Ambulatory and Laboratory Stress Detection Based on Raw Electrocardiogram Signals Using a Convolutional Neural Network

Relieving the burden of intensive labeling for stress monitoring in the wild by using semi-supervised learning

Stress, a natural process affecting individuals' wellbeing, has a profound impact on overall quality of life. Researchers from diverse fields employ various technologies and methodologies to investigate it and alleviate the negative effects of this phenomenon. Wearable devices, such as smart bands, capture physiological data, including heart rate variability, motions, and electrodermal activity, enabling stress level monitoring through machine learning models. However, labeling data for model accuracy assessment poses a significant challenge in stress-related research due to incomplete or inaccurate labels provided by individuals in their daily lives. To address this labeling predicament, our study proposes implementing Semi-Supervised Learning (SSL) models. Through comparisons with deep learning-based supervised models and clustering-based unsupervised models, we evaluate the performance of our SSL models. Our experiments show that our SSL models achieve 77% accuracy with a classifier trained on an augmented dataset prepared using the label propagation (LP) algorithm. Additionally, our deep autoencoder network achieves 76% accuracy. These results highlight the superiority of SSL models over unsupervised learning techniques and their comparable performance to supervised learning models, even with limited labeled data. By relieving the burden of labeling in daily life stress recognition, our study advances stress-related research, recognizing stress as a natural process rather than a disease. This facilitates the development of more efficient and accurate stress monitoring methods in the wild.

Electrocardiogram Monitoring Wearable Devices and Artificial-Intelligence-Enabled Diagnostic Capabilities: A Review

Worldwide, population aging and unhealthy lifestyles have increased the incidence of high-risk health conditions such as cardiovascular diseases, sleep apnea, and other conditions. Recently, to facilitate early identification and diagnosis, efforts have been made in the research and development of new wearable devices to make them smaller, more comfortable, more accurate, and increasingly compatible with artificial intelligence technologies. These efforts can pave the way to the longer and continuous health monitoring of different biosignals, including the real-time detection of diseases, thus providing more timely and accurate predictions of health events that can drastically improve the healthcare management of patients. Most recent reviews focus on a specific category of disease, the use of artificial intelligence in 12-lead electrocardiograms, or on wearable technology. However, we present recent advances in the use of electrocardiogram signals acquired with wearable devices or from publicly available databases and the analysis of such signals with artificial intelligence methods to detect and predict diseases. As expected, most of the available research focuses on heart diseases, sleep apnea, and other emerging areas, such as mental stress. From a methodological point of view, although traditional statistical methods and machine learning are still widely used, we observe an increasing use of more advanced deep learning methods, specifically architectures that can handle the complexity of biosignal data. These deep learning methods typically include convolutional and recurrent neural networks. Moreover, when proposing new artificial intelligence methods, we observe that the prevalent choice is to use publicly available databases rather than collecting new data.

Applications of artificial intelligence-machine learning for detection of stress: a critical overview

Psychological distress is a major contributor to human physiology and pathophysiology, and it has been linked to several conditions, such as auto-immune diseases, metabolic syndrome, sleep disorders, and suicidal thoughts and inclination. Therefore, early detection and management of chronic stress is crucial for the prevention of several diseases. Artificial intelligence (AI) and Machine Learning (ML) have promoted a paradigm shift in several areas of biomedicine including diagnosis, monitoring, and prognosis of disease. Here, our review aims to present some of the AI and ML applications for solving biomedical issues related to psychological stress. We provide several lines of evidence from previous studies highlighting that AI and ML have been able to predict stress and detect the brain normal states vs. abnormal states (notably, in post-traumatic stress disorder (PTSD)) with accuracy around 90%. Of note, AI/ML-driven technology applied to identify ubiquitously present stress exposure may not reach its full potential, unless future analytics focus on detecting prolonged distress through such technology instead of merely assessing stress exposure. Moving forward, we propose that a new subcategory of AI methods called Swarm Intelligence (SI) can be used towards detecting stress and PTSD. SI involves ensemble learning techniques to efficiently solve a complex problem, such as stress detection, and it offers particular strength in clinical settings, such as privacy preservation. We posit that AI and ML approaches will be beneficial for the medical and patient community when applied to predict and assess stress levels. Last, we encourage additional research to bring AI and ML into the standard clinical practice for diagnostics in the not-too-distant future.

An Innovative Random-Forest-Based Model to Assess the Health Impacts of Regular Commuting Using Non-Invasive Wearable Sensors

Regular commutes to work can cause chronic stress, which in turn can cause a physical and emotional reaction. The recognition of mental stress in its earliest stages is very necessary for effective clinical treatment. This study investigated the impact of commuting on human health based on qualitative and quantitative measures. The quantitative measures included electroencephalography (EEG) and blood pressure (BP), as well as weather temperature, while qualitative measures were established from the PANAS questionnaire, and included age, height, medication, alcohol status, weight, and smoking status. This study recruited 45 (n) healthy adults, including 18 female and 27 male participants. The modes of commute were bus (n = 8), driving (n = 6), cycling (n = 7), train (n = 9), tube (n = 13), and both bus and train (n = 2). The participants wore non-invasive wearable biosensor technology to measure EEG and blood pressure during their morning commute for 5 days in a row. A correlation analysis was applied to find the significant features associated with stress, as measured by a reduction in positive ratings in the PANAS. This study created a prediction model using random forest, support vector machine, naive Bayes, and K-nearest neighbor. The research results show that blood pressure and EEG beta waves were significantly increased, and the positive PANAS rating decreased from 34.73 to 28.60. The experiments revealed that measured systolic blood pressure was higher post commute than before the commute. For EEG waves, the model shows that the EEG beta low power exceeded alpha low power after the commute. Having a fusion of several modified decision trees within the random forest helped increase the performance of the developed model remarkably. Significant promising results were achieved using random forest with an accuracy of 91%, while K-nearest neighbor, support vector machine, and naive Bayes performed with an accuracy of 80%, 80%, and 73%, respectively.

A systematic review and Meta-data analysis on the applications of Deep Learning in Electrocardiogram

The success of deep learning over the traditional machine learning techniques in handling artificial intelligence application tasks such as image processing, computer vision, object detection, speech recognition, medical imaging and so on, has made deep learning the buzz word that dominates Artificial Intelligence applications. From the last decade, the applications of deep learning in physiological signals such as electrocardiogram (ECG) have attracted a good number of research. However, previous surveys have not been able to provide a systematic comprehensive review including biometric ECG based systems of the applications of deep learning in ECG with respect to domain of applications. To address this gap, we conducted a systematic literature review on the applications of deep learning in ECG including biometric ECG based systems. The study analyzed systematically, 150 primary studies with evidence of the application of deep learning in ECG. The study shows that the applications of deep learning in ECG have been applied in different domains. We presented a new taxonomy of the domains of application of the deep learning in ECG. The paper also presented discussions on biometric ECG based systems and meta-data analysis of the studies based on the domain, area, task, deep learning models, dataset sources and preprocessing methods. Challenges and potential research opportunities were highlighted to enable novel research. We believe that this study will be useful to both new researchers and expert researchers who are seeking to add knowledge to the already existing body of knowledge in ECG signal processing using deep learning algorithm.

Supplementary information

The online version contains supplementary material available at 10.1007/s12652-022-03868-z.

Real-Time Stress Level Feedback from Raw Ecg Signals for Personalised, Context-Aware Applications Using Lightweight Convolutional Neural Network Architectures

Human stress is intricately linked with mental processes such as decision making. Public protection practitioners, including Law Enforcement Agents (LEAs), are forced to make difficult decisions during high-pressure operations, under strenuous circumstances. In this respect, systems and applications that assist such practitioners to take decisions, are increasingly incorporating user stress level information for their development, adaptation, and evaluation. To that end, our goal is to accurately detect and classify the level of acute, short-term stress, in real time, for the development of personalized, context-aware solutions for LEAs. Deep Neural Networks (DNNs), and in particular Convolutional Neural Networks (CNNs), have been gaining traction in the field of stress analysis, exhibiting promising results. Furthermore, the electrocardiogram (ECG) signals, have also been widely adopted for estimating levels of stress. In this work, we propose two CNN architectures for the stress detection and 3-level (low, moderate, high) stress classification tasks, using ultra short-term raw ECG signals (3 s). One architecture is simple and with a low memory footprint, suitable for running in wearable edge-computing nodes, and the other is able to learn more complex features, having more trainable parameters. The models were trained on the two publicly available stress classification datasets, after applying pre-processing techniques, such as data pruning, down-sampling, and data augmentation, using a sliding window approach. After hyperparameter tuning, using 4-fold cross-validation, the evaluation on the test set demonstrated state-of-the-art accuracy both on the 3- and 2-level stress classification task using the DriveDB dataset, reporting an accuracy of 83.55% and 98.77% respectively.

Deep-Learning-Based Stress Recognition with Spatial-Temporal Facial Information

In recent times, as interest in stress control has increased, many studies on stress recognition have been conducted. Several studies have been based on physiological signals, but the disadvantage of this strategy is that it requires physiological-signal-acquisition devices. Another strategy employs facial-image-based stress-recognition methods, which do not require devices, but predominantly use handcrafted features. However, such features have low discriminating power. We propose a deep-learning-based stress-recognition method using facial images to address these challenges. Given that deep-learning methods require extensive data, we constructed a large-capacity image database for stress recognition. Furthermore, we used temporal attention, which assigns a high weight to frames that are highly related to stress, as well as spatial attention, which assigns a high weight to regions that are highly related to stress. By adding a network that inputs the facial landmark information closely related to stress, we supplemented the network that receives only facial images as the input. Experimental results on our newly constructed database indicated that the proposed method outperforms contemporary deep-learning-based recognition methods.

Trends in Heart-Rate Variability Signal Analysis

Heart rate variability (HRV) is the rate of variability between each heartbeat with respect to time. It is used to analyse the Autonomic Nervous System (ANS), a control system used to modulate the body's unconscious action such as cardiac function, respiration, digestion, blood pressure, urination, and dilation/constriction of the pupil. This review article presents a summary and analysis of various research works that analyzed HRV associated with morbidity, pain, drowsiness, stress and exercise through signal processing and machine learning methods. The points of emphasis with regards to HRV research as well as the gaps associated with processes which can be improved to enhance the quality of the research have been discussed meticulously. Restricting the physiological signals to Electrocardiogram (ECG), Electrodermal activity (EDA), photoplethysmography (PPG), and respiration (RESP) analysis resulted in 25 articles which examined the cause and effect of increased/reduced HRV. Reduced HRV was generally associated with increased morbidity and stress. High HRV normally indicated good health, and in some instances, it could signify clinical events of interest such as drowsiness. Effective analysis of HRV during ambulatory and motion situations such as exercise, video gaming, and driving could have a significant impact toward improving social well-being. Detection of HRV in motion is far from perfect, situations involving exercise or driving reported accuracy as high as 85% and as low as 59%. HRV detection in motion can be improved further by harnessing the advancements in machine learning techniques.

Classification of Mental Stress Using CNN-LSTM Algorithms with Electrocardiogram Signals

The mental stress faced by many people in modern society is a factor that causes various chronic diseases, such as depression, cancer, and cardiovascular disease, according to stress accumulation. Therefore, it is very important to regularly manage and monitor a person's stress. In this study, we propose an ensemble algorithm that can accurately determine mental stress states using a modified convolutional neural network (CNN)- long short-term memory (LSTM) architecture. When a person is exposed to stress, a displacement occurs in the electrocardiogram (ECG) signal. It is possible to classify stress signals by analyzing ECG signals and extracting specific parameters. To maximize the performance of the proposed stress classification algorithm, fast Fourier transform (FFT) and spectrograms were applied to preprocess ECG signals and produce signals in both the time and frequency domains to aid the training process. As the performance evaluation benchmarks of the stress classification model, confusion matrices, receiver operating characteristic (ROC) curves, and precision-recall (PR) curves were used, and the accuracy achieved by the proposed model was 98.3%, which is an improvement of 14.7% compared to previous research results. Therefore, our model can help manage the mental health of people exposed to stress. In addition, if combined with various biosignals such as electromyogram (EMG) and photoplethysmography (PPG), it may have the potential for development in various healthcare systems, such as home training, sleep state analysis, and cardiovascular monitoring.

Human stress classification during public speaking using physiological signals

Public speaking is a common type of social evaluative situation and a significant amount of the population feel uneasy with it. It is of utmost importance to detect public speaking stress so that appropriate action can be taken to minimize its impacts on human health. In this study, a multimodal human stress classification scheme in response to real-life public speaking activity is proposed. Electroencephalography (EEG), galvanic skin response (GSR), and photoplethysmography (PPG) signals of forty participants are acquired in rest-state and during public speaking activity to divide data into a stressed and non-stressed group. Frequency domain features from EEG and time-domain features from GSR and PPG signals are extracted. The selected set of features from all modalities are fused to classify the stress into two classes. Classification is performed via a leave-one-out cross-validation scheme by using five different classifiers. The highest accuracy of 96.25% is achieved using a support vector machine classifier with radial basis function. Statistical analysis is performed to examine the significance of EEG, GSR, and PPG signals between the two phases of the experiment. Statistical significance is found in certain EEG frequency bands as well as GSR and PPG data recorded before and after public speaking supporting the fact that brain activity, skin conductance, and blood volumetric flow are credible measures of human stress during public speaking activity.

Stress Classification Using Photoplethysmogram-Based Spatial and Frequency Domain Images

Stress is subjective and is manifested differently from one person to another. Thus, the performance of generic classification models that classify stress status is crude. Building a person-specific model leads to a reliable classification, but it requires the collection of new data to train a new model for every individual and needs periodic upgrades because stress is dynamic. In this paper, a new binary classification (called stressed and non-stressed) approach is proposed for a subject’s stress state in which the inter-beat intervals extracted from a photoplethysomogram (PPG) were transferred to spatial images and then to frequency domain images according to the number of consecutive. Then, the convolution neural network (CNN) was used to train and validate the classification accuracy of the person’s stress state. Three types of classification models were built: person-specific models, generic classification models, and calibrated-generic classification models. The average classification accuracies achieved by person-specific models using spatial images and frequency domain images were 99.9%, 100%, and 99.8%, and 99.68%, 98.97%, and 96.4% for the training, validation, and test, respectively. By combining 20% of the samples collected from test subjects into the training data, the calibrated generic models’ accuracy was improved and outperformed the generic performance across both the spatial and frequency domain images. The average classification accuracy of 99.6%, 99.9%, and 88.1%, and 99.2%, 97.4%, and 87.6% were obtained for the training set, validation set, and test set, respectively, using the calibrated generic classification-based method for the series of inter-beat interval (IBI) spatial and frequency domain images. The main contribution of this study is the use of the frequency domain images that are generated from the spatial domain images of the IBI extracted from the PPG signal to classify the stress state of the individual by building person-specific models and calibrated generic models.

Wireless Sensors System for Stress Detection by Means of ECG and EDA Acquisition

This paper describes the design of a two channels electrodermal activity (EDA) sensor and two channels electrocardiogram (ECG) sensor. The EDA sensors acquire data on the hands and transmit them to the ECG sensor with wireless WiFi communication for increased wearability. The sensors system acquires two EDA channels to improve the removal of motion artifacts that take place if EDA is measured on individuals who need to move their hands in their activities. The ECG channels are acquired on the chest and the ECG sensor is responsible for aligning the two ECG traces with the received packets from EDA sensors; the ECG sensor sends via WiFi the aligned packets to a laptop for real time plot and data storage. The metrological characterization showed high-level performances in terms of linearity and jitter; the delays introduced by the wireless transmission from EDA to ECG sensor have been proved to be negligible for the present application.