CNN-based framework using spatial dropping for enhanced interpretation of neural activity in motor imagery...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Interpreting brain activity from motor imagery (MI) is crucial for medical applications.
• Machine learning models face challenges due to high variability in identifying imagined actions.
• Enhanced spatial interpretation of neural patterns is needed for accurate MI classification.

Purpose of the Study:

• To develop a Convolutional Neural Network (CNN) architecture for improved spatial interpretation of brain neural patterns in MI tasks.
• To contrast Power Spectral Density and Continuous Wavelet Transform for 2D-feature extraction from EEG data.
• To enhance the classification accuracy and interpretability of MI tasks.

Main Methods:

• Utilized a CNN architecture with topographic interpolation for spatial EEG pattern preservation.
• Implemented a spatial dropping algorithm to refine feature selection.
• Compared Power Spectral Density and Continuous Wavelet Transform for feature extraction.
• Evaluated performance on bi-class and three-class MI tasks.

Main Results:

• Continuous Wavelet Transform combined with a thresholding strategy improved CNN accuracy and interpretability.
• The method identified significant contributions over the sensorimotor cortex.
• Differentiated behavior of rhythms [Formula: see text] and [Formula: see text] was observed.

Conclusions:

• The proposed CNN architecture with CWT and spatial dropping enhances MI classification.
• Improved spatial interpretability aids in understanding brain activity related to imagined movements.
• This approach shows promise for medical diagnosis and monitoring using EEG-based MI.