[Multi-source adversarial adaptation with calibration for electroencephalogram-based classification of meditation and...

Area of Science:

• Neuroscience
• Biomedical Informatics
• Machine Learning

Context:

• Meditation shows therapeutic potential, with electroencephalogram (EEG) changes during meditation suggesting feasibility for deep learning monitoring.
• Inter-subject variability in EEG signals presents a significant challenge for accurate meditation state monitoring systems.
• Existing methods struggle to generalize across different individuals due to unique EEG patterns.

Purpose:

• To develop a novel model-calibrated multi-source adversarial adaptation network (CMAAN) for robust EEG-based meditation monitoring.
• To address the challenge of inter-subject differences in EEG signals for improved classification accuracy.
• To enhance the performance and generalizability of deep learning models for analyzing meditation states.

Summary:

• A novel model-calibrated multi-source adversarial adaptation network (CMAAN) was proposed, training multiple domain-adversarial neural networks pairwise between subjects.
• A calibration process using limited target-domain labeled data integrated these networks, enhancing performance.
• The CMAAN model achieved 73.09% classification accuracy on an EEG dataset from 18 subjects in methamphetamine rehabilitation, identifying key EEG frequency bands and brain regions.

Impact:

• The CMAAN framework significantly improves the performance and robustness of EEG-based meditation monitoring systems.
• This advancement holds substantial promise for applications in biomedical informatics and clinical practice, particularly in rehabilitation settings.
• The study provides insights into the neural correlates of meditation by analyzing frequency bands and brain regions involved.