Signal processing and machine learning algorithms for stress monitoring using wearable sensor technologies
Date
2023-03-30Author
Iqbal, Talha
Metadata
Show full item recordUsage
This item's downloads: 599 (view details)
Abstract
In recent years, there has been a notable increase in depression, anxiety, pathological stress,
and other stress-related diseases. Stress is a known contributor to several life-threatening
medical conditions and triggers acute cardiovascular events, as well as one of the root causes
of several social problems. According to the statistics from the World Health Organization,
stress is associated with several medical and social problems, and these problems are seriously
affecting the health and well-being of not only adults but also children and youngsters. The
recent development of miniaturized and flexible biosensors has enabled the development of
connected wearable solutions to monitor stress and intervene in time to prevent the
progression of stress-induced medical conditions. Therefore, a vast interest has been
developed to investigate the underlying mechanisms of stress and monitor various
biophysiological and biochemical responses of the body to stress. The review of the literature
on different physiological and chemical indicators of stress, which are commonly used for
quantitative assessment of stress, and the associated sensing technologies shows that
prolonged exposure to stress triggers the adrenocorticotrophic hormonal (ACTH) system
and causes the release of cortisol hormones from the adrenal cortex that boosts the alertness
of the body. As a result, there is an increase in blood supply to muscles, heart rate, respiratory
rate, and cognitive activity, along with several other responses. The variable and
contradictory evidence in the literature on the use of either physiological or biochemical
stress markers leads to the conclusion that neither of these biomarkers in isolation can
provide sufficient means of monitoring stress. Therefore, a combination of physiological and
chemical stress biomarkers, with contextual information, can be a more reliable solution for
stress monitoring. The current standard for stress evaluation is based on self-reported questionnaires and
standardized stress scores. There is no gold standard to independently evaluate stress levels
despite the availability of numerous biophysiological stress indicators. Moreover, there is no
clear understanding of the relative sensitivity and specificity of these stress-related
biophysiological indicators of stress in the literature. An extensive statistical analysis and
classification modelling of biophysiological data gathered from healthy individuals,
undergoing various induced emotional states was performed to assess the relative sensitivity
and specificity of common biophysiological indicators of stress. The key indicators of stress,
such as heart rate, respiratory rate, skin conductance, RR interval, heart rate variability in the electrocardiogram, and muscle activation measured by electromyography, are evaluated as
gathered from an already existing, publicly available WESAD (Wearable Stress and Affect
Detection) dataset. Respiratory rate and heart rate were the two best features for
distinguishing between stressed and unstressed states. Both parameters can be estimated
using a single photoplethysmography (PPG) sensor. The heart rate is estimated by counting
the number of peaks in the PPG signal. Most of the existing algorithms for the estimation
of respiratory rate using photoplethysmography (PPG) are sensitive to external noise and
may require the selection of certain algorithm-specific parameters, through the trial-and-error
method. Thus, a new algorithm to estimate the respiratory rate using a
photoplethysmography sensor signal for health monitoring is proposed. The algorithm is
resistant to signal loss and can handle low-quality signals from the sensor. The results endorse
that integration of the proposed algorithm into a commercially available pulse oximetry
device would expand its functionality from the measurement of oxygen saturation level and
heart rate to the continuous measurement of the respiratory rate with good efficiency at
home and in a clinical setting.
Additionally, as the public availability of datasets for the development of stress monitoring
devices is limited, a clinical study was performed. The dataset created is an open-access
dataset named Stress-Predict dataset. The inclusion of an additional feature, i.e., respiratory
rate data along with stress and baseline labels within the dataset, makes the dataset more
desirable and unique from all the other publicly available Empatica E4-based datasets. The
dataset and outcomes of this study contribute to understanding any accuracy gaps in current
stress monitoring and help improve these technologies or develop new technologies for
stress monitoring. Most wearable stress monitoring systems are built on a supervised learning
classification algorithm trained on simple statistical features. For accurate stress monitoring,
it is essential that these features are not only informative but also well-distinguishable and
interpretable by the classification models. Thus, a correlation-based time-series feature
selection algorithm is proposed and evaluated on the stress-predict dataset. The outcome of
the study suggests that it is vital to have better analytical features rather than conventional
statistical features for accurate stress classification.
One of the most challenging tasks in physiological or pathological stress monitoring is the
labelling of the physiological signals collected during an experiment. Commonly, different
types of self-reporting questionnaires are used to label the perceived stress instances. These
questionnaires only capture stress levels at a specific point in time. Moreover, self-reporting
is subjective and prone to inaccuracies. Traditional supervised machine learning classifiers
require hand-crafted features and labels while on the contrary, the unsupervised classifier
does not require any labels of perceived stress levels and performs classification based on
clustering algorithms. The analysis and results of this comparative study demonstrate the
potential of unsupervised learning for the development of non-invasive, continuous, and
robust detection and monitoring of physiological and pathological stress.