2020/04/18 by Meysam Tavakoli, Tavakoli, Meysam, Faraz Kalantari +3
Medicine · #FOS: Electrical engineering #FOS: Physical sciences #Glaucoma and retinal disorders #Image and Video Processing (eess.IV) #Medical Physics (physics.med-ph) #Retinal Diseases and Treatments #Retinal Imaging and Analysis #Retinal and Optic Conditions #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.11696
openalex publication_date 2020/04/18 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Computer methods and image processing provide medical doctors assistance at\nany time and relieve their workload, especially for iterative processes like\nidentifying objects of interest such as lesions and anatomical structures from\nthe image. Vascular detection is considered to be a crucial step in some\nretinal image analysis algorithms to find other retinal landmarks and lesions,\nand their corresponding diameters, to use as a length reference to measure\nobjects in the retina. The objective of this study is to compare the effect of\ntwo preprocessing methods on retinal vessel segmentation methods,\nLaplacian-of-Gaussian edge detector (using second-order spatial\ndifferentiation), Canny edge detector (estimating the gradient intensity), and\nMatched filter edge detector either in the normal fundus images or in the\npresence of retinal lesions like diabetic retinopathy. From the accuracy\nviewpoint, compared to manual segmentation performed by ophthalmologists for\nretinal images belonging to a test set of 120 images, by using first\npreprocessing method, Illumination equalization, and contrast enhancement, the\naccuracy of Canny, Laplacian-of-Gaussian, and Match filter vessel segmentation\nwas more than 85% for all databases (MUMS-DB, DRIVE, MESSIDOR). The performance\nof the segmentation methods using top-hat preprocessing (the second method) was\nmore than 80%. And lastly, using matched filter had maximum accuracy for the\nvessel segmentation for all preprocessing steps for all databases.\n