The effects of layer-wise relevance propagation-based feature selection for EEG classification: a comparative study...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCI) enable device control via neural signals.
• Motor imagery (MI) is a key BCI paradigm using imagined movements.
• Electroencephalography (EEG) is common for MI-BCI but requires effective feature selection due to signal noise and inter-subject variability.

Purpose of the Study:

• To develop and evaluate a layer-wise relevance propagation (LRP)-based feature selection method for deep learning (DL) models in MI-BCI.
• To enhance the classification performance of EEG signals in subject-dependent scenarios.

Main Methods:

• A novel LRP-based feature selection technique integrated into DL models was designed.
• The method's effectiveness was assessed using two public EEG datasets.
• Performance was evaluated across various DL backbone models in a subject-dependent setting.

Main Results:

• LRP-based feature selection significantly improved MI classification performance on both datasets.
• The enhancement was consistent across all tested DL backbone models.
• The selected features were class-discriminative and robust.

Conclusions:

• The LRP-based feature selection method is effective for improving EEG-based MI classification.
• This approach offers a valuable tool for enhancing BCI performance.
• The method shows potential for broader application in other research domains requiring EEG signal analysis.