Artificial Intelligence and Deep Learning in Ophthalmology Education: A Scoping Review on Clinical Reasoning and Competence.

Read full text Summarise with VIRAT
Authors: Natalia Castaño Chaves
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Ophthalmology is a medical specialty based on visual skills, where diagnosis and therapeutic planning depend on expert interpretation of multiple image modalities combined with clinical and functional data such as visual data (1,2). The core competency of medical students or residents is the coherent clinical reasoning by identifying subtle alterations in the retina, optic nerve or anterior segment, which is difficult to accomplish, largely because education continues to rely on opportunistic patient exposure, case variability, and the limited time available for specialists to supervise and provide feedback on performance (3,4,5). Moreover, population aging and the increasing incidence of vision-threatening diseases (e.g., diabetic retinopathy, age-related macular degeneration (ARMD) and glaucoma) are generating an increasing demand for ophthalmologic care and straining traditional training systems, which are already considered insufficient, particularly in resource-limited regions (3,1,6). In this context, the emergence of artificial intelligence (AI), particularly deep learning (DL) for the automated analysis of medical images is one of the most significant technological transformations in contemporary medicine, and ophthalmology is positioned among the specialties best suited to benefit from these advances (2,7,8). To date, most applications of AI and DL in ophthalmology have been oriented towards automated clinical diagnosis and risk stratification, with particular emphasis on diabetic retinopathy screening, detection of ARMD, and identification of glaucomatous optic neuropathy from fundus photographs and OCT (1,2,9). Currently, emerging studies explore these AI capabilities in educational contexts, such as AI-assisted tutoring platforms, retinopathy reading and labelling systems, and algorithms for the identification of multiple fundus diseases or myopia, which improve performance and understanding compared with traditional teaching methods, enable residents and students to train in lesion grading with immediate feedback and personalize residency training experiences according to case history (11,12,13, 14). Furthermore, virtual reality and augmented reality simulators combined with intelligent tutoring systems have shown objective improvements in microsurgical skills, whereas interactive machine learning platforms, such as fundus image interpretation systems, allow trainees to compare their markings against algorithm outputs and practice with substantial case variability (5,15,16). More recently, synthetic images have been generated through generative models of different pathologies, as well as the use of large language models (LLMs) to create clinical cases and examination questions, further expanding the repertoire of potentially available didactic resources (17,18,19,20). Against this backdrop, the development of clinical reasoning and diagnostic criteria has become a critical competency for these physicians. Clinical reasoning may be understood as the process by which the clinician integrates the clinical history, physical examination and complementary tests to generate, contrast and refine diagnostic hypotheses and therapeutic decisions (21,22). Recent literature on ophthalmology education underscores that one of the essential training objectives is to cultivate accurate and adaptive diagnostic reasoning and describes how AI-based diagnostic decision support systems and simulation platforms are beginning to be utilized as “virtual mentors” that provide visual explanations, saliency maps, step-by-step feedback, and exposure to diverse virtual cases, in order to help learners internalize expert decision-making patterns (3,21). Concurrently, the emergence of LLMs has renewed interest in modelling clinical thinking (22,23). However, this sparks a critical pedagogical debate: while AI can serve as a scaffold for reasoning, concerns persist regarding algorithmic opacity (“black box”), bias, and the risk of fostering technological dependence or “deskilling” (3,1,24,25,26). There has been clear evidence of the impact of AI in ophthalmic education (27,28); however, most studies prioritize algorithmic performance metrics or exam passages over actual pedagogical outcomes, failing to address how these tools modify learners’ reasoning processes. This highlights a critical knowledge gap that justifies this review; the existing literature focuses predominantly on the technical performance of algorithms, leaving their real pedagogical impact largely unexplored. Consequently, a scoping review is needed to move beyond engineering metrics and systematically map the existing evidence regarding the implementation strategies of AI and DL aimed at enhancing medical education within the field of ophthalmology.

Fields of Study & Subjects

Ophthalmology and Visual Health Research Artificial Intelligence in Healthcare and Education Retinal Imaging and Analysis

Publication details

Publication date
1 January 2026
Language
EN
Document type
Preprint