Comparing teacher-made and AI-assisted teaching videos for thyroid eye disease: an exploratory study
Related Experiment Video
Updated: Nov 6, 2026
Comparing Eye-tracking Data of Children with High-functioning ASD, Comorbid ADHD, and of a Control Watching Social Videos
Published on: December 7, 2018
Shu Liu1, Yue Wang1, Yuqing Chen1
1Department of Ophthalmology, Changzheng Hospital of Naval Medical University, Shanghai, China.
Background:
Thyroid eye disease (TED), together with its acute and vision-threatening complication, dysthyroid optic neuropathy (DON), requires timely diagnosis and intervention. The limited availability of specialists in orbital diseases poses a significant challenge to ophthalmology residency training. In this context, generative artificial intelligence (AI) has created new opportunities for medical education, though its effectiveness in ophthalmology training remains to be explored.
Objective:
This exploratory study aims to evaluate the effectiveness of generative artificial intelligence (AI) in ophthalmology education, focusing on TED domains.
Methods:
The 46 residents (28 general residents and 18 ophthalmology residents) were randomly assigned to either an AI-assisted or a teacher-made 5-min teaching video on TED. All participants then completed three tests through online questionnaire experiments, each consisting of 20-item clinical questions about TED. Test 1 served as a baseline assessment of disease knowledge, Test 2 was conducted 5 min after learning to evaluate short-term learning outcomes, along with a questionnaire survey about AI tools, and Test 3 was administered 1 week later to assess long-term learning outcome.
Results:
Viewing the AI-assisted video significantly improved test scores from baseline to 5 min (mean difference = 13.04; p = 0.004), but the improvement was not sustained at 1 week (mean difference = 7.61; p = 0.16). Completion time also decreased significantly from baseline to both 5 min (p = 0.009; GMR = 0.57) and 1 week (p < 0.001; GMR = 0.45). No significant differences were observed between the AI-assisted and teacher-made video groups across all assessments (all p > 0.05). Additionally, learning experience was positively correlated with learning outcome (r = 0.60; P < 0.001).
Conclusion:
Our findings suggested that the AI-assisted and teacher-made videos for TED education showed no statistically significant difference. However, substantial expert oversight was required, and the AI-assisted workflow did not reduce total production time in this study. Given the limited long-term retention, such AI-assisted videos may be better positioned as supplementary resources combined with human expertise.
