AI and Explainable AI for Automatic Detection and Grading of Diabetic Retinopathy and Related Ophthalmic Diseases: A Review

Abstract

The management of retinal diseases within a national health service is greatly improved by the availability and development of good automatic systems for eye disease identification and grading. Identifying the characteristics of retinal diseases, particularly diabetic retinopathy, is a challenging and lengthy process. Artificial intelligence (AI) and explainable AI (XAI) are reshaping the landscape of automated ophthalmic diagnostics, with diabetic retinopathy (DR) emerging as a central benchmark for such innovations. This review paper offers a thorough and critical examination of AI and XAI methodologies in this domain, with DR as the primary core focus area. Artificial intelligence, and especially deep learning, now facilitates precise and swift identification of retinal lesions and disease classification, while explainable AI enhances interpretability, transparency, and clinical confidence. This review paper synthesizes recent AI and XAI developments for DR and other ophthalmic diseases, covering image pre-processing, deep learning architectures and feature extraction as well. In order to contextualize the discussion, relevant background information on diabetic retinopathy and related retinal diseases is presented, including a detailed overview of the characteristic pathological features and the screening procedures commonly adopted by healthcare services. Furthermore, this article examines the publicly available eye fundus image datasets that serve as valuable resources for research in this domain. In view of the inherent challenges in developing automated screening systems, the paper concludes with a critical evaluation of current limitations and an exploration of prospective directions for future developments on AI and XAI for diabetic retinopathy and other ophthalmic diseases.