Decoding personalized motor cortical excitability states from human electroencephalography
- Resource Type
- Authors
- Sara J. Hussain; Romain Quentin
- Source
- Subject
- Adult
Computer science
Brain activity and meditation
medicine.medical_treatment
brain stimulation
Electromyography
Electroencephalography
medicine
Humans
Multidisciplinary
medicine.diagnostic_test
[SCCO.NEUR]Cognitive science/Neuroscience
[SCCO.NEUR] Cognitive science/Neuroscience
Motor Cortex
Evoked Potentials, Motor
Transcranial Magnetic Stimulation
Neuromodulation (medicine)
Transcranial magnetic stimulation
machine learning
medicine.anatomical_structure
Brain state
Cortical Excitability
Neuroscience
Decoding methods
Motor cortex
- Language
OBJECTIVE: Brain state-dependent transcranial magnetic stimulation (TMS) requires real-time identification of cortical excitability states. Current state-of-the-art approaches deliver TMS during brain states shown to correlate with motor cortex (M1) excitability at the group level. Here, we hypothesized that machine learning classifiers could successfully discriminate between high and low M1 excitability states in individual subjects using power spectral features obtained from low-density electroencephalography (EEG) signals. METHODS: We analyzed a pre-existing publicly available dataset that delivered 600 single TMS pulses to the right M1 during EEG and electromyography (EMG) recordings. Personalized multivariate pattern classification was used to discriminate between brain states during which TMS evoked small and large motor-evoked potentials (MEPs). These brain states were labeled as low and high M1 excitability states, respectively. RESULTS: Personalized classifiers successfully discriminated between low and high M1 excitability states in every subject tested. MEPs elicited during classifier-predicted high excitability states were significantly larger than those elicited during classifier-predicted low excitability states in all subjects. Personalized classifiers trained using EEG features obtained immediately before each TMS pulse performed better than personalized classifiers trained using EEG features obtained from earlier time points. Personalized classifiers generalized weakly but significantly across subjects. CONCLUSION: Individual subjects exhibit unique brain activity patterns that correspond to low and high M1 excitability states. These patterns can be efficiently captured using power spectral features obtained from low-density EEG signals. Deploying individualized classifiers during brain state-dependent TMS may enable effective, fully personalized neuromodulation in the future.