The Association of EEG μ Rhythm Phase and Power with TMS-Assessed Cortical Excitability States

Wenshu Mai, Xingyu Zhao, Pisin Chen, Yashuang Zhao, He Wang, Xin Wang, Zhipeng Liu, Jingna Jin, Tao YinSensors 2025 · DOI

TL;DR

This study developed a closed-loop Transcranial Magnetic Stimulation (TMS) system that uses real-time EEG signals to trigger stimulation. They found that the power of the sensorimotor μ rhythm, but not its phase, significantly correlates with cortical excitability, suggesting that μ rhythm power is a more reliable feature for optimizing TMS efficacy.

What Problem Does This Paper Address?

Transcranial Magnetic Stimulation (TMS) is a powerful tool for modulating brain activity, but its effectiveness can vary significantly between individuals and even within the same individual at different times. This variability is thought to be due to the brain's dynamic, real-time state at the moment of stimulation. Researchers are exploring ways to make TMS more precise and effective by synchronizing stimulation with specific brain states. The sensorimotor μ rhythm (an 8-14 Hz oscillation) has been identified as a potential marker for cortical excitability. However, previous studies have yielded inconsistent results regarding whether the phase or power of the μ rhythm is more predictive of cortical excitability and how this relates to TMS outcomes. This paper aims to clarify this relationship and provide a foundation for developing more sophisticated, state-dependent TMS protocols.

Methods

The core innovation of this study is the development of a high-precision, closed-loop TMS system. This system relies on a real-time EEG signal processing pipeline:

  • EEG Signal Acquisition: 64-channel EEG was recorded at 5 kHz from healthy subjects.
  • μ Rhythm Phase Prediction: A Long Short-Term Memory (LSTM) network was trained to predict the μ rhythm phase in the sensorimotor cortex (specifically from the C3 channel, spatially filtered). This LSTM model was individualized for each participant. The prediction pipeline involved resampling, spatial filtering (Hjorth transform), detrending, zero-phase frequency-domain filtering, and finally, Hilbert transform to extract phase.
  • Closed-Loop TMS Triggering: When the predicted μ rhythm phase met predefined criteria (e.g., positive peak, negative peak, rising/falling phase), a TTL signal was generated to trigger a single-pulse TMS.
  • Cortical Excitability Assessment: Cortical excitability was measured using two methods:
  • Motor Evoked Potentials (MEPs): Amplitude of muscle responses in the hand after TMS.
  • TMS-Evoked Potentials (TEPs): EEG responses recorded after TMS, analyzed for specific components (N15, P30, etc.) and the Global Mean Field Potential (GMFP).
  • Pre-Stimulus Power Analysis: For each TMS trial, a 1000 ms epoch preceding stimulation was analyzed to extract μ rhythm power using spectral density estimation.

Datasets

The study utilized data from 30 healthy human subjects. No specific publicly available datasets were used for training or evaluation. The data collected consisted of simultaneous EEG and EMG recordings during TMS stimulation.

Key Insights and Ideas

  • μ Rhythm Power is Key, Not Phase: The most significant finding is that pre-stimulus μ rhythm power showed a strong positive correlation with MEP amplitude, indicating higher excitability during higher power states. In contrast, the μ rhythm phase did not show a consistent or significant correlation with MEP amplitudes or most TEP components. This challenges some prior assumptions and suggests a shift in focus for state-dependent TMS.
  • Power Modulates Late TEP Components: High μ rhythm power was associated with significantly enhanced amplitudes in the late P180 TEP component and early-to-mid components (N15-N100) of the GMFP. This suggests that power influences different aspects of the cortical response to TMS.
  • LSTM for Precise Phase Prediction: The use of an individualized LSTM network demonstrated high accuracy in predicting μ rhythm phase in real-time, enabling the construction of a reliable closed-loop TMS system. This highlights the potential of deep learning for capturing complex temporal dynamics in neural signals.
  • Individualized Frequency Bands: The study emphasized the importance of using individualized μ rhythm frequency bands (determined from resting-state EEG) for more accurate signal processing.

Key Results

  • Phase-MEP/TEP Correlation: No significant correlation was found between μ rhythm phase and MEP amplitudes or most TEP components.
  • Power-MEP Correlation: A significant positive correlation was observed between pre-stimulus μ rhythm power and MEP amplitude.
  • Power-TEP Modulation: High μ rhythm power led to significantly enhanced amplitudes in the P180 TEP component and early-to-mid GMFP components.
  • Closed-Loop System Performance: The phase-dependent closed-loop TMS system successfully delivered stimulation at the targeted μ rhythm phases with high precision.

Limitations and Future Work

  • Stimulus Intensity: The study used a fixed TMS intensity (120% RMT). The authors acknowledge that the relationship between μ rhythm phase and cortical excitability might be intensity-dependent, with weaker stimulation potentially showing different results. Future work should explore this across a range of intensities.
  • EEG Electrode Location: The C3 electrode, while commonly used for sensorimotor μ rhythm, might be more sensitive to sources in the postcentral gyrus (S1) than the precentral gyrus (M1), which is the primary target for MEPs. This potential mismatch could influence findings. Future studies could investigate S1-M1 connectivity.
  • Power-Dependent TMS System: While the study demonstrated the importance of power, it did not implement a power-dependent closed-loop TMS system. This is identified as a crucial next step to validate the findings and explore its therapeutic potential.
  • Suprathreshold Stimulation: The use of suprathreshold stimulation for MEPs might introduce confounding factors from peripheral muscle responses. Future research could explore TEPs during subthreshold TMS.
  • Individual Differences: While the LSTM was individualized, further exploration of how individual differences in neural connectivity and μ rhythm characteristics influence the observed relationships is warranted.

Summary updated

Abstract

The efficacy of transcranial magnetic stimulation (TMS) is influenced by the brain's real-time activity state. This study aimed to investigate the correlation between cortical excitability states and EEG features, specifically the phase and power of the sensorimotor μ rhythm. We developed a high-precision real-time phase prediction algorithm based on a Long Short-Term Memory (LSTM) network and constructed a closed-loop TMS system dependent on EEG phase and power. Thirty healthy subjects were recruited for single-pulse TMS experiments. Motor evoked potentials (MEPs) and TMS-evoked potentials (TEPs) were recorded simultaneously to assess cortical excitability states triggered in real time based on different EEG phase and power features. The results demonstrated no significant correlation between the μ rhythm phase and the amplitudes of MEPs or most TEP components. In contrast, pre-stimulus μ rhythm power showed a significant positive correlation with MEP amplitude. Under high-power conditions, the amplitude of the late P180 component in the sensorimotor cortex was significantly enhanced. The early-to-mid components (N15-N100) of the global mean field potential (GMFP) also exhibited significantly increased amplitudes. This study found that, compared to phase, EEG μ rhythm power exhibits a more significant correlation with TMS-assessed cortical excitability states. This finding provides a key basis for developing EEG power-dependent closed-loop TMS methods to enhance the efficacy of TMS modulation.
BibTeX
@inproceedings{mai2025association,
  author = {Wenshu Mai and Xingyu Zhao and Pisin Chen and Yashuang Zhao and He Wang and Xin Wang and Zhipeng Liu and Jingna Jin and Tao Yin},
  title = {{The Association of EEG μ Rhythm Phase and Power with TMS-Assessed Cortical Excitability States}},
  booktitle = {Sensors},
  year = {2025},
  doi = {10.3390/s25237187},
}
Figure