A Moderated Deep Learning Approach for Retinal Blood Vessel Segmentation Hybrid Neural Classification and Optimization Logic

Retinal blood vessel segmentation (RBVS) is a crucial task in ophthalmic and cardiovascular disease diagnosis, particularly in conditions like diabetic retinopathy, glaucoma, and hypertensive retinopathy. This paper presents a deep learning (DL)-based hybrid approach combining Convolutional Neural Networks (CNNs) and transformer models for accurate retinal vessel segmentation. The hybrid architecture integrates CNNs for local semantic feature extraction and transformers for capturing long-range spatial dependencies. Image quality was enhanced by the usage of pre processing techniques, like green channel extraction and histogram equalization. The model was trained and evaluated on three publicly available datasets: DRIVE, STARE, and CHASE_DB1. Hybrid CNN was proposed which attains state of the art performance with an accuracy of 96.05% on the DRIVE dataset, 96.20% on the STARE dataset and 95.85% on the CHASE_DB1 dataset. The model performed with sensitivity values of 78.20%, 79.00%, and 78.00% in the segmentation of the fine and delicate vessels. The results show that the hybrid CNN outperformed the traditional U-Net and other DL approaches in sensitivity, Dice coefficient, and F1 score. These results demonstrate the promise of the hybrid CNN-transformer approach for medical image segmentation, and also for improving early disease detection.