Classification of motor imagery using chaotic entropy based on sub-band EEG source localization

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) offers high temporal resolution for motor imagery (MI) research but suffers from low spatial resolution.
• EEG Source Localization (ESL) algorithms aim to improve spatial resolution by estimating cortical sources from scalp EEG.
• Enhanced spatial resolution is crucial for improving the accuracy of MI-based brain-computer interfaces (BCIs).

Purpose of the Study:

• To address the limitations of low spatial resolution in EEG for MI tasks.
• To propose and validate a novel feature extraction method combining sub-band ESL with chaotic entropy.
• To enhance the classification accuracy of MI tasks.

Main Methods:

• EEG signals were filtered into 8 sub-bands.
• Sub-band signals underwent source localization to identify activation patterns and brain region activities.
• Approximate entropy, fuzzy entropy, and permutation entropy were extracted from source signals to quantify complexity.
• Support vector machine (SVM) was used for classifying different MI tasks.

Main Results:

• The proposed sub-band ESL and chaotic entropy method demonstrated improved classification accuracies on two public MI datasets (BCI Competition III IVa and BCI Competition IV 2a).
• The results indicated superior performance compared to existing methods.
• The study successfully enhanced the spatial resolution of EEG signals.

Conclusions:

• Sub-band EEG source localization effectively improves spatial resolution, offering a new approach for EEG-based MI research.
• The integration of chaotic entropy features from localized sources enhances MI classification performance.
• This method provides a promising advancement for developing more accurate and reliable BCIs.