Blood Vessel Segmentation in Fundus Images Using Hessian Matrix for Diabetic Retinopathy Detection

Abstract

Diabetic Retinopathy (DR) is a severe eye disease that could lead to sight loss. This disease is caused by damages in the blood vessels of the retina due to prolonged high blood glucose level. DR is characterized by the presence lesions and the formation of abnormal blood vessels in the retina called neovascularization. Early detection of DR is essential to prevent the disease from worsening and avoid early loss of vision in diabetic patients. Identification and segmentation of retinal blood vessels from fundus images are crucial tasks for automatic DR detection. This paper presents a blood vessel segmentation technique using Hessian Matrix. First, the green channel is extracted from the fundus image in the pre-processing stage. A Gaussian filter is then used to smoothen the image. Next, the Hessian Matrix is constructed to calculate the maximum principal curvature of the image's intensity for extracting the blood vessels' structure. The retina's boundary is then removed to reduce false detection. In the post-processing stage, morphological erosion is used to remove noise from the image. Contrast-limited adaptive histogram equalization (CLAHE) is then applied to enhance the resulting image. Finally, Iterative Self-Organizing Data Analysis (ISODATA) thresholding technique is used to binarize the image. Experiments were conducted to evaluate the proposed method's performance using fundus images obtained from DRIVE, HRF, and STARE datasets. The results showed that the technique could provide good accuracy on average up to 0.95.