Consistency of the blind source separation computed with five common algorithms for magnetoencephalogram background...

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Blind Source Separation (BSS) is crucial for analyzing complex brain signals like magnetoencephalography (MEG).
• Limited comparative studies exist on BSS algorithm performance with real MEG data across various conditions.
• Understanding BSS decomposition consistency is vital for reliable brain activity analysis.

Purpose of the Study:

• To compare the consistency of five widely used BSS algorithms (AMUSE, SOBI, JADE, extended-Infomax, FastICA) on real MEG data.
• To investigate the influence of epoch length (10s-90s) on BSS decomposition similarity.
• To guide the selection of BSS algorithms for optimal brain signal analysis.

Main Methods:

• Real MEG data from 26 subjects were analyzed using five BSS algorithms.
• Epoch lengths of 10s, 20s, 40s, 60s, and 90s were utilized.
• A statistical criterion based on Factor Analysis determined the number of components for decomposition.

Main Results:

• AMUSE-SOBI and JADE-FastICA demonstrated the highest decomposition similarity.
• 'AMUSE-JADE' and 'SOBI-JADE' pairs exhibited the most dissimilar outcomes.
• BSS decomposition similarity increased with longer epoch lengths.

Conclusions:

• Algorithm choice significantly impacts MEG data decomposition; AMUSE and JADE offer complementary perspectives.
• Longer epochs enhance the reliability and consistency of BSS results in MEG analysis.
• Findings aid researchers in selecting appropriate BSS techniques for distinct brain signal components.