BCIwiki (bciwiki.com) — Developmental Neuroscience, Great Ormond Street Institute of Child Health, University College London, London, UK. posted the study as a preprint on medRxiv on July 24, 2026. The team developed a deep learning architecture for multichannel interictal intracranial EEG to generate hypotheses about epileptogenic-zone location. The model combines a Morlet-wavelet temporal Transformer with a spatial attention encoder.
The model was evaluated with 50.5 hours of intracranial EEG from 161 patients across 17,012 channels at 7 independent centers. Leave-one-center-out validation yielded a pooled AUROC of 0.778 with a 95% confidence interval of 0.748 to 0.808, with above-chance discrimination at every held-out center. The researchers report performance comparable to established electrophysiological baselines across centers and implantation modalities. However, the retrospective design and absence of structural imaging or effective-connectivity priors limit the conclusion, and prospective validation is still required. This study is a preprint and has not been peer reviewed.