Characterization of heart rate variability signal for distinction of meditative and pre-meditative states. (April 2021)

• Record Type: Journal Article
• Authors: Deka, Dipen; Deka, Bhabesh
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Heart rate variability -- Meditation -- Yoga -- Empirical mode decomposition -- Nonlinear dynamics -- Support vector machine; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102414
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We propose a data augmentation technique to handle short HRV data. A time-frequency-nonlinear analysis is conducted to characterize HRV signal during meditation. A new parameter SDSOD is proposed for HRV analysis. We found turning point count, which was earlier used to detect random signal, a very promising HRV feature. Classification performance of the Bayesian optimized SVM model: ACC = 95.31% and F1 Score = 0.952. Abstract: Meditation and yoga have growing popularity in recent times due to its effectiveness in reducing stress, anxiety, and elevating overall well-being. To study its exertion on human being, HRV analysis is found to be an appropriate tool. The results of meditation may vary heterogeneously for different subjects depending on experience, expertise of performance, besides the disparity in internal and external dynamic interactions. It poses the challenge to extract discriminating features and robust classifier for separating the meditative state from the non-meditative state. In this paper, we have proposed a new parameter, named as the standard deviation of second order differences of RR intervals to capture the underlying dynamics of HRV during meditation. Besides, we have selected 8 more discriminating and non-redundant parameters based on significance test and correlation coefficient measures to classify the two states. Considering the small data size, the SVM classification model is cross-validated using leave-one-subject's-one-state-out cross-validation (LOSOSOCV). Hyperparameters of the SVM model are chosen for each subject's each state (meditative/pre-meditative) using the Bayesian optimizer. The proposed approach provides a classification accuracy of 95.31% and F1 score of 0.9523. Furthermore, this study has also demonstrated the analysis of findings from different parameters in capturing the underlying dynamics of HRV during the practice of meditation.