Automated Diagnosis of Retinal Neovascularization Pathologies from Color Retinal Fundus Images

• Published in: Advances in Computer Graphics pp. 451 - 462
• Main Authors: Boukadida, Rahma, Elloumi, Yaroub, Kachouri, Rostom, Abdallah, Asma Ben, Bedoui, Mohamed Hedi
• Format: Book Chapter
• Language: English
• Published: Cham Springer Nature Switzerland
• Series: Lecture Notes in Computer Science
• Subjects: Deep Learning; Proliferative Diabetic Retinopathy; Retinal neovascularization; Transfer Learning; Wet Age-Related Macular Degeneration
• ISBN: 3031234723; 9783031234729
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-23473-6_35

The retinal Neo-Vascularization (NV) is the abnormal growth of new blood vessels in the retina, which leads to a severe reduction on visual acuity and blindness. It is a main biomarker to screening several diseases, where the Proliferative Diabetic Retinopathy (PDR) and Wet Age-related Macular Degeneration (WAMD) are the most common ones. The NV severity requires a fast screening to avoid severe degradation. However, it is labor intensive and time-consuming for the ophthalmologists. In this paper, we suggest an automated screening method that automatically detects NV from fundus photography and classify it as PDR, WAMD and Healthy. For this purpose, the image is preprocessed and then provided a transfer learned model of the VGG16 neural network. The method was evaluated using a dataset containing 395 fundus photographs of retinal images where an accuracy of 98.30%, a sensitivity of 98.66%, a specificity of 98.33% were achieved. In addition, the areas under curve in terms of classes were between 98% and 100%.