An Ensemble CNN for Subject-Independent Classification of Motor Imagery-based EEG

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), excels in classification tasks, including electroencephalogram-based Brain Computer Interfaces (BCIs).
• A significant challenge in BCIs is achieving subject-independent decoding due to substantial inter-subject variability in brain activity.

Purpose of the Study:

• To investigate the efficacy of an ensemble CNN classifier for decoding electroencephalogram (EEG) signals in motor imagery tasks.
• To enhance subject-independent classification performance in BCIs by integrating CNNs and ensemble learning.

Main Methods:

• An ensemble CNN classifier was developed and applied to EEG data from motor imagery experiments.
• Performance was evaluated against average base CNN classifiers and state-of-the-art subject-independent classification methods.

Main Results:

• The ensemble CNN demonstrated superior classification accuracy compared to average base CNNs, with improvements up to 9%.
• The proposed method outperformed several state-of-the-art techniques on benchmark BCI datasets (BCI Competition IV 2A and 2B).

Conclusions:

• Ensemble CNNs offer a promising approach to overcome inter-subject variability in EEG decoding for BCIs.
• This method advances the development of more robust and generalizable subject-independent brain-computer interfaces.