A supervised data-driven spatial filter denoising method for speech artifacts in intracranial electrophysiological...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Direct brain recordings in awake patients offer unique insights into human speech neurophysiology.
• Intracranial electroencephalography (iEEG) signals are susceptible to speech artifacts, particularly those related to fundamental frequency (F0).
• These artifacts overlap with high-gamma frequencies critical for speech production and perception.

Purpose of the Study:

• To develop and validate a spatial-filtering approach for identifying and removing acoustic-induced speech artifacts from iEEG recordings.
• To improve the signal quality of iEEG data acquired during overt speech in neurosurgical patients.
• To preserve authentic neural activity while effectively denoising speech-related contaminations.

Main Methods:

• A novel spatial-filtering technique was developed to isolate and remove speech artifacts.
• The method focuses on identifying and subtracting acoustic-induced signal contaminations.
• Data-driven approaches were employed to optimize artifact removal without compromising neural signals.

Main Results:

• The developed spatial-filtering method effectively identified and removed speech artifacts from iEEG signals.
• Traditional reference schemes were found to degrade signal quality.
• The data-driven spatial-filtering approach successfully denoised recordings while preserving underlying neural activity.

Conclusions:

• Acoustic-induced speech artifacts in iEEG can be effectively mitigated using a spatial-filtering approach.
• This data-driven method offers a superior alternative to traditional reference schemes for denoising iEEG during speech.
• The technique enhances the reliability of iEEG data for studying human speech neurophysiology.