Multimodal AI-assisted diagnosis and management of myocardial infarction

Background Acute coronary syndrome (ACS) remains a leading cause of global mortality, with guideline-adherent management critically influencing outcomes. Large language models show promise in clinical decision support, yet their cardiology applicability remains underexplored. Methods We retrospectively analyzed 478 myocardial infarction (MI) patients (January 2015–January 2025) at Etimesgut Şehit Sait Ertürk State Hospital, Ankara, Turkey. Three approaches were compared: a multimodal AI system integrating LLaVA-Med, Med-PaLM 2, and BioGPT; general-purpose ChatGPT-5.0; and actual clinical practice. Primary outcomes included diagnostic accuracy, ESC 2023 guideline adherence, MACE, and in-hospital mortality. Results Multimodal AI achieved 95.2% diagnostic accuracy versus 85.0% for ChatGPT-5.0 ( p < 0.001). STEMI detection sensitivity was 98.0% versus 88.1% (p < 0.001). ESC 2023 guideline adherence reached 96.4% for multimodal AI, compared to 85.2% for ChatGPT and 88.9% for clinical practice (both p < 0.001). Correct primary PCI timing was recommended in 100% of STEMI cases versus 92.4% in clinical practice ( p = 0.003). Non-guideline-adherent care was associated with higher in-hospital mortality (20.0% vs. 3.1%; p < 0.001). High-risk case identification reached 93.0% with multimodal AI versus 72.1% with ChatGPT ( p = 0.012). Conclusions Multimodal AI systems integrating specialized medical models demonstrate superior MI diagnostic accuracy and guideline adherence, potentially complementing physician expertise. Human oversight and prospective validation remain essential.