Informed decomposition of electroencephalographic data.

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Current electroencephalographic (EEG) decomposition methods, like blind source separation, struggle to integrate prior information.
• Existing constrained optimization techniques offer limited adherence to prior constraints and require manual parameter tuning.

Purpose of the Study:

• To develop an informed decomposition approach for EEG data that enhances the modeling and separation of distinct subspaces.
• To improve upon existing methods by adaptively determining optimal model sizes and incorporating prior knowledge.

Main Methods:

• An informed decomposition approach building on constrained optimization for independent component analysis (ICA) was developed.
• A likelihood function was employed to adaptively determine the optimal model size for each target subspace.

Main Results:

• The proposed method generated ordered independent subspaces with reduced residual mixing compared to conventional techniques.
• Improved modeling of specific EEG features was observed, alongside a decrease in the number of components required per model.

Conclusions:

• The informed decomposition approach was compared to standard methods (Infomax, FastICA, PCA, JADE, SOBI) for artifact removal (EOG, EMG) and neural activity modeling.
• This method demonstrates superior identification and separation of distinct subspaces within EEG data, with better preservation of remaining data, by incorporating problem-specific information.