Single-Trial Detection and Classification of Event-Related Optical Signals for a Brain-Computer Interface Application

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Event-related optical signals (EROS) provide high spatial and temporal resolution for measuring neuronal activity.
• EROS have potential applications in brain-computer interfaces (BCIs).
• Single-trial classification of EROS data remains an underexplored area.

Purpose of the Study:

• To evaluate the performance of neural network methods for single-trial classification of motor response-related EROS.
• To investigate the feasibility of using deep learning for EROS-based BCI applications.

Main Methods:

• Utilized a high-density recording montage covering the motor cortex.
• Employed a convolutional neural network (CNN) to extract spatiotemporal features from EROS data.
• Classified left and right motor responses from EROS phase and intensity data during a reaction time task.

Main Results:

• Subject-specific CNN classifiers trained on EROS phase data achieved an average single-trial classification accuracy of approximately 63%.
• Classification performance was significantly influenced by noise reduction in intensity data.
• CNNs demonstrated successful application to single-trial classification using high-spatial-resolution EROS signals.

Conclusions:

• Deep learning models, specifically CNNs, are effective for single-trial classification of motor EROS.
• EROS data, particularly phase information, can be utilized for developing advanced BCI systems.
• Further research into noise reduction and feature extraction from EROS data is warranted for optimizing BCI performance.