ChatGPT Helps Pinpoint Precise Locations of Seizures in the Brain

Researchers from the Stevens Institute of Technology and collaborating institutions have tested whether large language models (LLMs) like ChatGPT can help localize epileptogenic zones (EZs) in the brain by interpreting seizure semiology – clinical descriptions of symptoms and behaviors during seizures.

Their findings, published in the Journal of Medical Internet Research, suggest that when used carefully, LLMs may support clinical decision-making in the presurgical phase of epilepsy treatment.

Clinical context and challenges in epilepsy surgery

Epilepsy affects more than 70 million people globally, including 3.4 million in the United States. Approximately one-third of these cases are resistant to medical therapy, making surgical resection of EZs a potential treatment option. This involves removing brain tissue responsible for initiating seizures to reduce or eliminate their recurrence. However, successful outcomes depend on accurately identifying these zones, which is not always achieved. Current surgical success rates hover between 50% and 60%.

One barrier to accurate localization lies in the subjective nature of interpreting seizure semiology, which varies between epilepsy centers. Differences in terminology – for example, whether to describe a motor symptom as “asymmetric posturing” or “asymmetric tonic activity” – may lead to inconsistencies in diagnosis and surgical planning.

“Different epilepsy centers may use different terms describing the same seizure semiology,” explained Feng Liu, Assistant Professor at the Department of Systems and Enterprises, Schaefer School of Engineering and Science at Stevens Institute of Technology. “There are a lot of terms that can refer to the same thing, but different centers may use different terminology to describe it.”

"For example, the terms "asymmetric posturing" and "asymmetric tonic activity" can be used to describe the same thing,” Liu added, referring to a posture where one arm or one leg is extended while the other is flexed.

Evaluating ChatGPT and a fine-tuned model

The research team, which included collaborators from Case Western Reserve University, Rutgers University, University of California San Francisco and Goethe University, examined whether ChatGPT could improve the standardization and interpretation of seizure descriptions.

Five board-certified epileptologists completed an online survey with 100 questions, each requiring identification of the likely EZ based on a provided seizure description. ChatGPT was then tasked with the same questions. The AI's responses were evaluated against those of the clinicians.

In regions such as the frontal and temporal lobes – areas commonly implicated in epilepsy – ChatGPT’s answers were comparable to or better than those of the epileptologists. However, in less typical EZ locations, such as the insula and cingulate cortex, human experts were more accurate.

To further improve the performance of AI for this task, the team built the first LLM specially for interpreting seizure semiology, called EpiSemoLLM, which is hosted on a Stevens Institute of Technology GPU server.

Future applications and collaborative roles

The findings indicate that while LLMs alone are not sufficient for complete EZ localization, they can be useful adjuncts when paired with expert interpretation. The new EpiSemoLLM, for example, is intended to be a useful assistant in the decision-making during the presurgical workup phase for neurosurgeons and epileptologists.

Integrating the use of specialist LLMs into clinical workflows may help reduce subjectivity in semiology interpretation and support consistent surgical planning across epilepsy centers. 

“Our results demonstrate that LLM and fine-tuned LLM might serve as a valuable tool to assist in the preoperative assessment for epilepsy surgery,” Liu said. “The best results would be for the humans and AI to work together.”

Reference: Luo Y, Jiao M, Fotedar N, et al. Clinical value of chatgpt for epilepsy presurgical decision-making: systematic evaluation of seizure semiology interpretation. J Med Internet Res. 2025;27:e69173. doi: 10.2196/69173

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.

This content includes text that has been generated with the assistance of AI. Technology Networks' AI policy can be found here.