Inadequate blood supply in retinal vessels is a major cause of various visual impairments, including partial blindness. Accurate depiction of these blood vessels in Retinal Fundus Images (RFIs) is crucial for tasks like automated screening, computer-assisted laser surgery, and detecting conditions such as Diabetic Retinopathy (DR) and Age-related Macular Degeneration (AMD). Advances in Deep Learning (DL) have enhanced applications in retinal vessel segmentation. This article introduces the Long Short-Term Memory-Driven Blood Vessel Segmentation for Early Detection of Diabetic Retinopathy (LSTMBVS-EDDR) model. This model integrates blood vessel segmentation with DL-based classification to diagnose DR from RFIs. It uses Median Filter (MF) to remove image noise, followed by effective vessel segmentation using a modified SegNet model. The VGG16 network generates feature vectors, with hyperparameters optimized by the Grey Wolf Optimizer (GWO). The Long Short-Term Memory (LSTM) model is then used for DR detection and classification. Evaluations on the MESSIDOR dataset show its superior performance compared to recent techniques.