An international group of scientists have found that ChatGPT can help specialists identify where seizures are starting from.
In a paper published in the Journal of Medical Internet Research, Yaxi Luo and colleagues wanted to investigate how useful “large language models” such as ChatGPT are in epilepsy management.
The group wanted to see if ChatGPT could interpret different descriptions of seizures and accurately identify where the seizures start.
The researchers used epilepsy records of 852 seizure descriptions with a known seizure focus (area in the brain where seizures originate), from 193 other journal papers. The researchers then asked ChatGPT to evaluate the information and find the seizure focus.
The performance of ChatGPT was compared to that of eight epilepsy specialists who reviewed 100 of the epilepsy records at random.
According to the findings, ChatGPT outperformed the epilepsy specialists in identifying the more common areas where seizures tend to originate, such as the frontal and temporal lobes. However, the specialists were better at identifying more unusual areas where seizures started.
The team built a large language model specifically for seizure descriptions, called EpiSemoLLM, to support specialists in presurgical evaluation. They concluded that the “best results would be for the humans and AI to work together”.
The research group spanned the US, Greece, Taiwan and Germany. The paper is available on the Journal of Medical Internet Research website.
More articles
