Learning Spatial-Spectral-Temporal EEG Features With Recurrent 3D Convolutional Neural Networks for Cross-Task Mental...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Accurate mental workload assessment is crucial for safety and health.
• Existing models often fail in cross-task scenarios due to complex brain dynamics.
• Developing generalizable models requires learning features without prior knowledge.

Purpose of the Study:

• To propose a novel deep learning architecture for cross-task mental workload assessment.
• To enable learning of electroencephalography (EEG) features across tasks without human-designed features.
• To validate the model's effectiveness in distinguishing low and high mental workload conditions.

Main Methods:

• A Recurrent 3D Convolutional Neural Network (R3DCNN) model was developed.
• EEG data was transformed using Morlet wavelet to include frequency and time dimensions.
• The R3DCNN simultaneously learned spatial, spectral, and temporal EEG features.
• Binary classification was performed on spatial n-back and arithmetic tasks.

Main Results:

• The R3DCNN achieved an average accuracy of 88.9% in cross-task mental workload classification.
• This represents a significant improvement over state-of-the-art methods.
• Convolutional layer visualization confirmed the extraction of detailed EEG features.

Conclusions:

• The R3DCNN model demonstrates high capability in identifying mental workload levels across different tasks.
• This approach overcomes limitations of single-task models and prior knowledge dependency.
• The findings support the use of R3DCNN for robust, generalizable mental workload assessment.