Automatic detection of neovascularization on optic disk region with feature extraction and support vector machine.

Neovascularization (NV) is a definitive indicator for the onset of Proliferative Diabetic Retinopathy (PDR). The new vessels are fragile and prone to bleed, leading to high risk of sudden vision loss. Automatic detection of NV is an important task in automatic Diabetic Retinopathy (DR) screening as a consequence of the unmet requirement between the growing number of DR patients and limited number of ophthalmologists. This paper focuses on the computer aided detection of neovascularization in the optic disk region. We propose a novel image processing approach that involves vessel segmentation using multi-level Gabor filtering, feature extraction from vessel related features and texture features, and image classification based on machine learning. 21 features were extracted from each NVD image. The extracted features were trained and tested on 66 retinal images, which contains 16 NVD and 50 normal images, and achieved an sensitivity of 15/16 and specificity of 47/50.