Blood Vessels Segmentation in Retinal Fundus Image using Hybrid Method of Frangi Filter, Otsu Thresholding and Morphology

Abstract

Diagnosis of computer-based retinopathic hypertension is done by analyzing of retinal images. The analysis is carried out through various stages, one of which is blood vessel segmentation in retinal images. Vascular segmentation of the retina is a complex problem. This is caused by non-uniform lighting, contrast variations and the presence of abnormalities due to disease. This makes segmentation not successful if it only relies on one method. The aims of this study to segment blood vessels in retinal images. The method used is divided into three stages, namely preprocessing, segmentation and testing. The first stage, preprocessing, is to improve image quality with the CLAHE method and the median filter on the green channel image. The second stage, segmenting using a number of methods, namely, frangi filter, 2D-convolution filtering, median filtering, otsu's thresholding, morphology operation, and background subtraction. The last step is testing the system using the DRIVE and STARE dataset. The test results obtained sensitivity 91.187% performance parameters, 86.896% specificity, and area under the curve (AUC) 89.041%. Referring to the performance produced, the proposed model can be used as an alternative for blood vessel segmentation of retinal images.

Contrast enhancement for improved blood vessels retinal segmentation using top-hat transformation and otsu thresholding

“…The threshold in this method refered to the threshold at the maximum variance. The following equations were used in the Otsu Thresholding method [28]: calculate the probability for the background pixel r1(t), the blood vessel pixel r2(t), the variance for the background pixel σ1 2 (t), and the variance for blood vessel pixel σ2 2 (t) using Equations ( 7) -( 10) as followed [28]: ()…”

“…The average background pixels was μ1(t) and blood vessel pixels μ2(t) could be calculated using Equations ( 11) and ( 12) [28].…”

“…S(i) was a probability for the i th pixel defined as S(i)=n(i)/n, where n(i) was number of total grayscale pixels and n was a number of total pixels. After that, it needed to calculate the variance between blood vessel pixels and background pixels using Equation ( 13) [28].…”