ChatGPT outperforms radiologists in brain tumor diagnosis
ChatGPT, an AI model based on OpenAI's GPT-4 LLM, has shown some impressive skills in diagnosing brain tumors. Researchers from Osaka Metropolitan University's Graduate School of Medicine ran a study comparing the diagnostic accuracy of ChatGPT and radiologists using 150 preoperative brain tumor MRI reports. The results were pretty surprising: ChatGPT scored 73%, surpassing the accuracy of neuroradiologists (72%) and general radiologists (68%).
ChatGPT's performance varies based on report authorship
The study also found that ChatGPT's diagnostic accuracy changed based on who wrote the clinical report. When it was interpreting reports by neuroradiologists, the AI model hit an accuracy high of 80%. But when it came to reports by general radiologists, that number fell to 60%. So, it looks like the kind of radiologist behind the report can really affect how precise ChatGPT's diagnoses are.
Future prospects of AI in medical diagnosis
The research team, led by grad student Yasuhito Mitsuyama and Associate Professor Daiju Ueda, is all set to dive deeper into the potential of large language models like ChatGPT in other diagnostic imaging fields. Their goal? To lighten the load on doctors, boost diagnostic accuracy, and use AI to create better educational environments. Mitsuyama said, "These results suggest that ChatGPT can be useful for preoperative MRI diagnosis of brain tumors."
