluate the performance of the proposed method, and the accuracy can reach 98.5% and 98.6%, respectively. This method can reduce the possibility of early misdiagnosis of diabetic retinopathy, help doctors accurately identify the grade, and effectively prevent further impairment of visual function in patients.