Logo image
Back
DynamoDetrend: Removing Nonlinear Decay Artifacts in Concurrent Multimodal Neurostimulation and Electrophysiology
Abstract   Open access   Peer reviewed

DynamoDetrend: Removing Nonlinear Decay Artifacts in Concurrent Multimodal Neurostimulation and Electrophysiology

Raaj Chatterjee, Aaron Boes, Matthew Colbrook and Faranak Farzan
Brain stimulation, Vol.18(1), pp.419-420
01/2025
DOI: 10.1016/j.brs.2024.12.608
url
https://doi.org/10.1016/j.brs.2024.12.608View
Published (Version of record) Open Access

Abstract

Combining electromagnetic stimulation with electrophysiology provides valuable insights into brain circuits and networks' causal and dynamic nature. Multimodal stimulation, including transcranial magnetic stimulation (TMS) and direct cortical electrical stimulation (DCES), elicits characteristic neuronal responses detectable via scalp and intracranial electroencephalography (EEG). However, electromagnetic stimulation interacts with recording equipment and biological tissue, which produces significant nonlinear decays lasting tens of milliseconds post-pulse, obscuring early mono- and poly-synaptic responses. Standard signal reconstruction approaches, such as Independent Component Analysis (ICA) and (bi-)exponential detrending, primarily remove signal components that can be modeled using linear methods. Meanwhile, nonlinear approaches have focused on curve-fitting data to rational functions, which are limited in flexibility and only account for the capacitive effect of specific electrode lattice configurations. This study developed a novel algorithm, DynamoDetrend, to identify and remove complex nonlinear decay artifacts from multimodal electrophysiology data. DynamoDetrend is automatic, data-driven, and can be applied uniquely to each trial. The algorithm leverages time delay embedding and optimized dynamic mode decomposition (optDMD) to numerically approximate a Koopman Operator, which describes the evolution of underlying dynamics. Decaying eigenvalue-vector pairs from the approximation are clustered and removed, preserving the underlying oscillatory content. DynamoDetrend was validated against widely used open-source artifact reduction pipelines based on ICA, bi-exponential detrending, and rational curve-fitting. The methods were compared on their ability to reconstruct typical evoked and induced responses. Mean absolute errors from each pipeline were calculated relative to expected early responses in semi-simulated datasets. The validation datasets were composed of artifacts captured in phantom gel combined with human TMS-EEG, TMS-Intracranial EEG, and DCES-Intracranial EEG. DynamoDetrend offers researchers a new tool and establishes updated benchmarks for reducing decay artifacts, ultimately aiding the recovery of early neuronal responses in multimodal stimulation studies.

Details

Metrics

9 Record Views
Logo image