Automatic artifact component removal using a neural network in MCG signal

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Pulse-type artifacts frequently contaminate signals from 61-channel magnetoencephalography (MCG) systems.
• Accurate artifact removal is crucial for reliable MCG data analysis.
• Existing artifact rejection methods may lack automation or component-specific processing.

Purpose of the Study:

• To develop and validate an automated algorithm for removing pulse-type artifacts from MCG signals.
• To enhance the reliability and usability of MCG data by improving signal quality.
• To leverage the strengths of principal component analysis (PCA) and neural networks for artifact identification and removal.

Main Methods:

• A hybrid algorithm combining PCA and a neural network was developed.
• PCA decomposes the MCG signal into components.
• A neural network identifies and removes artifact-corrupted components based on extracted time-domain features (max, min, peak-to-peak, variance, mean, skewness, kurtosis).

Main Results:

• The proposed algorithm successfully removed pulse-type artifacts from MCG signals.
• The neural network achieved 97% agreement with human expert decisions in artifact identification.
• The artifact removal process was automated and performed on a component-by-component basis.

Conclusions:

• The combined PCA and neural network algorithm offers an effective and automated solution for pulse-type artifact removal in MCG.
• This technique significantly improves MCG signal quality, enabling more accurate downstream analysis.
• The component-based approach enhances the precision of artifact rejection.