Computer assisted vessel segmentation from retinal images

Systematic and eye diseases cause morphologic variations in the form of growing, narrowing and branching in retinal blood vessels. Screening of retinal blood vessels with imaging devices plays an important role in identification and being followed of these diseases. Segmentation of right vessel is hard for medical imaging applications. Computer aided algorithm is needed to analyze progression of eye diseases. The method that is proposed in this thesis, provides preprocessing and data incresing with deep learning model. Preprocessing was used for solving irregular clarification problem and forming a contrast between background and retinal blood vessels. Convolutional neural network (CNN) was designed and trained later on for determination of retinal blood vessels. Data augmentation procedure was applied for improving training performance. The proposed model of convolution neural network is trained and tested in the DRIVE database, which is commonly used in retinal blood vessel segmentation and is publicly available for studies in this area. According to the results obtained in this thesis, the proposed system accurately identified vessels with a sensitivity of 77.78% and an accuracy of 95.27%.