Stress is a term that can be related to both body and mind. Stress is essential for survival, but the modern-day lifestyle has led to stress becoming a major health concern. Stress is the leading contributor for heart attack, diabetes, depression. However, it is not feasible to monitor the stress levels continuously. Hence, we need a device that will keep track of stress over weeks/months. We don’t want the device to be conspicuous and for the user to be always aware of the device. The existing method such as electrodermal activity needs to place electrodes on the fingers, which is very obtrusive and not practical. Blood pressure measurement is obtrusive(cuffs) or invasive (needle in the artery). This paper tries to detect mental stress using unobtrusive wearables.
The unobtrusive wearable is a Heart Rate Monitor (HRM). This paper tries to estimate the state of autonomic nervous system. Autonomic control can one-to-one map onto psychological states. There is a need for decoupling SNS and PNS because, Beck’s work proved that there could be one to many relationships between psychophysiological conditions considering only heart rate. But separating the autonomic branches help in creating a mark. The existing methods assumes that SNS and PNS exert independent influences on HRV and tries to find the contributions from each independent component. But the assumption is the problem. And other methods that use blood pressure are pharmacological blockades.Proposed Solution:The Autonomic Nervous System (ANS) contains two branches, Sympathetic Nervous System (SNS) and Parasympathetic Nervous System (PNS). The PNS brings the body to rest state where as the SNS is used in response to threats. SNS increases the heart rate and PNS decreases heart rate. But it is difficult to find if heart rate change is because of SNS increase or PNS decrease. But, by analyzing beat period we can find Heart Rate Variability (HRV) we can differentiate SNS and PNS. Principal Dynamic Models (PDM), a nonlinear system identification technique is used to predict activation levels of 2 autonomic branches. The first two PDMs had similar spectral characteristics as that of SNS and PNS activity. We discard the eigenvalues whose contribution to the total energy is less than 5%. SNS eigenvalues are considered negative and PNS eigenvalues are considered positive, because SNS increases the heart rate and reduces the heart period.
Feasibility of HRV from heart rate monitor and how it can be used to correlate to stress has been validated experimentally. A wearable that was non-obtrusive was used, it was also modular, open, power efficient and cheap. Heart period measurements were made on three subjects with four experiments, two of them caused mental load and others mental relaxation. Heart signals were sampled, and peak detection algorithms were used to find the R waves and this was used to find the R-R period.
Results/Findings:HRV monitors were found to be consistent. Heart Rate Measurements (HRM) was determined using the HRV monitor as well as 3-lead ECG Monitor for two experimental conditions, one which is inclining the body by 70 degree so that causes PNS to be high and other in upright position so that SNS is high.
There was a good correlation between the two methods. Detection of Mental Stress was considered a binary classification problem. The goal was to differentiate between the mental load experiment from the mental relaxation experiment. The classifier was initially trained for a subject for four days and made to predict on the fifth day. The classifier was also trained with two subject’s data and experimented on the third subject to correctly differentiate. It was found that PDM features are more stable but less subject dependent than spectral features. But spectral features lead to higher classification performance. Critique:I feel that at the time this paper was written there was extensive works going on wearables. This paper is very critical because there were so many deaths that was caused by heart attacks and they were directly linked to stress. I will take better care of my body, but I need to know that my body is getting stressed. But for that matter I would not prefer to be constantly monitored by something that will affect my day to day activity. The idea of using wearable to solve this case seems to be like an age-old solution now, but I think at 2009 it must have been a novel idea. The analysis of the existing problems with heart rate measurement is something that I liked a lot. I feel that the problem statement was well defined. I also liked the use of artificial intelligence to detect mental stress given the R-R period.
The classification experiment was something that I would have liked to be a part of, to see how the experiment is going.Ideas for follow-up on work:I think the work in this field of wearables have saturated in the present days. But still for improvement purpose, we can try to work on the backend which is the wearable embedded system and the front-end which is the algorithm to classify mental stress. The backend can us the modern-day advancements to incorporate more artificial intelligence features and also can be made more power efficient. The front end can use more advanced classification than a simple binary classification. Although, these can be done, there is a need to evaluate the benefits reaped given that the prices of the wearable will go up because of using high computing architectures.