A new neurofeedback training method based on feature space clustering to control EEG features within target clusters

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Traditional neurofeedback methods rely on predefined thresholds for electroencephalogram (EEG) feature assessment.
• These threshold-based methods may lack precision in evaluating a subject's status during training.

Purpose of the Study:

• To propose and evaluate a new neurofeedback training method utilizing feature-space clustering for enhanced accuracy in assessing subject status.
• To compare the efficacy of soft-boundary and hard-boundary approaches within the proposed clustering framework.

Main Methods:

• Employed the neural gas algorithm for feature-space clustering of EEG data.
• Defined initial and target clusters, with a scoring index based on subjects' brain activity.
• Simulated the method in two versions: soft-boundary and hard-boundary.

Main Results:

• The hard-boundary version successfully guided subjects towards target cluster boundaries and stabilized their status amidst EEG fluctuations.
• The soft-boundary version guided subjects toward target boundaries without over-encouragement, even with significant feature changes.

Conclusions:

• The hard-boundary neurofeedback approach is effective for guiding subjects within and potentially beyond target EEG feature clusters.
• The soft-boundary approach offers utility for controlling EEG features within specified limits, enhancing training adaptability.