Machine learning based artifact rejection for transcranial magnetic stimulation electroencephalogram
Inventors
Etkin, Amit • Keller, Corey • Wu, Wei
Assignees
US Department of Veterans Affairs
Publication Number
US-11577090-B2
Publication Date
2023-02-14
Expiration Date
2037-12-19
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A method for machine learning based artifact rejection is provided. The method may include applying a machine learning model to identify artefactual independent components in transcranial magnetic stimulation electroencephalogram data collected during a transcranial magnetic stimulation procedure. Clean transcranial magnetic stimulation electroencephalogram data is generated by removing, from the transcranial magnetic stimulation electroencephalogram data, the artefactual independent components. Real-time adjustments to parameters of the transcranial magnetic stimulation procedure may be performed based on the clean transcranial magnetic stimulation electroencephalogram data. Related systems and articles of manufacture, including computer program products, are also provided.
Core Innovation
The invention provides a method for machine learning based artifact rejection in transcranial magnetic stimulation (TMS) electroencephalogram (EEG) data. This method involves applying a machine learning model to identify artefactual independent components within the TMS EEG data collected during a TMS procedure. Clean TMS EEG data is generated by removing these artefactual components, and this clean data is used to perform real-time adjustments to one or more parameters of the TMS procedure.
The problem addressed arises because TMS EEG data typically contains various artifacts from sources such as the magnetic stimulus itself, subject motion including scalp muscle activation and eye blinks, coil clicks, coil recharge, and others. These artifacts skew subsequent data analysis and manual artifact rejection processes are inconsistent, subjective, and time-consuming, preventing real-time use of the EEG data to guide TMS procedures. The invention solves this issue by providing a machine learning based technique that can rapidly and consistently remove artifacts, generating clean EEG data suitable for real-time procedural parameter adjustments.
The invention also details specific preprocessing methods including removing EEG data recorded immediately after stimulation (e.g., the first 10 milliseconds), filtering to remove spectrally irrelevant data such as low-frequency drifts and high-frequency line noise, epoching, and re-referencing EEG data to a common average. The EEG data is decomposed into independent components, each representing a non-Gaussian signal that may originate from artefactual or neural sources. A machine learning classifier, such as a Fisher linear discriminant analyzer trained on features extracted from independent components, identifies artefactual components for removal. The extracted features capture spatial, temporal, and spectral properties indicative of artifacts, enabling automatic and reliable artifact rejection that supports real-time TMS parameter adjustment.
Claims Coverage
The patent presents multiple inventive features across several independent claims centered on a machine learning based artifact rejection method and system for TMS EEG data.
Machine learning based artifact rejection method
Decomposing TMS EEG data collected during a TMS procedure into multiple non-Gaussian independent components; applying a machine learning model to identify one or more artefactual independent components; generating clean EEG data by removing identified artefactual components; performing adjustments to TMS procedure parameters based at least on the clean EEG data.
System for artifact rejection and TMS parameter adjustment
A system with processor and memory configured to decompose TMS EEG data into independent components, apply a machine learning model to identify artefactual components, generate clean EEG data by removing these components, and perform real-time adjustments to the TMS procedure parameters based on the clean data.
Preprocessing prior to decomposition
Preprocessing includes removing portions of EEG data recorded within a threshold time after TMS stimuli, filtering to remove spectrally irrelevant artifacts (including frequencies outside 8-12 Hz band), and epoching and/or re-referencing EEG data to a common average.
Feature vector generation for independent components
Generating a feature vector for the independent components capturing characteristics such as spatial range, regional activation, border activation, correlation with eye movement and blink templates, electrocardiogram artifact features, temporal features via wavelet decomposition, source activity complexity, maximum magnitude across trials, short-time magnitude features capturing TMS-evoked potentials, skewness, band-power in standard EEG frequency bands, and EEG spectral features.
The claims comprehensively cover methods and systems for artifact rejection of TMS EEG data using machine learning models applied to features extracted from independent components, with preprocessing steps and real-time adjustment of TMS procedure parameters based on clean EEG data.
Stated Advantages
Provides consistent artifact rejection that is faster and more reliable than manual methods.
Enables real-time availability of clean TMS EEG data to guide adjustments during TMS procedures.
Removes artifacts originating from multiple sources including TMS stimuli, subject motion, coil clicks, and cardiac artifacts.
Improves accuracy of EEG data analysis by removing spurious signals and artifacts.
Documented Applications
Used in transcranial magnetic stimulation procedures including single pulse, paired-pulse, and repetitive TMS.
Provides real-time guidance for adjusting TMS parameters such as stimulation duration, frequency, magnitude, area, and inter-stimulus intervals based on clean EEG data.
Applicable for diagnostics and treatment of disorders like depression, PTSD, neuropathic pain by analyzing brain responses to TMS.
Interested in licensing this patent?