Comparing physicians with artificial intelligence applications in decision-making for cranial CT scans in patients with mild head trauma

Objective: The study aims to investigate the usability of Chat Generative Pre-Training-4 (ChatGPT-4) in decision-making for cranial CT scans in patients presenting to the emergency department with mild head trauma and to evaluate whether it can contribute to clinicians' decision-making processes. Methods: Our study was designed as a prospective and observational study. A total of 957 patients of varying age groups who presented with mild head trauma to the emergency department of a tertiary university hospital were included. For all patients, the cranial Computed Tomography (CT) scan decisions made by physicians and ChatGPT-4 were recorded. In adult patients, CT scan indications were assessed based on the Canadian CT Head Rules, while in pediatric patients, the PECARN (Pediatric Emergency Care Applied Research Network) rules were utilized. The CT imaging decisions of physicians and ChatGPT-4 were compared using appropriate statistical methodologies to evaluate their concordance and diagnostic accuracy. Results: ChatGPT-4 demonstrated higher concordance with the Canadian CT Head Rules and PECARN guidelines for cranial CT scan decision-making in patients with mild head trauma compared to physicians. This higher concordance was observed across all patient groups: the adult patient group (κ=1.000, p<0.001 vs. κ=0.647, p<0.001), the pediatric patient group aged over 2 years (κ=0.977, p<0.001 vs. κ=0.655, p<0.001), and the pediatric patient group under 2 years of age (κ=0.979, p<0.001 vs. κ=0.781, p<0.001). These differences were found to be statistically significant. Conclusion: Artificial intelligence applications are being utilized for various purposes in the field of healthcare. The results of our study suggest that ChatGPT-4 can be used in decision-making for cranial CT scans in patients with mild head trauma. Keywords: Head Trauma, ChatGPT-4, Canadian CT Head Rules, PECARN