EEG Spectral Correlates of Rapid and Deep Slow Breathing States and classification using ML

Abstract

One interpretation of breathing exercise is to enforce mind-body harmony, when someone feels well and healthy, different organs of our body function harmoniously. One dysfunctional organ may disturb the resonating mechanism across multiple organs. There are different breathing techniques, and recent scientific evidence encourages understanding the neural correlates of breathing. This research investigates breathing exercises at two paces: Rapid and Deep Slow using neural signals. We collect Electroencephalography (EEG) recordings of 14 participants performing breathing tasks. EEG signals are primarily decomposed in frequency bands that designate different cognitive functions. We extract six primary frequency bands, including delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), low beta (13-20 Hz), high beta (21-30 Hz), and gamma (30-40 Hz). Two different techniques are utilized to report the findings encompassing power spectral analysis and employing machine learning classifiers to discriminate features among different stages of inhalation and exhalation with the significance of different frequencies bands. Lowered beta power in Slow Deep breathing is observed compared to Rapid Breathing, which may suggest increased relaxation, calmness, and anxiety reduction. Differences between the two conditions observed in the frontoparietal cortex may be attributed to differences in voluntary control between the two tasks. We observed classification accuracy of 72 % using low beta between Rapid and Deep Slow breathing using Decision Tree. Several interesting findings are observed in different scalp regions suggesting future direction for further investigation. This study contributes to the understanding of neural signatures for different breathing practices. The implication of this research in health care is to design personalized therapies and to design better breathing mobile applications for daily use.