Domain Adaptation Model for EEG Analysis: Mitigating Spatial and Spectral Variability in Heterogeneous Datasets.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Signal Processing

Background:

• Electroencephalography (EEG) is vital for diagnosing neurological disorders.
• Model generalization in EEG analysis is often limited by domain shifts due to variations in data acquisition and standards.

Purpose of the Study:

• To develop a robust domain adaptation model for EEG analysis that overcomes domain shifts.
• To improve the generalization and classification performance of EEG-based diagnostic models across heterogeneous datasets.

Main Methods:

• Proposed a novel domain adaptation model integrating the Mean Teacher Model (MTM), Maximum Mean Discrepancy (MMD), Topology Loss, Frequency Regularization, and Attractor Loss.
• Utilized EEG-specific properties to align heterogeneous datasets, specifically aligning CHB-MIT (source) with TUH Seizure / MDD datasets (targets).

Main Results:

• Demonstrated significant improvements in classification performance (F1 score) and domain alignment (Jaccard coefficient).
• Successfully preserved critical neurophysiological markers, including 20Hz epilepsy-related spikes and parietal alpha suppression associated with depression.
• Constructed a latent space that effectively retains epileptic and depressive EEG features, highlighting spatial contributions from key channels like P8.

Conclusions:

• The proposed domain adaptation model effectively addresses domain shifts in EEG data, leading to enhanced diagnostic accuracy.
• The model's ability to preserve neurophysiological markers and construct a meaningful latent space offers a promising approach for robust EEG analysis in clinical settings.