Estimation of Task-Related Dynamic Brain Connectivity via Data Inflation and Classification Model Explainability.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Analyzing brain function requires understanding dynamic switching between intrinsic brain networks.
• Traditional functional connectivity methods face limitations in temporal resolution and explainability due to statistical constraints.

Purpose of the Study:

• To develop a novel approach for functional connectivity analysis that improves explainability in electroencephalography (EEG) classification.
• To introduce a method that enhances raw EEG data decomposition for better process interpretation.

Main Methods:

• Proposed a novel approach for functional connectivity analysis via EEG classification explainability.
• Introduced Dynamic Influence Data Inflation (DIDI) to extract interaction signals between electrode regions.
• Utilized an end-to-end neural network classifier for raw EEG signals and employed saliency map estimation.

Main Results:

• Demonstrated improved classification accuracy on two public datasets (imagined motor movement and emotion classification).
• Revealed connectivity dynamics affecting classification decisions through dynamic connectivity support maps.
• Showcased the dual benefits of enhanced interpretability and increased classification accuracy.

Conclusions:

• The proposed methodology offers a powerful tool for understanding brain connectivity dynamics.
• This approach enhances both the accuracy and interpretability of EEG-based brain function analysis.
• Dynamic connectivity support maps provide valuable insights into the neural processes underlying classification tasks.