This diagnostic accuracy study examined whether multivariate resting-state EEG network measures in the 6–9 Hz range could help distinguish non-lesional epilepsy from functional/dissociative seizures (FDS). The analysis included 148 age- and sex-matched adults with suspected seizure disorders who were later diagnosed with either epilepsy (n=75) or FDS (n=73). The goal was to see if these EEG features could separate the two groups better than chance.
The models reached a maximum balanced accuracy of 67.5%, which was reported as significantly above chance. Sensitivity was higher for epilepsy (81.8%) than for FDS (53.3%). Using epoch-wise averaging improved classification accuracy from 62.6% to 67.5%, and 77.5% of individuals received a consistent diagnostic label from the top models.
No safety issues were reported because the study used EEG recordings, not treatments. However, important limitations include that the dataset had been used in a prior study, model choice strongly influenced accuracy, and dimensionality reduction did not provide a clear advantage. The study does not show that these EEG measures cause or confirm any diagnosis.
Realistically, these results suggest EEG network features may support clinical decision-making by improving post-test probability, but they are not diagnostic on their own. More research in independent, diverse groups is needed to see how well these measures perform in real-world settings and whether they can guide care.