An application of image processing to the diagnosis of diabetic retinopathy for the automatic detection of microaneurysms and hemorrhages in fundus images of the human retina
This study focused on developing a new algorithm for segmenting lesions in retinal images, in order to prepare for lesion feature extraction, in the automatic diagnosis of Diabetic Retinopathy (DR) using computer vision. Tests conducted to evaluate algorithm performance included a sensitivity-and-specificity test, a ground-truth-based test and a subjective fidelity-criteria test. Sensitivity and specificity were 90% and 100%, respectively. Target lesions fell within ground-truth-labeled regions. On a scoring scale of 1 to 5, the subjective fidelity-criteria mean, standard deviation and variance were 3.829, 0.6128 and 0.375, respectively. Tests showed the algorithm has strong potential in helping to detect microaneurysms and hemorrhages, as a means of automatically diagnosing DR’s early symptoms and treating it in a timely manner.
Key Words: Blindness, diabetic retinopathy, fundus, retinal blood vessels, microaneurysms, hemorrhages, image processing, image segmentation