Comparing Different Preprocessing Methods in Automated Segmentation of Retinal Vasculature

Abstract

Computer methods and image processing provide medical doctors assistance at any time and relieve their work load, especially for iterative processes like identifying objects of interest such as lesions and anatomical structures from the image. Vescular detection is considered to be a crucial step in some retinal image analysis algorithms to find other retinal landmarks and lesions, and their corresponding diameters, to use as a length reference to measure objects in the retina. The objective of this study is to compare effect of two preprocessing methods on retinal vessel segmentation methods, Laplacian-of-Gaussian edge detector (using second-order spatial differentiation), Canny edge detector (estimating the gradient intensity), and Matched filter edge detector either in the normal fundus images or in the presence of retinal lesions like diabetic retinopathy. The steps for the segmentation are as following: 1) Smoothing: suppress as much noise as possible, without destroying the true edges, 2) Enhancement: apply a filter to enhance the quality of the edges in the image (sharpening), 3) Detection: determine which edge pixels should be discarded as noise and which should be retained by thresholding the edge strength and edge size, 4) Localization: determine the exact location of an edge by edge thinning or linking. From the accuracy view point, comparing to manual segmentation performed by ophthalmologists for retinal images belonging to a test set of 120 images, by using first preprocessing method, Illumination equalization, and contrast enhancement, the accuracy of Canny, Laplacian-of-Gaussian, and Match filter vessel segmentation was more than 85% for all databases (MUMS-DB, DRIVE, MESSIDOR). The performance of the segmentation methods using top-hat preprocessing (the second method) was more than 80%. And lastly, using matched filter had maximum accuracy for the vessel segmentation for all preprocessing steps for all databases.