Affiliations: Department of Electrical Engineering, Birjand Branch, Islamic Azad University, Birjand, Iran | Department of Medical Radiation Engineering, College of Engineering, Borujerd Branch, Islamic Azad University, Borujerd, Iran | Department of Electrical Engineering, College of Engineering, Borujerd Branch, Islamic Azad University, Borujerd, Iran | Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran
Note: [] Corresponding author. Hadi Chahkandi Nejad, Department of Electrical Engineering, Birjand Branch, Islamic Azad University, Birjand, Iran. Tel.: +98 9155623083; E-mail: [email protected]
Abstract: Diabetic retinopathy is one of the most important causes of visual impairment. In this paper, a supervised automatic lesion detection in digital retina images for diagnosis and screening purposes. The aim of this study is to present a supervised approach for exudate detection in fundus images and also to analyze the method to find the optimum structure. Cellular automata model is used as the base for this task. To improve the adaptability and efficiency of the cellular automata, the rules are updating by a learning process to produce the cellular learning automata. Then, the algorithm is transferred to fuzzy domain for the task of digital retina image analysis. Automaton is created with simple and extended Moore neighborhood. Rule selection and rule updating are performed automatically and the score and penalty assignments are applied to the cells toward a segmentation process. To evaluate the proposed method, statistical parameters of sensitivity, specificity and accuracy are used. A comprehensive experiment is then executed comprising two main phases. First all structural parameters of the automaton are optimized in an investigation study and then a comparison is made between the proposed method with six other well-known methods to verify the results. In the best structure the statistical parameters of sensitivity, specificity and accuracy are computed as 96.3%, 98.7% and 96.1% for STARE retina image dataset.
