Diabetic Retinopathy Detection and Grading AI for Mobile and Hand-held Devices: A Readiness Survey

Diabetic Retinopathy (DR) is an eye complication arising from diabetic mellitus, leading to complete blindness as it progresses over time. The progression of diabetic retinopathy can be effectively treated when detected in its early stages. With the advancements in artificial intelligence, deep learning has opened the door for more research to be featured in this area. This interest has been mostly motivated by the feature extraction efficacy witnessed in deep learning architectures. Concerning the complexity of this task, stemming from intricate features, image quality, and dataset distribution, more challenges have emerged, necessitating innovative solutions to aid in a more robust automation of diabetic retinopathy detection and grading. This paper reviews DR detection and grading methods and evaluates their readiness for hand-held and mobile devices for increased accessibility. Our work focuses on four main methodologies proposed in recent years for DR detection, including ensemble fusion, cascaded methods, feature segmentation, and hyperparameter tuning, to identify these methodologies' strengths and highlight areas for improvement. The paper also addresses the challenges associated with these methodologies, offering insights for a deep understanding of these constraints and suggesting potential ways to address them in future research.