HRV Feature selection related to mental state monitoring
Beings inherently sociable, human beings use emotions as fundamental ways to communicate. Humans need to express emotions in order to survive and create bonding among themselves. Social interaction is very related to emotion regulation abilities. In addition, self-reporting information is not so reliable and exact that can be used alone for research works. Whether healthy or not, there are some people who are not able to distinguish or express their feelings. Although particularly some disorders such as ADHD, Parkinson’s, and social-emotional agnosia directly effect on emotions.
When it comes to emotion, the ANS is viewed as a major component. Besides, ANS has a role of regulate the heart rate so HRV has gained a currency as a new method to assess ANS function.
Three aspects of human emotion are investigated in research works: arousal, valence and dominance [1]. In this study by using a real and a virtual museum just by HRV tracking to investigate cardiovascular dynamics during high and low arousing, it has been concluded that the accuracy of arousal recognition is 70.39%. the HRV signal has been extracted using Pan-Tompkins’ algorithm. Additionally, after detrending by smoothness prior method, Kubios HRV software is used to correct the ectopic beats of HRV. In real museum scenario 3 features of frequency, i.e., HF peak, HF power n.u. and HF power%, domain were the most dominant by using SVM. So vagal activity is assumed to be the most important item to determine arousal. Also, some nonlinear features including SampEn, ApEn and DFA α1 contributed to detection power. However, in virtual scenario classification of arousal only by using cardiovascular dynamics is challenging [1].
Virtual Reality (VR) has gained a currency in recent years in different fields. Among this fields are treatment applications such as social phobias and also as an emotion elicitation method. Because it can generate environments similar to the real world.
Although HRV is a non-invasive tool to assess cardiovascular dynamics regulation by ANS and different emotions like physical or mental stress affects the function of ANS, it has been shown that HRV is not adequate to achieve the best classification [2]. There are some other physiological signals which are impressed by ANS including EDA, PPG and skin temperature (SKT). Combining the time, frequency and non-linear features of these different biosignals could lead in the best result of stress level examination [2]. Accordingly, in this study by using Kernel-based Extreme Learning Machine (K-ELM) classification method, it is reported that in order to discriminate different levels of stress, these features acquired great accuracy of mean 95.3% for 5 levels of stress.
Many researches have investigated the emotion monitoring and prediction from HRV, most of which have utilized showing video clips as emotion stimulation [3].
After calculation of power spectral density, HF (High Frequency) and LF (Low Frequency) components are obtained. Also, SDNN (Standard Deviation of NN intervals), SDSD (Standard Deviation of RR interval differences), RMSSD (Root Mean Square of successive RR interval differences) and PNN50 (The percentage of adjacent of NN intervals that differ from each other by more than 50 ms).
[4] The goal of this study was to see how virtual reality exergaming experiences are linked to autonomic nervous system responses as an engaging alternative to traditional physical activity workouts. They used an open-source Python library for preprocessing, and Kubios software for HRV analysis. The mean RR interval, mean HR, SDNN, and RMSSD were included as time domain variables. The LF and HF components were chosen for frequency domain analysis. The stress index (SI) of Belsky was also evaluated.
To identify significant differences in HRV measurements between different conditions, a one-way analysis of variance (ANOVA) was used separately. The ratio LF/HF increased twice as much in in the game playing group as it did in the control group, indicating sympathetic activation.
The stimulation of the parasympathetic branch of the ANS was most noticeable at lower difficulty levels. Higher-intensity gaming, on the other hand, was found to cause less parasympathetic recovery during post-exercise and resting times.
[1] The goal of this research is to create an emotion detection system that can automatically distinguish between affective states elicited by an immersive virtual environment.
The artefacts were removed using the Kubios software’s threshold base artefacts correction technique. The QRS detection technique developed by Pan and Tompkins was employed. The smoothness before detrending method was used to extract the distinct trend components.
Mean RR, STDN, RMSSD, PNN50, TINN, RR triangular index, total power, VLF, LF, HF, approximate entropy, sample entropy, DFA, correlation dimension, Pointcaré SD1, and Pointcaré SD2 features were extracted. They used PCA to reduce the features from 29 to 3 and then classified them by SVM recursive feature elimination.
[6] studied physiological responses to a stressful virtual reality environment. RMSSD, HR, and LF/HF ratio features were extracted. The results showed that RMSSD was increased, and the LF-HF ratio was decreased in the stressor group compared to the control group. It is unclear why they observe this pattern of responses!
In this study, [5], the following features were extracted for categorizing emotions into two-dimensional states (positive and negative): mean RR, standard deviation RR, ratio (mean RR/STD RR), VLF, HF, LF, ratio LF/HF, mean magnitude, phase entropy, normalized bispectral entropy, normalized bispectral squared entropy, sum of logarithmic bispectral amplitude, non-linear sympatho-vagal interaction, min and max and mean of wavelet. Six features coming from Wavelet were used.
References
[1] J. Marín-Morales et al., “Affective computing in virtual reality: emotion recognition from brain and heartbeat dynamics using wearable sensors,” Scientific reports, vol. 8, no. 1, pp. 1-15, 2018.
[2] D. Cho et al., “Detection of stress levels from biosignals measured in virtual reality environments using a kernel-based extreme learning machine,” Sensors, vol. 17, no. 10, p. 2435, 2017.
[3] J. W. Yee Chung, H. C. Fuk So, M. M. Tak Choi, V. C. Man Yan, and T. K. Shing Wong, “Artificial Intelligence in education: Using heart rate variability (HRV) as a biomarker to assess emotions objectively,” Computers and Education: Artificial Intelligence, vol. 2, p. 100011, 2021/01/01/ 2021.
[4] M. J. Rodrigues, O. Postolache, and F. Cercas, “Autonomic Nervous System Assessment Based on HRV Analysis During Virtual Reality Serious Games,” in Computational Collective Intelligence, Cham, 2021, pp. 756-768: Springer International Publishing.
[5] Z. Cheng, L. Shu, J. Xie, and C. L. P. Chen, “A novel ECG-based real-time detection method of negative emotions in wearable applications,” in 2017 International Conference on Security, Pattern Analysis, and Cybernetics (SPAC), 2017, pp. 296-301.
[6] M. A. Martens, A. Antley, D. Freeman, M. Slater, P. J. Harrison, and E. M. Tunbridge, “It feels real: physiological responses to a stressful virtual reality environment and its impact on working memory,” Journal of Psychopharmacology, vol. 33, no. 10, pp. 1264-1273, 2019.
