Comparison of Grok and ChatGPT in Temporomandibular Joint Magnetic Resonance Images Interpretation: Sectional Study

Abstract

Objective: With the rapid development of artificial intelligence, large language models (LLMs) have emerged as powerful tools capable of processing and interpreting complex medical data. These models can assist in the interpretation of imaging data, especially in conditions that require detailed anatomical analysis such as temporomandibular joint disorders. This study evaluated and compared the performance of 2 LLMs, Chat Generative Pre-trained Transformer-4 Omni (ChatGPT-4o) and Grok, in diagnosing temporomandibular joint disc displacement using magnetic resonance imaging (MRI). Material and Methods: A total of 129 sagittal MRI, including T1- and T2- weighted sequences, were retrospectively analyzed. The images were annotated to identify the disc and mandibular condyle, with diagnoses confirmed by oral and maxillofacial radiology experts. Both models were tasked with identifying anatomical structures and assessing the disc-condyle relationship. Results: Among the analyzed images, 65 showed disc displacement and 64 did not. ChatGPT-4o achieved an overall diagnostic accuracy of 67.4%, with a perfect sensitivity of 100% but lower specificity and precision. In contrast, Grok demonstrated an accuracy of 49.7% (p<0.005), but outperformed ChatGPT-4o in specificity (76.9%), precision (61.5%), and F1-score (58.1%). While ChatGPT-4o showed superior performance in identifying all pathological cases, Grok exhibited greater balance in reducing false positives. Conclusion: This study highlights the potential of LLMs as supplementary tools in oral and maxillofacial radiology while emphasizing the need for further advancements to improve their diagnostic capabilities.

Ähnliche Arbeiten