Comparative Evaluation of AI Models Such as ChatGPT 3.5, ChatGPT 4.0, and Google Gemini in Neuroradiology Diagnostics

Aims and objective: Advances in artificial intelligence (AI), particularly in large language models (LLMs) like ChatGPT (versions 3.5 and 4.0) and Google Gemini, are transforming healthcare. This study explores the performance of these AI models in solving diagnostic quizzes from "Neuroradiology: A Core Review" to evaluate their potential as diagnostic tools in radiology. Materials and methods: We assessed the accuracy of ChatGPT 3.5, ChatGPT 4.0, and Google Gemini using 262 multiple-choice questions covering brain, head and neck, spine, and non-interpretive skills. Each AI tool provided answers and explanations, which were compared to textbook answers. The analysis followed the STARD (Standards for Reporting of Diagnostic Accuracy Studies) guidelines, and accuracy was calculated for each AI tool and subgroup. Results: ChatGPT 4.0 achieved the highest overall accuracy at 64.89%, outperforming ChatGPT 3.5 (62.60%) and Google Gemini (55.73%). ChatGPT 4.0 excelled in brain, head, and neck diagnostics, while Google Gemini performed best in head and neck but lagged in other areas. ChatGPT 3.5 showed consistent performance across all subgroups. Conclusion: This study found that advanced AI models, including ChatGPT 4.0 and Google Gemini, vary in diagnostic accuracy, with ChatGPT 4.0 leading at 64.89% overall. While these tools are promising in improving diagnostics and medical education, their effectiveness varies by area, and Google Gemini performs unevenly across different categories. The study underscores the need for ongoing improvements and broader evaluation to address ethical concerns and optimize AI use in patient care.