Enhancing the performance of a deep convolutional neural network model for motor imagery classification using EEG...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-electroencephalography (MI-EEG) is crucial for brain-computer interfaces (BCIs) aiding individuals with motor disabilities.
• Decoding MI-EEG signals is challenging due to non-stationarity, noise, and low signal-to-noise ratio.
• Deep learning models show promise but require enhancements for robust MI-EEG analysis.

Purpose of the Study:

• To propose and evaluate a novel EEG channel-wise attention module (ECWAM) integrated into a deep convolutional neural network (deep CNN).
• To enhance the accuracy of motor imagery-electroencephalography decoding using the proposed ECWAM.
• To compare the performance of the ECWAM against conventional channel-wise attention methods.

Main Methods:

• A novel EEG channel-wise attention module (ECWAM) was developed and incorporated into a deep CNN architecture.
• The ECWAM calculates channel scores for mu band EEG channels, amplifying prominent channels.
• The proposed deep CNN with ECWAM was evaluated on a binary class MI dataset from 54 subjects.

Main Results:

• The proposed deep CNN with ECWAM achieved a statistically significant improvement in average classification accuracy from 63.96% to 68.98% (p=0.02).
• The ECWAM demonstrated superior performance compared to conventional channel-wise attention modules.
• Scalp map analysis revealed that the ECWAM yields higher channel rankings in the motor cortex region, crucial for MI activity.

Conclusions:

• The novel EEG channel-wise attention module (ECWAM) effectively enhances deep CNN performance for MI-EEG classification.
• The ECWAM improves the decoding accuracy of brain-computer interfaces by focusing on relevant EEG channels.
• This approach offers a promising advancement for assistive technologies relying on motor imagery decoding.