Recent progress in generative AI, especially ChatGPT, is being evaluated for its role in diagnostic imaging. A study led by Dr. Daisuke Horiuchi and Associate Professor Daiju Ueda from Osaka Metropolitan University’s Graduate School of Medicine compared ChatGPT’s diagnostic performance with that of radiologists.

The study, published in European Radiology, analyzed 106 musculoskeletal radiology cases, which included patient histories, images, and imaging findings. The AI models GPT-4 and GPT-4 with vision (GPT-4V) were used to generate diagnoses, which were then compared to those provided by a radiology resident and a board-certified radiologist.

The results showed that GPT-4 outperformed GPT-4V and matched the diagnostic accuracy of the radiology resident, but still fell short compared to the board-certified radiologist’s accuracy.

Dr. Horiuchi remarked that while ChatGPT holds promise for diagnostic imaging, it does not yet reach the diagnostic precision of experienced radiologists. He emphasized the need for additional research to fully evaluate the AI’s capabilities before it can be more broadly integrated into clinical practice.

The study highlights the emerging role of generative AI in medical diagnostics, suggesting it could become a useful supplementary tool in the future, pending further validation of its performance.