Robust blood vessel detection with image enhancement using relative intensity order transformation and deep learning

Abstract

Background:An eye-related diabetic microvascular condition is called diabetic retinopathy (DR). Injury to the retinal blood vessels, one of the main causes of blindness worldwide, is a severe public health problem. The vessel growth within the retina, and the likelihood of change in the estimation of retinal veins are important for understanding the stage of diabetic retinopathy. Segmenting retinal blood vessels is necessary to see the changes. Purpose: To develop a framework to improve the quality of the segmentation findings over diabetic retinal datasets. Method:Deep learning-based approaches now show outstanding performance on blood vessel segmentation. However, the majority of them concentrate on designing powerful deep learning architectures rather than capturing the underlying curvilinear structure feature (e.g., the curvilinear structure is darker than the background). In this paper, a robust blood vessel segmentation technique is proposed that captures the characteristics of the blood vessel using relative intensity order transformations(RIOT) and statistical edge-based features. The RIOT extract the thick vessels considering 16 pixels on the horizontal and vertical directions, and the statistical edge features are used to extract the thin blood vessels. The proposed method balances the extraction of thin and thick blood vessels providing robust blood vessel segmentation insensitive to contrast variance in images. Results:The proposed method is validated on the DRIVE, CHASEDB1 and STARE datasets. The proposed model achieved a segmentation accuracy of 96% on the DRIVE dataset, 99% on CHASEDB1 dataset and 87% on STARE dataset. Conclusions:The paper has presented a framework that is contrast invariant in extracting the thick blood vessels using RIOT and thin vessels using statistical edge features and proposed framework has outperformed the existing models.