Use of Expert-Validated AI-Generated Forensic Images in Case-Based Learning and Clinical Reasoning Among Undergraduate Medical Students: A Two-Arm Educational Intervention Study

BackgroundClinical reasoning is an essential competency in undergraduate medical education, while forensic medicine requires accurate visual recognition and interpretation of injury patterns.Case-based learning (CBL) provides a structured approach to developing higher-order reasoning skills.Generative AI offers a potential method for creating customized visual learning resources, although evidence regarding the educational value of AI-generated clinical and forensic images remains limited. ObjectiveThe study aimed to evaluate whether expert-validated AI-generated forensic images used as visual triggers within standardized CBL were associated with differences in clinical reasoning among Phase II and Phase III MBBS students compared with conventional forensic images. MethodsThis two-arm parallel-group non-randomized controlled educational intervention study enrolled and allocated 249 Phase II and Phase III MBBS students to AI-generated image or conventional-image groups.Images were generated using GPT-Image-1 (OpenAI, San Francisco, CA).The primary T0-T1 analysis included 227 students, while 218 students with complete T0-T1-T2 data contributed to the longitudinal analysis.The intervention group participated in standardized CBL sessions using expert-validated AIgenerated forensic images, while the comparison group used conventional forensic images.Case scenarios, learning objectives, questioning sequence, faculty facilitation, session duration, and debriefing were standardized across groups.Clinical reasoning was the prespecified primary outcome.Secondary outcomes included knowledge, engagement, confidence, satisfaction, and short-term retention.The primary T0-T1 analysis included 227 students, while 218 students with complete T0-T1-T2 data contributed to the longitudinal analysis.Between-group comparisons were adjusted for baseline clinical reasoning and academic phase. ResultsBaseline characteristics were comparable between groups.The adjusted post-intervention clinical reasoning score was significantly higher in the AI-generated image group than in the conventional-image group (25.3 ± 3.9 vs 23.1 ± 4.2), with an adjusted mean difference of 2.4 points (95% CI, 1.4-3.4;p < 0.001; Cohen's d = 0.59).The AI-generated image group also demonstrated higher knowledge scores (25.1 ± 2.9 vs 23.7 ± 3.1; adjusted mean difference, 1.3; 95% CI, 0.5-2.1;p = 0.002), engagement (64.6 ± 6.2 vs 62.5 ± 7.2; p < 0.001), confidence (21.4 ± 2.3 vs 19.6 ± 2.8; p < 0.001), and satisfaction (22.8 ± 2.1 vs 20.5 ± 2.8; p < 0.001).At five to six weeks, clinical reasoning scores were 23.2 ± 4.1 in the AI-generated image group and 20.9 ± 4.3 in the conventional-image group, with a significant group-by-time interaction (p < 0.001).Knowledge also demonstrated a significant group-by-time interaction (p = 0.004). ConclusionThis study found that students receiving expert-validated AI-generated forensic images as visual triggers within standardized CBL demonstrated greater improvement in measured clinical reasoning than students receiving conventional forensic images.These findings represent an association within a specific facultysupervised educational setting and do not establish that AI-generated imagery alone caused the observed differences.AI-generated visual resources may be considered as an adjunct to competency-based forensic medicine education when incorporated within a structured pedagogical framework and subjected to expert