<![CDATA[Retinal disease that has a major impact on human vision is diabetic retinopathy. Diabetic retinopathy is a disease caused by advanced diabetic mellitus. Early detection of the disease is very importance. An automated system that can recognize retinal blood vessel abnormalities is very useful for providing quick information to prevent further damage to the retina. In this study, we propose an automated system for segmenting the blood vessels in retinal fundus images using semantic segmentation based on pre-trained from VGG transfer learning and using median frequency balancing weights for the cross entropy loss function. The median frequency weights are to balance the importance of blood vessel and background pixels to get more accurate training results. The integration of encoder and decoder networks utilizing VGG transfer learning and semantic segmentation can segment retinal blood vessels with a sensitivity value of 85.48% using the DRIVE and STARE database.]]>
Copyrights © 2025
