An Integrated Vessel Segmentation and Machine Learning approach for abnormal vasculature detection in Retinal Images

Retinal image analysis is gaining more attention in recent years due to its non-invasive nature. The information encompassed in vascular structure helps in the detection and diagnosis of various ocular and systemic diseases. In this work, 34-dimensional feature vectors extracted from retinal images are used to segment the blood vessels by implementing an unsupervised K-means clustering algorithm and a U-net architecture. Nine attributes including spatial information are extracted for identifying the abnormalities from the segmented images. The parameters are examined using machine learning algorithms to predict the abnormality. The performance of four classification algorithms, namely, Decision Tree, Support Vector Machine, K - Nearest Neighbor and Ensemble learning approach are analyzed. Digital Retinal Image for Extraction (DRIVE) and Structure Analysis of the Retina (STARE) databases are used in this work. Out of the four methods, Support Vector Machine achieves the highest accuracy of 96.81 % for both datasets in detecting the abnormalities.