Intelligent identification and classification of diabetic retinopathy using fuzzy inference system

• Published in: Computer methods in biomechanics and biomedical engineering. 2023-11, Vol.11 (6), p.2386-2399
• Main Authors: Medhi, Jyoti Prakash, Sandeep, R., Datta, Pranami, Nizami, Tousif Khan
• Format: Article
• Language: English
• Subjects: Diabetic retinopathy; fundus image; haemorrhages; Mamdani fuzzy inference system; microaneurysms
• ISSN: 2168-1163; 2168-1171
• DOI: 10.1080/21681163.2023.2235014

Persistent diabetes results in diabetic retinopathy (DR), affecting the retinal blood vessels (BVs), causing lesions. Rapid identification and treatment are crucial for preventing vision loss. Low ophthalmologist to patient's ratio results automating the DR detection a dire need. Therefore, a feature extraction method is proposed using a Mamdani fuzzy inference system (FIS) classifier for efficient identification. Mathematical morphology, region growth, and 12-region search computation have been used to mask the BVs and macula. The masked green plane image was subjected to Nick's thresholding to locate the dark lesions, from which statistical features were extracted and employed in the Mamdani FIS to classify the DR. On evaluating a total of 909 images from the MESSIDOR database shows, average sensitivity, specificity, area under the curve receiver operating characteristics, and accuracy of 99.7%, 99.8%, 99.4%, and 99.6%, respectively. The algorithm performs well in real-time images from two local hospitals. The proposed technique provides a powerful yet flexible tool for improving the diagnosis and treatment of this condition that threatens vision, as it combines the strengths of fuzzy logic, clinical knowledge, and adaptive learning to provide precise, timely, non-invasive, and economical solutions.