Machine learning validation of EEG+tACS artefact removal

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Simultaneous electroencephalography (EEG) and transcranial alternating current stimulation (tACS) is crucial for studying brain dynamics during stimulation.
• Large stimulation artefacts corrupt EEG data, necessitating effective artefact removal algorithms.
• Previous artefact removal methods face scrutiny regarding performance and residual artefacts.

Purpose of the Study:

• To investigate the efficacy of machine learning in validating EEG artefact removal algorithms used with tACS.
• To determine if residual artefacts, if present, interfere with the identification of experimental conditions and tasks.
• To build confidence in the integrity of EEG data post-artefact removal.

Main Methods:

• Ten participants performed working memory tasks (nBack, digital recall) during simultaneous EEG+tACS.
• EEG data was cleaned using a previously established tACS artefact removal algorithm.
• Linear discriminant analysis was employed to classify EEG segments from different experimental conditions and tasks.

Main Results:

• Machine learning successfully differentiated baseline, tACS, and task conditions with 65%-94% accuracy.
• EEG data from distinct working memory tasks (nBack, digital recall) were separable during stimulation (accuracy >72%).
• Classification performance indicated residual artefacts did not dominate the cleaned EEG data.

Conclusions:

• The findings build confidence that genuine EEG information is preserved after artefact removal.
• This machine learning approach offers a novel method for validating tACS artefact removal techniques.
• The study supports the use of cleaned EEG data for reliable analysis of brain activity during tACS.