Key takeaways:
- ChatGPT was higher at predicting seizure semiology for frontal, temporal lobes vs. epileptologists.
- Epileptologists have been extra correct when decoding semiology for the cingulate cortex.
AI-based giant language fashions could also be possible to help clinicians for pre-surgical determination making in these with drug-resistant focal epilepsy, based on a examine printed within the Journal of Web Medical Analysis.
“Seizure semiology has a scientific worth to pinpoint the places of epileptogenic zones,” Feng Liu, PhD, MEng, an assistant professor within the division of techniques and enterprises on the Schaefer Faculty of Engineering and Science on the Stevens Institute of Know-how in Hoboken, N.J., advised Healio.
Information have been derived from Liu F, et al. J Med Web Res. 2025;doi:10.2196/69173.
“Deciphering seizure semiology requires intensive experience and coaching, and the variations of their descriptions make the interpretation extra sophisticated,” Liu stated.
Giant language fashions (LLMs) corresponding to ChatGPT have been explored for his or her potential as AI-based evaluation instruments, which can be suited to interpret detailed seizure semiology descriptions for correct epileptogenic zone (EZ) localization, Liu and colleagues wrote.
As such, they sought to guage ChatGPT’s capacity to appropriately interpret seizure semiology forward of epilepsy surgical procedure in contrast with the efficiency of epileptologists.
Their examine comprised two distinct information cohorts: a publicly sourced cohort of 852 semiology-EZ pairs (common age, 23.66 years; 47.4% males) from 193 peer-reviewed journal publications from the earlier 20 years, and a personal cohort of 184 semiology-EZ pairs (common age, 28.67 years; 51.1% males) collected from Far Japanese Memorial Hospital (FEMH) in Taiwan from 2017 to 2021.
ChatGPT-4 was utilized to foretell the almost definitely EZ places utilizing each zero-shot prompting (ZSP) and few-shot prompting (FSP).
Concurrently, eight epileptologists crammed out on-line surveys asking them to interpret 100 randomly chosen semiology data. Responses from human and synthetic topics have been subsequently in contrast with regional sensitivity (RSens), weighted sensitivity (WSens) and internet optimistic inference fee (NPIR) metrics for all areas of the mind.
Within the publicly sourced cohort, outcomes confirmed that, for RSens, ChatGPT registered 0.88 in FSP and 0.9 in ZSP within the frontal lobes; for the temporal lobes, it was 0.83 for FSP and 0.81 for ZSP. Within the occipital lobe, ZSP and FSP have been 0.42 and for the parietal lobe, 0.26 (ZSP) and 0.22 (FSP). For WSens, ZSP and FSP values have been 0.69 and 0.67, respectively.
The RSens for the privately sourced cohort have been as follows: within the frontal lobe, ZSP and FSP have been each 0.87; within the temporal lobe, FSP was 0.83 and ZSP 0.81; within the occipital lobe ZSP and FSP achieved 0.38 and within the parietal lobe, ZSP was 0.34 and FSP 0.32. WSens values have been 0.74 for FSP and 0.73 for ZSP.
Researchers moreover reported that ChatGPT-4 outperformed a choice of 5 epileptologists in seizure semiology for frontal and temporal lobes, with comparable outcomes evaluating the parietal and occipital lobes.
Conversely, epileptologists have been extra correct when decoding semiology for the cingulate cortex, indicating that LLM efficiency wanes when analyzing much less ceaselessly noticed areas of the mind.
Information additional confirmed ChatGPT-4’s superiority in WSens, with an FSP of 0.63 and ZSP of 0.61 in contrast with a variety of 0.49 to 0.51 from medical professionals.
Nonetheless, Liu and colleagues wrote, no important variations have been noticed between ChatGPT and the epileptologists relating to NPIR.
“This examine confirmed that leveraging giant language fashions, corresponding to ChatGPT, can present a really cheap interpretation of seizure semiology in comparison with epileptologists with years of expertise,” Liu stated. “This consequence motivated us to develop a fine-tuned LLM for decoding the seizure semiology and prediction of epileptogenic zones.”
Reference:
ChatGPT helps pinpoint exact places of seizures within the mind, aiding neurosurgeons. https://www.stevens.edu/information/chatgpt-helps-pinpoint-precise-locations-of-seizures-in-the-brain-aiding. Revealed Could 12, 2025. Accessed Could 14, 2025.
For extra info:
Feng Liu, PhD, MEng, will be reached at neurology@healio.com.
