Retinal Blood Vessel Segmentation through Morphology Cascaded Features and Supervised Learning

RETINAL BLOOD VESSEL SEGMENTATION THROUGH SUPERVISED LEARNING

Abstract

Retinal blood vessels are the most important attributes in the automatic diagnosis of Diabetic Retinopathy (DR). Since the advanced stages of DR are diagnosed through blood vessels, their segmentation followed by clear analysis is required. Such process can be accompanied through the classification. Segmentation of minor and thin vessels is a challenge because they are analogous to background pixels in fundus image. To solve this issue, this paper proposes a three-stage retinal vessel segmentation mechanism from fundus images. In the first stage, the fundus image is pre-processed for enhancement and then major blood vessels are processed, after extracting them through filtering and morphological transformation. Each pixel of the resulting image is represented by a set of composite features and then processed for pixel level classification. A total of five different features are used to signify each pixel and then for classification Support Vector Machine (SVM) algorithm is used. In the final and post-processing stage, the outputs of first two stages are fused to get the complete retinal vessel structure. Using DRIVE dataset, the proposed method’s experimental validation proves the effectiveness of segmentation accuracy and computational time. The average improvement in the Accuracy, Specificity and Sensitivity is observed as 2.3645%, 1.3365% and 5.2314% respectively from past recent vessel segmentation methods.