Functional Graph Image Representation applied to EEG-based Mental Workload Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Machine learning and statistical signal processing increasingly use graph representations for inference and estimation problems.
• Functional connectivity analysis from Electroencephalography (EEG) signals is a key application area.
• Current functional connectivity metrics often suffer from redundant information due to volume conduction and overlook electrode locations.

Purpose of the Study:

• To develop an innovative approach for functional connectivity analysis from EEG signals.
• To improve Mental Workload (MW) classification by leveraging image representation of functional graphs.
• To explicitly encode electrode locations and sparse functional connectivity within the graph representation.

Main Methods:

• Functional graphs are learned from EEG signals under sparsity and structural constraints, represented as images.
• Electrode locations and sparse functional connectivity are explicitly encoded in the image representation.
• A convolutional neural network (CNN) processes these images to extract latent features for inference.

Main Results:

• The proposed method demonstrates promising performance in Mental Workload (MW) classification on a public dataset.
• The approach shows improved results compared to traditional spatial filtering techniques.
• Performance is also superior to methods relying on hand-crafted functional connectivities.

Conclusions:

• The image representation of functional graphs learned from EEG signals offers an effective approach for connectivity analysis.
• Explicitly encoding spatial information and sparse connectivity enhances machine learning model performance.
• This method shows potential for advancing EEG-based applications like Mental Workload classification.