3D CNN to Estimate Reaction Time from Multi-Channel EEG.

Area of Science:

• Neuroscience
• Biomedical Signal Processing
• Machine Learning

Background:

• Human reaction time (RT) is crucial for understanding sensory-motor functions and developing brain-computer interfaces (BCI).
• Advancements in sensor technology, computation, and neural networks drive biomedical signal processing.
• Existing models often overlook inter-channel brain signal relationships.

Purpose of the Study:

• To explore the relationship between behavioral responses and electroencephalogram (EEG) signals during perceptual decision-making.
• To introduce a novel 3D convolutional neural network (CNN) architecture for estimating RT from single-trial multi-channel EEG.
• To improve the accuracy of RT prediction by incorporating spatial inter-channel relationships.

Main Methods:

• Utilized a generalized 3D CNN architecture for analyzing multi-channel EEG data.
• Applied the model to estimate RT for a simple visual task.
• Focused on leveraging both spectral information and spatial relationships among EEG channels.

Main Results:

• The 3D CNN model achieved a root mean square error (RMSE) of 91.5 ms for RT prediction.
• A correlation coefficient of 0.83 was obtained between predicted and actual RT.
• These results significantly surpass previous benchmarks in comparable studies.

Conclusions:

• The developed 3D CNN model effectively estimates RT from EEG, outperforming existing methods.
• Incorporating inter-channel spatial relationships enhances the accuracy of brain signal decoding.
• This approach holds promise for advancing BCI, psychology, and neuroscience research, aiding in the development of assistive devices.