Neurological Diagnosis in the AI Era: A Comparative Assessment of ChatGPT 3.5, Google Gemini, Being AI, and Perplexity AI

Abstract

Background. In neurological diagnostics, where complexity, data volume, and diagnostic urgency present major obstacles, artificial intelligence (AI) systems have the potential to revolutionize the field. Despite widespread use, there is a lack of comparable performance assessments of publicly available AI tools for integrated clinical reasoning in neurology. Methods. This cross-sectional study evaluated five AI platforms (ChatGPT 3.5, Google Gemini, Bing AI, Perplexity AI, DeepSeek) utilizing 15 standardized neurological cases from Case Files: Neurology, Third Edition. Each platform was given identical prompts imitating clinical consultations. Responses were evaluated (maximum 6 points per case; total 90) in three domains: diagnosis, subsequent diagnostic step, and therapeutic/molecular foundation. Nonparametric statistical methods (Kruskal-Wallis, Chi-square) assessed performance disparities. Results. ChatGPT achieved the highest overall score (88/90, 97.8%), followed by DeepSeek (86/90, 95.6%), Perplexity (84/90, 93.3%), Google Gemini (78/90, 86.7%), and Microsoft Copilot (73/90, 81.1%). Therapeutic accuracy was 100% for ChatGPT, DeepSeek, and Gemini, whereas it was 80% for Copilot. Although there were disparities in performance, inferential statistics revealed no significant differences between platforms (Kruskal-Wallis p = .423; Chi-square p = .374). Verbosity showed significant variation: DeepSeek averaged 488 words per response, whereas Copilot and Perplexity averaged 239 to 240 words. Conclusion. Popular AI platforms (ChatGPT, DeepSeek) exhibit significant proficiency in neurological diagnosis and treatment planning, but there is a huge difference in the depth and structure of responses across all of the tools. AI should be used as complementary healthcare assistance, with future integration requiring better explainability and real-world validation.

Ähnliche Arbeiten