Application of a deep learning-based system for eyelid margin signs identification training and testing in dry eye disease education
In brief
AI teaching platform lifts dry-eye exam scores by about three points
In a randomized trial of 40 medical students, those who trained with an interactive deep-learning system scored on average 3.6 points higher on a timed eyelid-margin exam than peers who watched video lectures. The AI tool especially improved identification of meibomian gland orifice plugging, suggesting it could augment traditional ophthalmology education, though larger studies are needed.
- Journal
- Frontiers in medicine (Q1)
- Published
- 31 July 2026
- Study design
- Randomized controlled trial
- Evidence level
- Level 1, High (CEBM 1b)
- Authors
- Ningkai Tang, Zhixin Duan, Yuexin Wang
- PMID
- 42602288
- DOI
- 10.3389/fmed.2026.1896446
Why clinicians should know about it
- Picked for Ophthalmology (top studies of the week, 25 October 2026): RCT shows AI improves eyelid margin identification
Abstract
BACKGROUND: Dry eye disease (DED) is a common ocular surface disease, and identifying abnormal eyelid margin signs is essential for clinical diagnosis. However, traditional teaching methods lack effective feedback and often lead to low learning efficiency. This study aimed to develop an interactive teaching platform based on a deep learning model for eyelid margin signs and to evaluate its effectiveness. METHODS: This randomized controlled trial included 40 medical students from the Peking University Health Science Center. Participants were randomly assigned to a conventional learning group (n = 19) or a platform learning group (n = 21). After baseline training, the platform group practiced using the AI-assisted system with real-time feedback, while the conventional group reviewed video lectures. Two weeks later, all students took a timed final exam evaluating eight core abnormal eyelid margin signs. RESULTS: The platform learning group achieved a significantly higher total score than the conventional learning group (56.95 ± 5.95 vs. 53.37 ± 4.32, p = 0.035). Specifically, the platform group showed a significantly higher accuracy rate in identifying meibomian gland orifice (MGO) plugging (77.6% vs. 67.4%, p = 0.021). Subgroup analysis revealed that the platform significantly improved the accuracy of rounding of posterior lid margin for undergraduate students (82.1% vs. 69.1%, p = 0.016), and the accuracy of MGO plugging for graduate students (85.7% vs. 67.5%, p = 0.009). CONCLUSION: The deep learning-integrated teaching platform may improve students' overall accuracy in identifying abnormal eyelid margin signs, particularly MGO plugging. This platform may serve as a useful supplement to traditional clinical ophthalmology education and may help improve the efficiency of learning eyelid margin assessment.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.