Systems and methods to remove brain stimulation artifacts in neural signals

Inventors

Brunner, Peter • Xie, Tao • Willie, Jon • Leuthardt, Eric

Assignees

Washington University in St. Louis

Abstract

Computing systems and computer-implemented methods for removing brain stimulation artifacts in neural signals are disclosed. The method makes use of a matching pursuit algorithm to accurately extract the stimulation artifact. The disclosed method removes the stimulation artifact associated with individual stimulation pulses without needing additional information from previous stimulation pulses or other electrodes. The disclosed method is compatible for use with various stimulation frequencies, does not need any filtering, and can recover neural signals almost immediately after stimulation. The disclosed method is compatible for use in has great potential in closed-loop systems used in various neurological diagnostic and therapeutic procedures.

Core Innovation

The invention provides a system for removing brain stimulation artifacts in electrophysiological signals. A raw signal trace comprising a time history of electrophysiological signals is decomposed by matching pursuit into a plurality of line noise atoms representative of line noise, and a summation of the line noise atoms is subtracted from the raw signal trace to produce a de-noised trace. A stimulation time window containing a portion of the denoised trace that includes a brain stimulation event is then identified, and the portion within the stimulation window is replaced with a line connecting pre-stimulation window and post-stimulation window portions to produce a stimulation-free trace.

Next, the stimulation-free trace is decomposed by matching pursuit into a plurality of evoked potential atoms representative of an evoked potential. The evoked potential-free trace is produced by subtracting a summation of the plurality of evoked potential atoms from the de-noised trace. The evoked potential-free trace is then decomposed by matching pursuit into a plurality of stimulation artifact atoms representative of a stimulation artifact, and a summation of the stimulation artifact atoms is subtracted from the evoked potential-free trace to produce an artifact-free trace.

In dependent variations, the system characterizes the brain stimulation event as a single pulse electrical signal. The decomposition uses a stored dictionary of atoms including atom types such as Gabor atoms, Dirac atoms, sharp Gaussian atoms, and Fourier atoms. Further constraints define line-noise atoms by atom frequency greater than 55 Hz persisting through the full duration of the raw signal trace, define stimulation artifact atoms by frequencies greater than 70 Hz or Dirac atoms centered at the onset of the brain stimulation event spanning less than the full length of the raw signal trace, and may isolate portions of the raw signal trace within selected biologically relevant frequency bands including theta 4–7 Hz, alpha 8–12 Hz, low beta 13–20 Hz, high beta 20–30 Hz, low gamma 30–50 Hz, and broadband gamma 70–170 Hz.

Claims Coverage

The independent claim is clm-00001 and it contains a sequential, multi-stage artifact-removal pipeline that uses matching pursuit decompositions for different components. It includes at least three main inventive stages: line-noise removal, stimulation-event removal by stimulation-window interpolation, and matching-pursuit removal of evoked potentials followed by stimulation-artifact removal to produce an artifact-free trace.

Clm-00001 covers removing brain stimulation artifacts by sequentially (1) decomposing and subtracting line noise atoms, (2) interpolating across a stimulation time window to remove the stimulation event, (3) decomposing and subtracting evoked potential atoms, and (4) decomposing and subtracting stimulation artifact atoms to produce an artifact-free trace.

Stated Advantages

Documented Applications