425 Generalized Instantaneous Functional Characterization of the Human Brain With Large Brain-State Modeling (LBM) for Adaptive Closed-Loop Neuromodulation

View at publisher Summarise with VIRAT
Authors: Graham Walter Johnson, Ghassan S. Makhoul, Derek J. Doss, Bruno Hidalgo Monroy Lerma, Angela N. Crudele, Robert Partlow Naftel, Victoria L. Morgan, Rushna Ali, Jamie J. Van Gompel, Sarah K. B. Bick, Václav Křemen, Gelareh Mohammed Zadeh
+3 more Gregory A. Worrell, Kai Joshua Miller, Dario J. Englot
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: The instantaneous functional state of a human brain remains difficult to characterize, much less predict. Regardless, humans have engineered techniques to electrically neuromodulate the brain with neurosurgically implanted devices to treat a subset of neurologic and psychiatric diseases with moderate efficacy. Accurate characterization of a brain's instantaneous state has stymied the development of more effective neuromodulation paradigms. Advanced computational methods are required to address this gap and enable large-scale neuroscience. Intracranial electroencephalography recordings from people undergoing neurosurgical workup for drug-resistant epilepsy offer high-resolution recordings of large-scale electrical phenomena of the human brain for use in developing a generalized model of a brain's functional state. METHODS: We utilized a custom deep learning transformer-based autoencoder to abstract generalized brain-states from a large cohort of patients with stereotactic-EEG (SEEG) implanted electrodes for neurosurgical workup of drug-resistant epilepsy. This large-scale unsupervised model was trained on data from 45 patients (17.4 billion tokens over 405 days of SEEG recordings with 555 ictal events), and pre-ictal density scoring was validated on 13 patients with completely unseen data (5.9 billion tokens over 138 days of SEEG data and 99 ictal events). The source code and trained generalized model will soon be made available on github.com to promote collaboration and foster further advancement of this novel paradigm. RESULTS: The 99 pre-ictal epochs from 13 completely withheld patients generalized to the predicted brain-state domain from the 555 pre-ictal events from the training data on 45 patients. This serves as a generalized brain-state marker for quantitatively driven closed-loop neuromodulation of drug-resistant epilepsy. Furthermore, this paradigm can be readily expanded to other neurologic and psychiatric diseases. CONCLUSIONS: Large Brain-State Modeling (LBM) offers a generalized framework to characterize, interpret, and predict future functional brain-states for use in basic inquiry and adaptive closed-loop neuromodulation.

Fields of Study & Subjects

EEG and Brain-Computer Interfaces Epilepsy research and treatment Neural dynamics and brain function

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
121–121
Publication date
26 March 2026
Language
EN