Biology subjects

Haxel, L.

Publications and source records attributed to Haxel, L..

4 recordsLinked to original sources

bioRxiv2025

Personalized real-time inference of momentary excitability from human EEG

The efficacy of transcranial magnetic stimulation (TMS) is often limited by non-adaptive protocols that disregard instantaneous brain states, potentially constraining therapeutic outcomes. Current EEG-guided approaches are hindered by their reliance on motor-evoked potentials (MEPs), which confound cortical and spinal excitability and restrict applications to the motor cortex, and a dependence on static biomarkers that cannot adapt to changing neurophysiological patterns. We introduce PRIME (Personalized Real-time Inference of Momentary Excitability), a deep learning framework that predicts cortical excitability, quantified by TMS-evoked potential (TEP) amplitude, from raw EEG signals. By targeting cortical excitability directly, PRIME enables brain state-dependent stimulation across any cortical region. PRIME incorporates transfer learning and continual adaptation to automatically identify personalized biomarkers, allowing stimulation timing to be adapted across individuals and sessions. PRIME successfully predicts cortical excitability with minimal latency, providing a computational foundation for next-generation, personalized closed-loop TMS interventions.

neuroscience↗

bioRxiv2025

Predictive modeling of TMS-evoked responses: Unraveling instantaneous excitability states

Transcranial magnetic stimulation (TMS) combined with electroencephalography (EEG) and electromyography (EMG) provides a unique window into instantaneous cortical and corticospinal excitability states. We investigated 50 healthy participants to determine how fluctuations in pre-stimulus brain activity influence single-trial TMS-evoked potentials (TEPs) and motor-evoked potentials (MEPs). We developed a novel automated source-level TEP extraction method using individualized spatiotemporal priors that is robust against poor single-trial signal-to-noise ratios (SNRs) and ongoing oscillations. TEP and MEP amplitudes were predicted with linear mixed-effects models based on pre-stimulation EEG band-powers (theta to gamma), while accounting for temporal drifts (within-session trends), coil control, and inter-subject differences. We found that higher pre-stimulus sensorimotor alpha, beta, and gamma power were each associated with larger TEPs, indicating a more excitable cortical state. Increases in alpha and gamma power immediately before stimulation specifically predicted larger MEPs, reflecting increased corticospinal excitability. These results reveal relationships between ongoing oscillatory brain states and TMS response amplitudes, identifying EEG biomarkers of high- and low-excitability states. In conclusion, our study demonstrates the feasibility of single-trial source-level TMS-EEG analysis and shows that spontaneous alpha-, beta-, and gamma-band oscillations modulate motor cortical and corticospinal responsiveness. These findings pave the way for EEG-informed, brain-state-dependent TMS protocols to optimize neuromodulatory interventions in clinical and research applications.

neuroscience↗

bioRxiv2025

NeuroSimo: an open-source software for closed-loop EEG- or EMG-guided TMS

ObjectiveOur goal was to create open-source software for closed-loop EEG-TMS that allows researchers to rapidly prototype and develop novel stimulation paradigms in a high-level programming language. This addresses the limitations of current solutions, which often rely on proprietary hardware and software, limiting their accessibility and customizability, or comprise ad-hoc pipelines tailored to specific use cases. ApproachWe developed NeuroSimo, a software platform that enables arbitrary EEG-TMS stimulation protocols written in Python, leveraging Pythons ecosystem of scientific, neuroimaging, and machine learning libraries. The core software is written in C++ with an embedded Python interpreter and employs the Robot Operating System (ROS 2) for inter-process communication. NeuroSimo runs on real-time-enabled Linux Ubuntu, using LabJack T4 for pulse triggering, and supports two EEG device models (Bittium NeurOne, BrainProducts actiCHamp) and TMS devices that deliver pulses via trigger signals. The software includes a graphical user interface for configuration and performance monitoring, and supports GPU processing for neural network computations. Main resultsIn brain-state-dependent stimulation using the Phastimate algorithm, which targets TMS pulses to the trough of sensorimotor -rhythm, NeuroSimo achieved a median timing error of 0.2 ms (95% CI: 0.2-0.2 ms), a 99th-percentile of 0.6 ms (0.6-0.6 ms), and a maximum of 1.4 ms. SignificanceAs an open-source platform combining Pythons flexibility with real-time closed-loop EEG-TMS, NeuroSimo enables researchers to develop and implement novel therapeutic approaches, marking a significant advance in personalized brain stimulation.

bioengineering↗

bioRxiv2024

Decoding Motor Excitability in TMS using EEG-Features:An Exploratory Machine Learning Approach

BackgroundWith the burgeoning interest in personalized treatments for brain network disorders, closed-loop transcranial magnetic stimulation (TMS) represents a promising frontier. Relying on the real-time adjustment of stimulation parameters based on brain signal decoding, the success of this approach depends on the identification of precise biomarkers for timing the stimulation optimally. ObjectiveWe aimed to develop and validate a supervised machine learning framework for the individualized prediction of motor excitability states, leveraging a broad spectrum of sensor and source space EEG features. MethodsOur approach integrates multi-scale EEG feature extraction and selection within a nested cross-validation scheme, tested on a cohort of 20 healthy participants. We assessed the frameworks performance across different classifiers, feature sets, and experimental protocols to ensure robustness and generalizability. ResultsPersonalized classifiers demonstrated a statistically significant mean predictive accuracy of 72 {+/-} 11%. Consistent performance across various testing conditions highlighted the sufficiency of sensor-derived features for accurate excitability state predictions. Subtype analysis revealed distinct clusters linked to specific brain regions and oscillatory features as well as the need for a more extensive feature set for effective biomarker identification than conventionally considered. ConclusionsOur machine learning framework effectively identifies predictive biomarkers for motor excitability, holding potential to enhance the efficacy of personalized closed-loop TMS interventions. While the clinical applicability of our findings remains to be validated, the consistent performance across diverse testing conditions and the efficacy of sensor-only features suggest promising avenues for clinical research and wider applications in brain signal classification.

neuroscience↗