Automated Detection of Diabetic Retinopathy Vascular Lesions (Microaneurysms) in Fluorescein Angiography by Image Processing Methods

اولين همايش تحقيقات چشم پزشكي و علوم بينايي ايران

19 تیر 1389، تهران - ايران

Presentation Type: Speech

Purpose:The main objective of this research is to aid in developing automated screening systems for diabetic retinopathy. Such systems will significantly help ophthalmologists in diagnosing and treating patients.
Methods:A set of selected images consisting of 35 training and 85 test images were used for analysis. Gold standard was defined by manual detection of microaneurysms on selected images by and image reader which is stored in a database named "Mash had University of Medical Science Diabetic Retinopathy Image Database" (MUMS-DB).Firstly, we detected optic nerve head by using radon transform and multi overlapping windows after masking. In preprocessing stages at first we used top-hat transformation and subsequently, we applied averaging filter and subtracted the result of filtering and top-hat. In the main section we detected and localized the vascular lesion related to diabetic retinopathy by dividing the whole preprocessed images to multi overlapping sub images (windows) and radon transformation of these sub-blocks.After retinal vascular tree and optic nerve head are detected and masked, microaneurysms will be detected and numbered by using appropriate thresholding.
Results:In this study the sensitivity of diagnosis for diabetic retinopathy was 94%; with a specificity of 75% and sensitivity of precise microaneurysm localization was 92%.
Conclusion:This project addresses a novel method in detection of retinal landmarks and lesions to diagnose diabetic retinopathy. High sensitivity of this method, encourage the usage of it as a first phase screening for diagnosis of diabetic retinopathy.