Machine Learning Based Automatic Neovascularization Detection on Optic Disc Region

In this paper, the automatic detection of neovascularization in the optic disc region (NVD) for color fundus retinal image is presented. NV is the indicator for the onset of proliferative diabetic retinopathy and it is featured by the presence of new vessels in the retina. The new vessels are fragile and pose a high risk for sudden vision loss. Therefore, the importance of accurate and timely detection of NV cannot be underestimated. We propose an automatic image processing procedure for NVD detection that involves vessel segmentation using multilevel Gabor filtering, feature extraction of vessel morphological features and texture features, and image classification with support vector machine. Forty two features are extracted from each NVD image and feature selection procedure further reduce the optimal feature dimension to 18. The selected features are trained and tested on 424 retinal images, which contains 134 NVD and 290 non-NVD images. We achieved an average accuracy of 95.23%, specificity of 96.30%, sensitivity of 92.90%, and area under curve value of 98.51% on the randomly selected test set.