A New Technique using Deep Learning to Segment Retinal Blood Vessels

Diabetic retinopathy which causes severe vision loss which results in lesions on the eye. This problem will lead to blindness. No permanent treatment for DR we can save eyesight if we detect in earlier stage. Comparing with latest computer-based diagnosis methods, the traditional diagnosis of DR with the help of eye specialists and the process will take more time and which cause human error. In existing research, the location of the optic disc is not adequately highlighted, adding attention increases the computational complexity, requires a large number of annotated images lead to overfitting, retinal features are still limited. Segmenting the vessels or identifying illnesses cannot have performed. Noise in the input image leads to inaccuracies, hemorrhages are complicated and segmentation precision is less accurate. Deep learning and Machine learning are advanced methods are popularly using in this area Nowadays. They are especially used to identify and classify medical images. In this paper, we proposed a significant method is U-net network for segmenting retinal blood vessels. A novel different method of U-Net that uses 18 Convolutional layers and which used to improve the performance. The developed model is tested on IDRID and DRIVE datasets, which is able to detect vessel acc of 0.9893, PRE of 0.7469 and REC of 0.6587.