Computer-assisted diagnosis (CAD) system for Diabetic Retinopathy screening using color fundus images using Deep learning

Abstract

Diabetes is a serious medical condition and regular screening for diabetes is of great importance as treatment options are most effective in the early stages of diabetes. Digital imaging of retina is considered as a low-cost method for screening and could be used in conjunction with computer-based image processing techniques to automatically detect early signs of diabetes utilizing diabetes-related pathologies visible in retinal fundus images. This research proposes a novel computer-assisted diagnosis (CAD) system for assisting with the screening of the population as up to 50% of the affected population are not aware of having diabetes. Moreover, these screenings are often carried out by an optometrist who receives some training with the patients being referred to an ophthalmologist if they show symptoms. Having a computer-assisted diagnosis system assisting the optometrist during the screening can greatly increase the detection rate for patients with diabetes by providing a second opinion and highlighting any suspicious pathologies. For achieving the highest detection rate possible, a hybrid machine learning approach is proposed in this research by combining Deep Learning with the AdaBoost classifier. The proposed computer-assisted diagnosis system starts with the segmentation of the blood vessels. Then, microaneurysms and exudates are segmentation from the image. Statistical and regional features are then extracted utilizing first, second, and higher-order image features. A Deep Learning framework will be utilized for extracting additional statistical image descriptors as a Deep Learning has superior contextual analysis capabilities compared to other machine learning techniques. Finally, the most informative features are selected by a minimal-redundancy maximal-relevance feature selection approach with an AdaBoost classifier analyzing all the features and informing the operator regarding the patient’s condition. Ethereum Swarm blockchain-based decentralized cloud file storage provides the proposed CAD users with a secure storage olution to access the patient information and related images. The sensitivity, specificity, and accuracy of the classification will be measured under clinical conditions. Healthcare, government, and public users would receive the most benefit from this project.