Summary: A study in Japan found that ChatGPT matches neuroradiologists in brain tumor MRI diagnosis accuracy, highlighting its potential to enhance healthcare diagnostics and ease physician workloads.

Key Takeaways

  • ChatGPT matched neuroradiologists’ accuracy (73%) in diagnosing brain tumors from MRI reports, surpassing general radiologists (68%).
  • ChatGPT’s accuracy was higher (80%) when using neuroradiologist-written reports, indicating its potential is enhanced with specialized input.
  • The study suggests that ChatGPT can support preoperative MRI diagnoses, reduce physician workload, and improve diagnostic accuracy in healthcare.

——————————————————————————————————————————————————————————————

Using real-world Japanese data, large language models like ChatGPT now match neuroradiologists in accuracy, transforming healthcare by enhancing diagnostics and easing the workload on professionals.

To explore this potential, Yasuhito Mitsuyama, a graduate student, and associate professor Daiju Ueda, MD, PhD, led a research team at Osaka Metropolitan University in Japan. They compared GPT-4-based ChatGPT’s diagnostic performance with that of radiologists on 150 preoperative brain tumor MRI reports. Using daily clinical notes written in Japanese, ChatGPT, two neuroradiologists, and three general radiologists provided differential and final diagnoses, which were then compared to the actual tumor diagnoses confirmed after removal.

Brain Tumor Diagnosis Accuracy

The accuracy results showed 73% for ChatGPT, a 72% average for neuroradiologists, and a 68% average for general radiologists. Interestingly, ChatGPT’s accuracy varied depending on the report source: it achieved 80% accuracy with neuroradiologist reports but only 60% when using general radiologist reports, suggesting potential for improvement with specialized input.

“These results suggest that ChatGPT can be useful for preoperative MRI diagnosis of brain tumors,” stated Mitsuyama. “In the future, we intend to study large language models in other diagnostic imaging fields to reduce physician burden, improve diagnostic accuracy, and use AI to support educational environments, ultimately enhancing patient care.”