Thalamic stereo-EEG detects sleep stages N2 and N3 with discrimination approaching scalp EEG
- 1Department of Neurological Surgery, University of Chicago
Abstract
Sleep and epilepsy interact strongly during sleep stages N2 and N3, when synchronization of brain rhythms is increased. Sleep staging relies on scalp EEG, which is often absent or incomplete during stereo-EEG monitoring and unavailable from implanted devices, while thalamic leads are increasingly implanted for epilepsy neuromodulation. Given the role of the thalamus in spindle generation and slow oscillation dynamics, we hypothesized that thalamic channels would identify N2 and N3 (N2+N3) better than other intracranial sites. We analyzed 477 h of stereo-EEG with simultaneous scalp EEG from six adults, including 46 thalamic bipolar channels on 11 electrodes. Two classifiers were evaluated on identical channels: a random forest with hidden Markov decoding (RF + HMM) trained leave-one-participant-out, and the pretrained SleepSEEG algorithm developed without thalamic channels. The primary endpoint was discrimination of N2+N3 from other stages as classified from scalp EEG by a sleep expert. Among regions sampled in all participants, thalamic channels performed best (area under the curve [AUC] 0.94 with RF + HMM and 0.93 with SleepSEEG across whole recordings; 0.88 and 0.86 at night). With RF + HMM, thalamic channels outperformed mesial temporal and temporal neocortical channels in all six participants across whole recordings. The advantage persisted against channel-count-matched subsets and a white-matter control with both classifiers. Individual thalamic electrodes reached AUCs of 0.79-0.99 with RF + HMM (10 of 11 at least 0.89). These offline findings support thalamic N2+N3 detection and motivate real-time evaluation for sleep-targeted neuromodulation.
Related
- Waterborne eDNA metabarcoding of the upper Llano River watershed (Texas, USA) detects regionally unreported green plant taxa
- Visualization of intra-thalamic nuclei with optimized white-matter-nulled MPRAGE at 7T
- Blunted cerebello-basal ganglia-thalamic circuit connectivity in melancholic depression
- An in silico cortico-basal ganglia-thalamic testbed for comparing visual and deep brain stimulation in Parkinson’s disease
- Scalp EEG detects motor sequence replay signatures during wakeful rest
BibTeX
@misc{alipour2026thalamic,
title = {{Thalamic stereo-EEG detects sleep stages N2 and N3 with discrimination approaching scalp EEG}},
author = {M. Alipour and W. van Drongelen and P. C. Warnke and N. P. Issa and K. Lee and D. Satzer},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.31.763362},
url = {https://www.biorxiv.org/content/10.64898/2026.10.31.763362v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Thalamic stereo-EEG detects sleep stages N2 and N3 with discrimination approaching scalp EEG
AU - M. Alipour
AU - W. van Drongelen
AU - P. C. Warnke
AU - N. P. Issa
AU - K. Lee
AU - D. Satzer
PY - 2026
DA - 2026/10/31
PB - bioRxiv
DO - 10.64898/2026.10.31.763362
UR - https://www.biorxiv.org/content/10.64898/2026.10.31.763362v1
AB - Sleep and epilepsy interact strongly during sleep stages N2 and N3, when synchronization of brain rhythms is increased. Sleep staging relies on scalp EEG, which is often absent or incomplete during stereo-EEG monitoring and unavailable from implanted devices, while thalamic leads are increasingly implanted for epilepsy neuromodulation. Given the role of the thalamus in spindle generation and slow oscillation dynamics, we hypothesized that thalamic channels would identify N2 and N3 (N2+N3) better than other intracranial sites. We analyzed 477 h of stereo-EEG with simultaneous scalp EEG from six adults, including 46 thalamic bipolar channels on 11 electrodes. Two classifiers were evaluated on identical channels: a random forest with hidden Markov decoding (RF + HMM) trained leave-one-participant-out, and the pretrained SleepSEEG algorithm developed without thalamic channels. The primary endpoint was discrimination of N2+N3 from other stages as classified from scalp EEG by a sleep expert. Among regions sampled in all participants, thalamic channels performed best (area under the curve [AUC] 0.94 with RF + HMM and 0.93 with SleepSEEG across whole recordings; 0.88 and 0.86 at night). With RF + HMM, thalamic channels outperformed mesial temporal and temporal neocortical channels in all six participants across whole recordings. The advantage persisted against channel-count-matched subsets and a white-matter control with both classifiers. Individual thalamic electrodes reached AUCs of 0.79-0.99 with RF + HMM (10 of 11 at least 0.89). These offline findings support thalamic N2+N3 detection and motivate real-time evaluation for sleep-targeted neuromodulation.
ER -