ChatGPT Aids Neurosurgeons in Pinpointing Epileptic Seizure Locations

ChatGPT Shows Promise in Epilepsy Surgery Planning

Researchers at Stevens Institute of Technology have demonstrated that large language models (LLMs) like ChatGPT could significantly improve the accuracy of identifying epileptogenic zones (EZs) for epilepsy surgery. This breakthrough could potentially increase the success rate of surgical interventions for patients with drug-resistant epilepsy 1

1

The Challenge of Epilepsy Treatment

Epilepsy affects over 70 million people worldwide, with approximately 3.4 million in the United States alone. For about one-third of patients, medications fail to control seizures effectively. In these cases, surgical resection of the EZ can be a viable option to reduce or eliminate seizures 2

2

However, the current success rate for resective surgery hovers around 50% to 60%, largely due to inaccurate identification of EZs. The process of locating EZs involves a series of tests, including MRI, EEG, and intracranial EEG, with epileptologists interpreting the results to describe seizure semiology 3

3

Inconsistencies in Seizure Semiology Description

One major challenge in EZ identification is the inconsistent terminology used by different epilepsy centers to describe seizure semiology. As Feng Liu, Assistant Professor at Stevens Institute of Technology, explains, "Different epilepsy centers may use different terms describing the same seizure semiology." This variability can lead to confusion and potential misidentification of EZs 1

1

ChatGPT's Performance in EZ Localization

To evaluate the potential of LLMs in addressing this issue, Liu and his team conducted a study comparing ChatGPT's performance to that of board-certified epileptologists. The study involved a survey of 100 questions about EZ localization based on seizure semiology descriptions 2

2

The results, published in the Journal of Medical Internet Research, showed that ChatGPT matched or outperformed epileptologists in identifying EZs in commonly affected areas such as the frontal and temporal lobes. However, epileptologists were more accurate in identifying EZs in rarely affected regions like the insula and cingulate cortex 3

3

Related Stories

AI Tool MELD Graph Revolutionizes Epilepsy Diagnosis by Detecting Hidden Brain Lesions

AI Tool Revolutionizes Epilepsy Treatment in Children by Detecting Tiny Brain Lesions

AI Revolutionizes Brain Tumor Diagnosis: Outperforms Radiologists and Enhances Preoperative MRI Analysis

AI epilepsy detection identifies early warning signs in EEG data without visible seizures

EpiSemoLLM: A Specialized Model for Seizure Semiology

Building on these promising results, the research team developed EpiSemoLLM, the first LLM specifically designed for interpreting seizure semiology. This specialized model, hosted on a Stevens GPU server, aims to further improve the accuracy of EZ localization and serve as a valuable tool in the presurgical planning process 1

1

The Future of AI in Epilepsy Treatment

Liu emphasizes that the goal is not to replace human expertise but to enhance it: "Our results demonstrate that LLM and fine-tuned LLM might serve as a valuable tool to assist in the preoperative assessment for epilepsy surgery. The best results would be for the humans and AI to work together" 2

2

This research highlights the potential of AI to improve patient outcomes in epilepsy treatment by providing more accurate and consistent interpretations of seizure semiology, ultimately leading to more successful surgical interventions.