Portable brain-computer interface based on novel convolutional neural network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) offers high temporal resolution for brain activity monitoring.
• Traditional EEG methods involve uncomfortable procedures like skin preparation, conductive gels, and restrictive hats.
• These conventional methods present challenges for subject comfort and adherence.

Purpose of the Study:

• To develop a novel, portable, dry-electrode, wireless brain-computer interface (BCI) for comfortable EEG signal acquisition.
• To implement a 3D convolutional neural network (CNN) for classifying motor imagery (MI) tasks using EEG data.
• To evaluate the reliability and classification performance of the proposed EEG system and CNN algorithm.

Main Methods:

• Designed a portable BCI system featuring 24-bit ADCs and a wireless microprocessor for comfortable EEG acquisition.
• Transmitted EEG signals wirelessly via Bluetooth to a personal computer.
• Utilized wavelet package decomposition (WPD) for frequency domain representation and engineered a 3D input for CNN training.
• Classified motor imagery (MI) experiments using the developed 3D CNN algorithm.

Main Results:

• The portable EEG system demonstrated reliable signal acquisition.
• The 3D CNN achieved a classification performance of kappa = 0.564 for MI tasks.
• Statistical analysis (t-test) confirmed significant performance improvement of the 3D CNN over state-of-the-art methods.

Conclusions:

• The proposed portable EEG system offers a comfortable and reliable solution for brain activity measurement.
• The 3D CNN approach significantly enhances classification accuracy for motor imagery tasks compared to existing methods.
• This integrated system and algorithm advance the field of noninvasive brain-computer interfaces.