EEG denoising during transcutaneous auricular vagus nerve stimulation across simulated, phantom and human data

Abstract

Objective: The acquisition of electroencephalogram (EEG) data during neurostimulation, particularly concurrent transcutaneous electrical stimulation of the auricular vagus nerve, introduces unique challenges for data preprocessing and analysis due to the presence of significant stimulation artifacts. This study evaluates various denoising techniques to address these challenges effectively. Methods: A variety of denoising techniques were investigated, including interpolation methods, spectral filtering, and spatial filtering techniques. The techniques evaluated included low-pass and notch filtering, spectrum interpolation, average artifact subtraction, the Zapline algorithm, and advanced methods such as independent component analysis (ICA), signal-space projection (SSP), and generalized eigendecomposition with stimulation artifact source separation (GED/SASS). The efficacy of these algorithms was evaluated across three distinct datasets: simulated data, data from a gelatin phantom model, and real human subject data. Results: Our findings indicate that GED (SASS) and SSP significantly outperformed other methods in reducing artifacts while preserving the integrity of the EEG signal. ICA and Zapline were effective too, but came with important limitations. These methods demonstrated robustness across different data types and conditions, providing effective artifact mitigation with minimal disruption to other essential signal components. Conclusion: This comprehensive analysis demonstrates the efficacy of advanced spatial filtering techniques in the preprocessing of EEG data during auricular vagus nerve stimulation. These methods offer promising avenues for enhancing the quality and reliability of neurostimulation-associated EEG data, facilitating a deeper understanding and wider applications in clinical and research settings.