Computer analysis of retinal images for vessel anomaly detection

Many important eye diseases, as well as systemic diseases such as diabetes, hypertension and arteriosclerosis manifest themselves in the retinal. An automatic analysis of the vascular system of the retina can assist to monitor the effects of corresponding diseases on the visual system. In this thesis, we have presented the robust method for retinal blood vessel segmentation and an automatic algorithms for analyzing the vessel network with regard to the vessel diameter and tortuosity. A supervised method is presented for segmentation of retinal blood vessels using an ensemble classifier of random forest decision trees. The 17 dimension feature vector is constructed for successfully handling the both normal and abnormal retinal images with different kind of lesions. Forty images of the retina from DRIVE database and twenty images from STARE database are used to evaluate the performance of the method. The results of performance metrics illustrate that the proposed method outperforms most of the other segmentation methods. Moreover, our method needs fewer samples for training than other methods and it is independent to the training set as it offers a better performance in the cross-training section. The automatic method for measuring the retinal vessel width and tortuosity is also presented based on the vessel edges and centerline in order to analysis the vessel network and monitor the presence of different abnormalities in the structure of vessels. This application may assist to quick diagnosis and treatment planning in the clinical procedures.