Ensemble Deep Neural Network for Automatic Classification of EEG Independent Components.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Independent Component Analysis (ICA) is crucial for removing artifacts from multi-channel scalp electroencephalogram (EEG) signals.
• Manual classification of independent components (ICs) by experts is time-consuming and requires specialized availability.
• Current automated IC classification methods often rely on power spectrum densities (PSDs) and topoplots, potentially overlooking time-series data.

Purpose of the Study:

• To develop an improved automated method for classifying ICs in EEG signals.
• To investigate the utility of incorporating time-series data alongside PSDs and topoplots for IC classification.
• To evaluate the effectiveness of transfer learning approaches for IC classification models.

Main Methods:

• A novel ensemble deep neural network was developed.
• The model integrates time-series, PSDs, and topoplot data for IC classification.
• The study explored the application of transfer learning with the proposed model.

Main Results:

• Incorporating time-series data significantly improved IC classification accuracy.
• Transfer learning approaches demonstrated superior performance compared to training models from scratch.
• The developed model enhances the automatic removal of EEG artifacts.

Conclusions:

• Future IC classifiers should leverage time-series, PSD, and topoplot information.
• Transfer learning is a recommended strategy for developing new deep learning models in this domain.
• This work facilitates more efficient EEG artifact removal and encourages further research due to the potential for transfer learning.