Overview

• Epilepsy surgery success rates remain at 50–60% due to challenges in accurately identifying epileptogenic zones (EZs).
• ChatGPT matched or outperformed board-certified epileptologists in localizing common EZ regions like the frontal and temporal lobes.
• Human experts still outperformed ChatGPT in identifying EZs in rare regions such as the insula and cingulate cortex.
• Researchers introduced EpiSemoLLM, the first large language model fine-tuned for seizure semiology interpretation, hosted on a Stevens GPU server.
• The study highlights the potential of AI-human collaboration in improving preoperative assessments and surgical outcomes for epilepsy patients.