A Comparative Analysis of the Clinical Anatomy Case Design Performance of Generative AI Models in Medical Education
4th International Congress on Educational Sciences and Effective Practices, Eskişehir, Türkiye, 14 - 18 Ekim 2026, ss.368-369, (Özet Bildiri)
- Yayın Türü: Bildiri / Özet Bildiri
- Basıldığı Şehir: Eskişehir
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.368-369
- Eskişehir Osmangazi Üniversitesi Adresli: Evet
Özet
The integration of clinical anatomy into medical education forms the foundation of problem-based learning (PBL) processes. However, developing clinical case scenarios that meet standards and fully reflect anatomical terminology requires significant academic effort. The aim of this study is to comparatively evaluate the performance of two current large language models (GPT-4o and Claude 3.5 Sonnet) in generating structured clinical scenarios for anatomy education. This study is an original methodological analysis that evaluates open-source artificial intelligence outputs without involving human or animal participants. In the study, a standard prompt template was developed regarding the Anatomy of the Nervous System, which is part of the second-year medical school anatomy curriculum. Both models were given the same set of prompts to generate a total of 10 original clinical case scenarios and associated sets of anatomical questions. The resulting outputs were objectively scored using a 5-point Likert scale based on the parameters of “Anatomical Terminology,” “Clinical Correlation,” “Pedagogical Guidance,” and “Absence of Hallucinations.” Data analysis was performed using the Mann-Whitney U test. Based on the analyses conducted, no statistically significant difference was found between the models in the parameters of Anatomical Terminology (GPT: 3.60, Claude: 3.40; p=0.631) and Absence of Hallucinations (GPT: 3.80, Claude: 3.00; p=0.083). However, in the Clinical Correlation (GPT: 3.80, Claude: 4.80; p=0.021) and Pedagogical Guidance (GPT: 3.80, Claude: 4.80; p=0.021), the Claude 3.5 Sonnet model was found to perform significantly better than GPT-4o. Furthermore, it was found that the Claude 3.5 Sonnet model provided significantly richer and more detailed content in terms of the Total Number of Words Used per Case (GPT: 892 words, Claude: 1,261 words; p=0.016). This study demonstrates that generative AI models have high potential in the process of preparing clinical anatomy educational materials. While the Claude 3.5 Sonnet model stands out in terms of establishing clinical connections and providing pedagogical depth, objective data has proven that both models must undergo expert review prior to their direct integration into the educational curriculum. Keywords: Clinical Anatomy, Artificial Intelligence Performance Analysis, Medical Education