Ballistocardiogram Artifact Reduction in Simultaneous EEG-fMRI Using Deep Learning.

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Machine Learning

Background:

• Concurrent electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) offers high temporal and spatial resolution for brain activity.
• Ballistocardiogram (BCG) artifacts, caused by cardiac-induced movement, significantly contaminate EEG signals during simultaneous EEG-fMRI.
• Existing BCG removal methods often rely on linear decomposition, which is suboptimal due to the nonlinear nature of BCG propagation relative to the electrocardiogram (ECG).

Purpose of the Study:

• To present a novel deep learning-based method for suppressing ballistocardiogram (BCG) artifacts in EEG during simultaneous EEG-fMRI recordings.
• To address the limitations of linear decomposition methods for BCG artifact removal.
• To develop a software-based solution that does not require additional hardware.

Main Methods:

• Recurrent neural networks (RNNs) were trained to learn the nonlinear mappings between the electrocardiogram (ECG) and BCG-corrupted EEG signals.
• The performance of the RNN model was evaluated against the established Optimal Basis Set (OBS) method.
• Model generalization across different subjects was investigated.

Main Results:

• The proposed RNN algorithm demonstrated superior performance in reducing BCG artifact power at critical frequencies compared to the OBS method.
• The artifact suppression led to improved classification accuracy for task-relevant EEG signals.
• The method effectively reduced BCG-related artifacts without additional hardware.

Conclusions:

• The presented deep learning architecture, utilizing RNNs, provides an effective means to reduce BCG artifacts in EEG-fMRI recordings.
• This novel approach offers a promising alternative to existing methods, with potential for real-time application and integration with current hardware techniques.