Convolutional spiking neural networks for intent detection based on anticipatory brain potentials using...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Spiking neural networks (SNNs) mimic biological systems for computational efficiency.
• Convolutional layers enhance feature extraction, leading to convolutional SNNs (CSNNs).
• Anticipatory slow cortical potentials (SCPs) in EEG signals may indicate future actions like braking.

Purpose of the Study:

• To investigate the feasibility of using CSNNs for detecting anticipatory SCPs related to braking intention.
• To compare CSNN performance against other neural network models for EEG-based intention detection.

Main Methods:

• Collected EEG data from participants performing a simulated driving task with braking events.
• Utilized a CSNN architecture to analyze EEG signals and detect SCPs.
• Performed 10-fold cross-validation to compare CSNN with CNN, EEGNet, and graph neural networks.

Main Results:

• The CSNN achieved a predictive accuracy of 99.06%, outperforming all other tested neural networks.
• The CSNN demonstrated high performance metrics: 98.50% true positive rate, 99.20% true negative rate, and an F1-score of 0.98.
• CSNN performance remained robust even when EEG data was converted to spike trains, indicating computational efficiency.

Conclusions:

• CSNNs are highly effective for detecting anticipatory SCPs from EEG data, showing significant potential for brain-computer interfaces.
• The study validates the use of CSNNs in real-time intention recognition tasks, offering a computationally efficient alternative to traditional methods.