Research Article: Effect of AI-driven simulation integrated with virtual case-based training on diagnostic and therapeutic reasoning homogenization among junior ophthalmologists: a prospective cohort study
Abstract:
Conventional training models for diabetic retinopathy (DR) inadequately address the dynamic decision-making demands required for diagnostic reasoning standardization among junior ophthalmologists.
To evaluate whether AI-driven simulation integrated with virtual DR case training is associated with greater homogenization of diagnostic and therapeutic reasoning compared with conventional case-based instruction.
This single-center prospective cohort study enrolled 100 junior ophthalmologists (1–5 years’ clinical experience) assigned to an AI-integrated cohort ( n =?50) or conventional cohort ( n =?50). Both underwent 12-week, 48-hour curricula. Outcomes were assessed at baseline, post-intervention, and one-month post-intervention using blinded expert evaluation and algorithmic benchmarking.
Post-intervention, participants in the AI-integrated cohort showed significantly superior scheme-decision Kappa values (0.93 vs. 0.52; p <?0.001), key decision-point concordance (95.5% vs. 59.0%; p <?0.001), and differential diagnosis coverage (97.0% vs. 62.0%; p <?0.001). Diagnostic confidence (4.78 vs. 3.56; p <?0.001), complex-case diagnostic accuracy (94.0% vs. 60.0%; p <?0.001), and prognostic concordance with an independently validated reference model (RMSE: 0.31 vs. 1.52; p <?0.001) were markedly superior. Acute-complication response time was halved (11.4 vs. 21.3?min; p <?0.001), and high-risk-factor discrimination reached the high-discrimination classification (AUC: 0.95 vs. 0.72; p <?0.001). Individualized treatment-plan adaptation (90.4?±?6.2 vs. 64.8?±?7.4 on a 0–100 scale; p <?0.001) and comorbidity-association recognition (95.0% vs. 63.0%; p <?0.001) confirmed comprehensive practice competency gains. All advantages persisted at one-month follow-up (all p <?0.001), with cross-center consultation Kappa of 0.89 versus 0.49 and follow-up management accuracy of 92.0% versus 58.0%.
AI-driven simulation integrated with virtual DR case training was associated with significantly enhanced reasoning homogenization, clinical confidence, and practical competency among junior ophthalmologists, suggesting that this integrated approach may offer a scalable, guideline-anchored model for standardized ophthalmic specialty training.
Introduction:
Conventional training models for diabetic retinopathy (DR) inadequately address the dynamic decision-making demands required for diagnostic reasoning standardization among junior ophthalmologists.
Read more