BLOOD VESSEL SEGMENTATION IN RETINAL IMAGES USING CONVOLUTIONAL NEURAL NETWORK VV-NET METHOD

The retina is susceptible to various diseases that can be fatal if not treated quickly. Image processing is currently very helpful for doctors to detect retinal diseases faster so that retinal diseases can be treated immediately. The first step in image processing is to improve the quality of retinal images affected by noise, aiming to increase accuracy in the process of segmentation and image extraction. accurate segmentation of retinal blood vessels is the first step in disease detection. The process of segmentation and analysis of retinal blood vessels has an important role in assisting medical professionals in identifying the severity of a disease. Image quality improvement steps in preprocessing use grayscale, median filter (denoising), and clahe. The method used for blood vessel segmentation is CNN VV-Net. Evaluation of the results of applying image quality enhancement and segmentation techniques using the VV-Net method was performed on the DRIVE, STARE, and CHASEDB_1 datasets at both stages, training and testing. The measurement results of blood vessel segmentation using the CNN VV-net method on the DRIVE dataset (accuracy 96.27%, sensitivity 84.38%, precission 75.95%, and jaccard score 66.28%), STARE dataset (accuracy 96.58%, sensitivity 82.78%, precission 76.73%, and jaccard score 65.38%), and CHASEDB_1 dataset (accuracy 97.04%, sensitivity 83.55%, precission 76.72%, and jaccard score 66.40%). From the three datasets used, the CHASEDB_1 dataset obtained better results than the DRIVE and STARE datasets.