AUTOMATED RETINAL BLOOD VESSEL SEGMENTATION VIA U-NET AND VGG-BASED MODELS
Keywords:
Medical Image Segmentation, Precision Diagnostic, Hybrid Segmentation Architecture, Vascular Structure Analysis, U-Net, Visionary Vessel MappingAbstract
Diagnosis of human diseases especially eye disease is a challenging task. Automatedretinal blood vessel segmentation helps detect and treat ophthalmological illnesses like diabetic retinopathy and glaucoma. This study uses a mixed deep learning strategy with U-Net and VGG16 architectures to segment retinal blood vessels precisely. The dataset of 100 retinal pictures with segmentation masks was contrast-boosted and normalized for uniformity. The U-Net model had 87.92% accuracy and 0.4243 loss, whereas the VGG16 model had 87.68% accuracy and 0.4085 loss. The proposed combination model performed well, with 89.75% accuracy, 88% precision, and 89% recall. The hybrid architecture uses U-Net's segmentation and VGG16's deep feature extraction to outperform standalone models in complex vessel structures. This powerful model can improve retinal vascular segmentation, potentially changing clinical diagnostic procedures.
