Artificial intelligence for improving cardiovascular surgery training: A multiyear national exam analysis

Le résumé fourni par la source

Background: Cardiovascular surgery residency training requires the integration of extensive theoretical knowledge with advanced visuospatial reasoning and technical skills. This study analyzed data from the National Resident Assessment Examinations conducted by the Turkish Society of Cardiovascular Surgery (TSCVS) Board to identify specific training domains in which residents experience the greatest challenges. Furthermore, it explores the potential of artificial intelligence (AI) to support theoretical knowledge acquisition, thereby enabling residents to devote more time to developing visuospatial reasoning and operative decision-making skills. Methods: In this retrospective observational study, 300 multiple-choice questions from six National Resident Assessment Examinations conducted by the TSCVS Board between 2023 and 2025 were analyzed. Residents were categorized into Junior (13-30 months) and Senior (31-48 months) cohorts according to residency duration. The same questions were answered by the ChatGPT 5.2 model using a zero-shot approach, and AI and resident performances were statistically compared. Questions were also categorized according to topic and cognitive domain, and visual questions were further subclassified. Results: Residents achieved a mean examination score of 58.1±11.8, whereas the AI model significantly outperformed the residents, achieving a score of 81.3±6.4 (p<0.001). Junior and Senior residents demonstrated relatively similar mean scores (60.6 vs. 55.6). AI showed high and consistent performance on text-based and guideline-oriented questions; however, its performance declined on examinations requiring visuospatial reasoning and angiographic interpretation. This trend suggests that visually oriented clinical reasoning remains a challenging domain for both human trainees and AI models. AI performance was highest on clinical management and guideline-based questions, whereas lower accuracy was observed on imaging-based and inferencedriven questions. Conclusion: These findings suggest that AI may serve as a valuable educational adjunct for theoretical knowledge acquisition, potentially allowing greater emphasis on simulation-based visuospatial reasoning and operative decision-making skills during residency training.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.