DR-Net: Diabetic Retinopathy detection with fusion multi-lesion segmentation and classification

Diabetic Retinopathy (DR) is one of the most common microvascular complications of diabetes mellitus and is a major cause of blurred vision, vision loss, and blindness. Depending on the severity of the disease, DR is divided into non-proliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR). Current research has focused on using Deep Learning (DL) models to classify fundus images based on DR severity. To make the lesions in DR images more visible and to make DR detection easier, this study proposes a two-phase classification model (DR-Net). SR-Net (SE-Block-ResNet) is the first phase of the network in this study, the second phase consists of MT-SNet (Multiple lesions-TransUnet-Segmentation-Net) and SRVGG (SE-Block-RepVGG). The first phase uses ST-Net to classify NPDR images with PDR images, while the second phase first implements segmentation of multiple lesions, followed by classification of the processed NPDR images. The accuracy on the DDR dataset is improved by 2.21% compared to the new study.