A facile and flexible motor imagery classification using electroencephalogram signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Mind-machine interfaces (MMI) rely on accurately identifying motor imagery (MI) tasks from brain activity.
• Electroencephalogram (EEG) signals are crucial for MMI, but multi-channel analysis is computationally intensive.
• Extracting information from complex EEG signals requires effective decomposition and classification methods to prevent data loss and misclassification.

Purpose of the Study:

• To develop a novel, efficient single-channel method for identifying right-hand and right-foot MI tasks.
• To reduce computational burden in MMI by utilizing adaptive decomposition and classification techniques.
• To improve the reliability of MMI systems through precise discrimination of brain activity.

Main Methods:

• Employed unsupervised learning for significant channel selection.
• Utilized flexible variational mode decomposition (F-VMD) for adaptive signal decomposition into narrow-band modes.
• Extracted Hjorth, entropy, and quartile features from decomposed modes and classified them using a flexible extreme learning machine (F-ELM) with adaptive hyperparameter selection.

Main Results:

• Achieved perfect performance metrics: 100% accuracy (ACC), sensitivity (SEN), specificity (SPE), and Mathew's correlation coefficient (MCC), with an F-1 score of 1.
• Demonstrated superior performance compared to existing methodologies on the same dataset.
• Validated the efficacy of the single-channel adaptive approach for MI task identification.

Conclusions:

• The proposed method is highly promising and efficient for single-channel EEG-based MMI.
• This framework enables the development of real-time MMI applications, such as robotic arms and wheelchairs.
• The adaptive decomposition and classification strategy effectively addresses the challenges of EEG signal analysis in MMI.