Removing the cardiac field artifact from the EEG using neural network regression

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Investigating central nervous and cardiovascular interactions using EEG is challenging due to cardiac field artifacts (CFA).
• CFA contaminates EEG signals when analyzed time-locked to cardiac events, confounding neuro-cardiac data.
• Accurate analysis requires disentangling cardiac electrical activity from neural signals.

Purpose of the Study:

• To present a novel nonlinear regression method using neural networks for CFA removal from EEG signals.
• To enable accurate analysis of EEG data time-locked to cardiac events.
• To provide a replicable, data-driven solution for single-trial CFA removal.

Main Methods:

• Developed neural network models to predict EEG episodes based on ECG and CFA-related information.
• Trained models to predict and remove CFA from EEG signals containing visual stimuli time-locked to ECG.
• Conducted an extensive grid search to determine optimal model hyperparameters.

Main Results:

• The proposed method effectively removes CFA from EEG signals on a single-trial level.
• Stimulus-evoked activity and intertrial phase coherence remained unaffected by CFA removal.
• The approach is purely data-driven, ensuring replicable results.

Conclusions:

• The neural network-based method successfully removes cardiac field artifacts from EEG.
• This technique allows for precise investigation of neuro-cardiac interactions without confounding artifacts.
• The method offers a robust and replicable solution for single-trial EEG artifact removal.