Stress Monitoring Using Wearable Sensors: A Pilot Study and Stress-Predict Dataset

Examining the challenges of blood pressure estimation via photoplethysmogram

The use of observed wearable sensor data (e.g., photoplethysmograms [PPG]) to infer health measures (e.g., glucose level or blood pressure) is a very active area of research. Such technology can have a significant impact on health screening, chronic disease management and remote monitoring. A common approach is to collect sensor data and corresponding labels from a clinical grade device (e.g., blood pressure cuff) and train deep learning models to map one to the other. Although well intentioned, this approach often ignores a principled analysis of whether the input sensor data have enough information to predict the desired metric. We analyze the task of predicting blood pressure from PPG pulse wave analysis. Our review of the prior work reveals that many papers fall prey to data leakage and unrealistic constraints on the task and preprocessing steps. We propose a set of tools to help determine if the input signal in question (e.g., PPG) is indeed a good predictor of the desired label (e.g., blood pressure). Using our proposed tools, we found that blood pressure prediction using PPG has a high multi-valued mapping factor of 33.2% and low mutual information of 9.8%. In comparison, heart rate prediction using PPG, a well-established task, has a very low multi-valued mapping factor of 0.75% and high mutual information of 87.7%. We argue that these results provide a more realistic representation of the current progress toward the goal of wearable blood pressure measurement via PPG pulse wave analysis. For code, see our project page: https://github.com/lirus7/PPG-BP-Analysis.

Wearable Data From Subjects Playing Super Mario, Taking University Exams, or Performing Physical Exercise Help Detect Acute Mood Disorder Episodes via Self-Supervised Learning: Prospective, Exploratory, Observational Study

BACKGROUND

Personal sensing, leveraging data passively and near-continuously collected with wearables from patients in their ecological environment, is a promising paradigm to monitor mood disorders (MDs), a major determinant of the worldwide disease burden. However, collecting and annotating wearable data is resource intensive. Studies of this kind can thus typically afford to recruit only a few dozen patients. This constitutes one of the major obstacles to applying modern supervised machine learning techniques to MD detection.

OBJECTIVE

In this paper, we overcame this data bottleneck and advanced the detection of acute MD episodes from wearables' data on the back of recent advances in self-supervised learning (SSL). This approach leverages unlabeled data to learn representations during pretraining, subsequently exploited for a supervised task.

METHODS

We collected open access data sets recording with the Empatica E4 wristband spanning different, unrelated to MD monitoring, personal sensing tasks-from emotion recognition in Super Mario players to stress detection in undergraduates-and devised a preprocessing pipeline performing on-/off-body detection, sleep/wake detection, segmentation, and (optionally) feature extraction. With 161 E4-recorded subjects, we introduced E4SelfLearning, the largest-to-date open access collection, and its preprocessing pipeline. We developed a novel E4-tailored transformer (E4mer) architecture, serving as the blueprint for both SSL and fully supervised learning; we assessed whether and under which conditions self-supervised pretraining led to an improvement over fully supervised baselines (ie, the fully supervised E4mer and pre-deep learning algorithms) in detecting acute MD episodes from recording segments taken in 64 (n=32, 50%, acute, n=32, 50%, stable) patients.

RESULTS

SSL significantly outperformed fully supervised pipelines using either our novel E4mer or extreme gradient boosting (XGBoost): n=3353 (81.23%) against n=3110 (75.35%; E4mer) and n=2973 (72.02%; XGBoost) correctly classified recording segments from a total of 4128 segments. SSL performance was strongly associated with the specific surrogate task used for pretraining, as well as with unlabeled data availability.

CONCLUSIONS

We showed that SSL, a paradigm where a model is pretrained on unlabeled data with no need for human annotations before deployment on the supervised target task of interest, helps overcome the annotation bottleneck; the choice of the pretraining surrogate task and the size of unlabeled data for pretraining are key determinants of SSL success. We introduced E4mer, which can be used for SSL, and shared the E4SelfLearning collection, along with its preprocessing pipeline, which can foster and expedite future research into SSL for personal sensing.

Personalized Stress Detection Using Biosignals from Wearables: A Scoping Review

Stress is a natural yet potentially harmful aspect of human life, necessitating effective management, particularly during overwhelming experiences. This paper presents a scoping review of personalized stress detection models using wearable technology. Employing the PRISMA-ScR framework for rigorous methodological structuring, we systematically analyzed literature from key databases including Scopus, IEEE Xplore, and PubMed. Our focus was on biosignals, AI methodologies, datasets, wearable devices, and real-world implementation challenges. The review presents an overview of stress and its biological mechanisms, details the methodology for the literature search, and synthesizes the findings. It shows that biosignals, especially EDA and PPG, are frequently utilized for stress detection and demonstrate potential reliability in multimodal settings. Evidence for a trend towards deep learning models was found, although the limited comparison with traditional methods calls for further research. Concerns arise regarding the representativeness of datasets and practical challenges in deploying wearable technologies, which include issues related to data quality and privacy. Future research should aim to develop comprehensive datasets and explore AI techniques that are not only accurate but also computationally efficient and user-centric, thereby closing the gap between theoretical models and practical applications to improve the effectiveness of stress detection systems in real scenarios.

Predicting stress in first-year college students using sleep data from wearable devices

Consumer wearables have been successful at measuring sleep and may be useful in predicting changes in mental health measures such as stress. A key challenge remains in quantifying the relationship between sleep measures associated with physiologic stress and a user's experience of stress. Students from a public university enrolled in the Lived Experiences Measured Using Rings Study (LEMURS) provided continuous biometric data and answered weekly surveys during their first semester of college between October-December 2022. We analyzed weekly associations between estimated sleep measures and perceived stress for participants (N = 525). Through mixed-effects regression models, we identified consistent associations between perceived stress scores and average nightly total sleep time (TST), resting heart rate (RHR), heart rate variability (HRV), and respiratory rate (ARR). These effects persisted after controlling for gender and week of the semester. Specifically, for every additional hour of TST, the odds of experiencing moderate-to-high stress decreased by 0.617 or by 38.3% (p<0.01). For each 1 beat per minute increase in RHR, the odds of experiencing moderate-to-high stress increased by 1.036 or by 3.6% (p<0.01). For each 1 millisecond increase in HRV, the odds of experiencing moderate-to-high stress decreased by 0.988 or by 1.2% (p<0.05). For each additional breath per minute increase in ARR, the odds of experiencing moderate-to-high stress increased by 1.230 or by 23.0% (p<0.01). Consistent with previous research, participants who did not identify as male (i.e., female, nonbinary, and transgender participants) had significantly higher self-reported stress throughout the study. The week of the semester was also a significant predictor of stress. Sleep data from wearable devices may help us understand and to better predict stress, a strong signal of the ongoing mental health epidemic among college students.

Advances in electrochemical biosensor design for the detection of the stress biomarker cortisol

The monitoring of stress levels in humans has become increasingly relevant, given the recent incline of stress-related mental health disorders, lifestyle impacts, and chronic physiological diseases. Long-term exposure to stress can induce anxiety and depression, heart disease, and risky behaviors, such as drug and alcohol abuse. Biomarker molecules can be quantified in biological fluids to study human stress. Cortisol, specifically, is a hormone biomarker produced in the adrenal glands with biofluid concentrations that directly correlate to stress levels in humans. The rapid, real-time detection of cortisol is necessary for stress management and predicting the onset of psychological and physical ailments. Current methods, including mass spectrometry and immunoassays, are effective for sensitive cortisol quantification. However, these techniques provide only single measurements which pose challenges in the continuous monitoring of stress levels. Additionally, these analytical methods often require trained personnel to operate expensive instrumentation. Alternatively, low-cost electrochemical biosensors enable the real-time detection and continuous monitoring of cortisol levels while also providing adequate analytical figures of merit (e.g., sensitivity, selectivity, sensor response times, detection limits, and reproducibility) in a simple design platform. This review discusses the recent developments in electrochemical biosensor design for the detection of cortisol in human biofluids. Special emphasis is given to biosensor recognition elements, including antibodies, molecularly imprinted polymers (MIPs), and aptamers, as critical components of electrochemical biosensors for cortisol detection. Furthermore, the advantages and limiting factors of various electrochemical techniques and sensing in complex biofluid matrices are overviewed. Remarks on the current challenges and future perspectives regarding electrochemical biosensors for stress monitoring are provided, including matrix effects (pH dependence and biological interferences), wearability, and large-scale production.

Mental Health for All: The Case for Investing in Digital Mental Health to Improve Global Outcomes, Access, and Innovation in Low-Resource Settings

Mental health disorders are an increasing global public health concern that contribute to morbidity, mortality, disability, and healthcare costs across the world. Biomedical and psychological research has come a long way in identifying the importance of mental health and its impact on behavioral risk factors, physiological health, and overall quality of life. Despite this, access to psychological and psychiatric services remains widely unavailable and is a challenge for many healthcare systems, particularly those in developing countries. This review article highlights the strengths and opportunities brought forward by digital mental health in narrowing this divide. Further, it points to the economic and societal benefits of effectively managing mental illness, making a case for investing resources into mental healthcare as a larger priority for large non-governmental organizations and individual nations across the globe.

Generalizable machine learning for stress monitoring from wearable devices: A systematic literature review

INTRODUCTION

Wearable sensors have shown promise as a non-intrusive method for collecting biomarkers that may correlate with levels of elevated stress. Stressors cause a variety of biological responses, and these physiological reactions can be measured using biomarkers including Heart Rate Variability (HRV), Electrodermal Activity (EDA) and Heart Rate (HR) that represent the stress response from the Hypothalamic-Pituitary-Adrenal (HPA) axis, the Autonomic Nervous System (ANS), and the immune system. While Cortisol response magnitude remains the gold standard indicator for stress assessment [1], recent advances in wearable technologies have resulted in the availability of a number of consumer devices capable of recording HRV, EDA and HR sensor biomarkers, amongst other signals. At the same time, researchers have been applying machine learning techniques to the recorded biomarkers in order to build models that may be able to predict elevated levels of stress.

OBJECTIVE

The aim of this review is to provide an overview of machine learning techniques utilized in prior research with a specific focus on model generalization when using these public datasets as training data. We also shed light on the challenges and opportunities that machine learning-enabled stress monitoring and detection face.

METHODS

This study reviewed published works contributing and/or using public datasets designed for detecting stress and their associated machine learning methods. The electronic databases of Google Scholar, Crossref, DOAJ and PubMed were searched for relevant articles and a total of 33 articles were identified and included in the final analysis. The reviewed works were synthesized into three categories of publicly available stress datasets, machine learning techniques applied using those, and future research directions. For the machine learning studies reviewed, we provide an analysis of their approach to results validation and model generalization. The quality assessment of the included studies was conducted in accordance with the IJMEDI checklist [2].

RESULTS

A number of public datasets were identified that are labeled for stress detection. These datasets were most commonly produced from sensor biomarker data recorded using the Empatica E4 device, a well-studied, medical-grade wrist-worn wearable that provides sensor biomarkers most notable to correlate with elevated levels of stress. Most of the reviewed datasets contain less than twenty-four hours of data, and the varied experimental conditions and labeling methodologies potentially limit their ability to generalize for unseen data. In addition, we discuss that previous works show shortcomings in areas such as their labeling protocols, lack of statistical power, validity of stress biomarkers, and model generalization ability.

CONCLUSION

Health tracking and monitoring using wearable devices is growing in popularity, while the generalization of existing machine learning models still requires further study, and research in this area will continue to provide improvements as newer and more substantial datasets become available.

Wrist-Based Electrodermal Activity Monitoring for Stress Detection Using Federated Learning

With the most recent developments in wearable technology, the possibility of continually monitoring stress using various physiological factors has attracted much attention. By reducing the detrimental effects of chronic stress, early diagnosis of stress can enhance healthcare. Machine Learning (ML) models are trained for healthcare systems to track health status using adequate user data. Insufficient data is accessible, however, due to privacy concerns, making it challenging to use Artificial Intelligence (AI) models in the medical industry. This research aims to preserve the privacy of patient data while classifying wearable-based electrodermal activities. We propose a Federated Learning (FL) based approach using a Deep Neural Network (DNN) model. For experimentation, we use the Wearable Stress and Affect Detection (WESAD) dataset, which includes five data states: transient, baseline, stress, amusement, and meditation. We transform this raw dataset into a suitable form for the proposed methodology using the Synthetic Minority Oversampling Technique (SMOTE) and min-max normalization pre-processing methods. In the FL-based technique, the DNN algorithm is trained on the dataset individually after receiving model updates from two clients. To decrease the over-fitting effect, every client analyses the results three times. Accuracies, Precision, Recall, F1-scores, and Area Under the Receiver Operating Curve (AUROC) values are evaluated for each client. The experimental result shows the effectiveness of the federated learning-based technique on a DNN, reaching 86.82% accuracy while also providing privacy to the patient's data. Using the FL-based DNN model over a WESAD dataset improves the detection accuracy compared to the previous studies while also providing the privacy of patient data.

Improved Stress Classification Using Automatic Feature Selection from Heart Rate and Respiratory Rate Time Signals

Time-series features are the characteristics of data periodically collected over time. The calculation of time-series features helps in understanding the underlying patterns and structure of the data, as well as in visualizing the data. The manual calculation and selection of time-series feature from a large temporal dataset are time-consuming. It requires researchers to consider several signal-processing algorithms and time-series analysis methods to identify and extract meaningful features from the given time-series data. These features are the core of a machine learning-based predictive model and are designed to describe the informative characteristics of the time-series signal. For accurate stress monitoring, it is essential that these features are not only informative but also well-distinguishable and interpretable by the classification models. Recently, a lot of work has been carried out on automating the extraction and selection of times-series features. In this paper, a correlation-based time-series feature selection algorithm is proposed and evaluated on the stress-predict dataset. The algorithm calculates a list of 1578 features of heart rate and respiratory rate signals (combined) using the tsfresh library. These features are then shortlisted to the more specific time-series features using Principal Component Analysis (PCA) and Pearson, Kendall, and Spearman correlation ranking techniques. A comparative study of conventional statistical features (like, mean, standard deviation, median, and mean absolute deviation) versus correlation-based selected features is performed using linear (logistic regression), ensemble (random forest), and clustering (k-nearest neighbours) predictive models. The correlation-based selected features achieved higher classification performance with an accuracy of 98.6% as compared to the conventional statistical feature’s 67.4%. The outcome of the proposed study suggests that it is vital to have better analytical features rather than conventional statistical features for accurate stress classification.