Decoding of Hand Gestures from Electrocorticography with LSTM Based Deep Neural Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain Computer Interfaces (BCI) are crucial for prosthetic control.
• Electrocorticography (ECoG) signals offer rich temporal information for decoding hand gestures.
• Previous methods often focused on temporal aspects, potentially overlooking frequency-specific power variations.

Purpose of the Study:

• To improve hand gesture classification accuracy using ECoG signals.
• To explore the utility of power variations across multiple frequency bands (4-200 Hz).
• To develop a BCI system that equally weights temporal and spatial information from different frequency bands.

Main Methods:

• Feature reduction using Statistical and Principal Component Analysis (PCA) was applied to six distinct frequency bands of ECoG data.
• A Long Short-Term Memory (LSTM) neural network was employed to leverage both temporal and spatial features.
• The proposed method was evaluated on the 'fingerflex' dataset, comprising ECoG recordings from seven subjects performing five finger flexions.

Main Results:

• Observed significant power variations across six frequency bands in ECoG recordings during hand gestures.
• Achieved an average classification accuracy of 82.4% using statistical feature reduction.
• Demonstrated superior performance compared to existing state-of-the-art methods in hand gesture decoding.

Conclusions:

• Integrating frequency-specific power variations alongside temporal information enhances ECoG-based hand gesture decoding.
• The proposed LSTM architecture with statistical feature selection provides a robust and accurate BCI control method.
• This approach represents a significant advancement in developing more intuitive and responsive prosthetic devices.