Wearable Device Heart Rate and Activity Data in an Unsupervised Approach to Personalized Sleep Monitoring: Algorithm Validation

Data Preprocessing Techniques for AI and Machine Learning Readiness: Scoping Review of Wearable Sensor Data in Cancer Care

BACKGROUND

Wearable sensors are increasingly being explored in health care, including in cancer care, for their potential in continuously monitoring patients. Despite their growing adoption, significant challenges remain in the quality and consistency of data collected from wearable sensors. Moreover, preprocessing pipelines to clean, transform, normalize, and standardize raw data have not yet been fully optimized.

OBJECTIVE

This study aims to conduct a scoping review of preprocessing techniques used on raw wearable sensor data in cancer care, specifically focusing on methods implemented to ensure their readiness for artificial intelligence and machine learning (AI/ML) applications. We sought to understand the current landscape of approaches for handling issues, such as noise, missing values, normalization or standardization, and transformation, as well as techniques for extracting meaningful features from raw sensor outputs and converting them into usable formats for subsequent AI/ML analysis.

METHODS

We systematically searched IEEE Xplore, PubMed, Embase, and Scopus to identify potentially relevant studies for this review. The eligibility criteria included (1) mobile health and wearable sensor studies in cancer, (2) written and published in English, (3) published between January 2018 and December 2023, (4) full text available rather than abstracts, and (5) original studies published in peer-reviewed journals or conferences.

RESULTS

The initial search yielded 2147 articles, of which 20 (0.93%) met the inclusion criteria. Three major categories of preprocessing techniques were identified: data transformation (used in 12/20, 60% of selected studies), data normalization and standardization (used in 8/20, 40% of the selected studies), and data cleaning (used in 8/20, 40% of the selected studies). Transformation methods aimed to convert raw data into more informative formats for analysis, such as by segmenting sensor streams or extracting statistical features. Normalization and standardization techniques usually normalize the range of features to improve comparability and model convergence. Cleaning methods focused on enhancing data reliability by handling artifacts like missing values, outliers, and inconsistencies.

CONCLUSIONS

While wearable sensors are gaining traction in cancer care, realizing their full potential hinges on the ability to reliably translate raw outputs into high-quality data suitable for AI/ML applications. This review found that researchers are using various preprocessing techniques to address this challenge, but there remains a lack of standardized best practices. Our findings suggest a pressing need to develop and adopt uniform data quality and preprocessing workflows of wearable sensor data that can support the breadth of cancer research and varied patient populations. Given the diverse preprocessing techniques identified in the literature, there is an urgency for a framework that can guide researchers and clinicians in preparing wearable sensor data for AI/ML applications. For the scoping review as well as our research, we propose a general framework for preprocessing wearable sensor data, designed to be adaptable across different disease settings, moving beyond cancer care.

Case example of a female solo sailor competing in the Vendée Globe reveals extraordinary biological demands

The Vendée Globe is a non-stop, unassisted, single-handed round the world sailing race. It is regarded as the toughest sailing race, requiring high cognitive functioning and constant alertness. Little is known about the amount of sleep restriction and nutritional deficit experienced at sea and effects that fatigue have on sailors' performance. This report aimed to investigate these aspects by monitoring one of the female participants of the latest Vendée Globe. Sleep, food intake and stress were self-reported daily using specific app. Cognitive assessments were digitally completed. Heart rate and activity intensity were measured using a wrist-worn wearable device. Mean self-report sleep duration per 24 h was 3 hours 40 minutes. By the end of the 95 race days, the sailor reached a caloric deficit of 27,900 kcal. On average, the sailor spent 50 minutes per day in moderate-to-vigorous activity. Cognitive assessments did not show any effect of fatigue or stress on completion time or performance. Recent technological and communication advancement for offshore sailors, enabled continuous data to be monitored in near real time, even from the Southern Ocean. Moving forward this will enable greater understanding of when sailors will be at risk of poor decision making, illness or injury.

Personalized Non-contact Sleep Stage Estimation with Weighted Probability Estimation by Ultradian Rhythm

This paper focused on ultradian rhythms (a sleep cycle of approximately 60 to 120 minute) for personalizing sleep stage estimation, and proposed a personalized sleep stage estimation method that weights the results estimated by machine learning with the predicted ultradian rhythms. The ultradian rhythms are predicted by the body movement density which is correlated with ultradian rhythm. To investigate the effectiveness of the proposed method, this paper conducts human subjects experiment for eight subjects.Clinical relevance- The proposed method is compared with the results estimated by conventional ML, and the result of the proposed method is competitive with their conventional counterparts. This indicates that the ultradian rhythm has the potential for developing personalized sleep stage estimation.

Exploring Digital Biomarkers of Illness Activity in Mood Episodes: Hypotheses Generating and Model Development Study

BACKGROUND

Depressive and manic episodes within bipolar disorder (BD) and major depressive disorder (MDD) involve altered mood, sleep, and activity, alongside physiological alterations wearables can capture.

OBJECTIVE

Firstly, we explored whether physiological wearable data could predict (aim 1) the severity of an acute affective episode at the intra-individual level and (aim 2) the polarity of an acute affective episode and euthymia among different individuals. Secondarily, we explored which physiological data were related to prior predictions, generalization across patients, and associations between affective symptoms and physiological data.

METHODS

We conducted a prospective exploratory observational study including patients with BD and MDD on acute affective episodes (manic, depressed, and mixed) whose physiological data were recorded using a research-grade wearable (Empatica E4) across 3 consecutive time points (acute, response, and remission of episode). Euthymic patients and healthy controls were recorded during a single session (approximately 48 h). Manic and depressive symptoms were assessed using standardized psychometric scales. Physiological wearable data included the following channels: acceleration (ACC), skin temperature, blood volume pulse, heart rate (HR), and electrodermal activity (EDA). Invalid physiological data were removed using a rule-based filter, and channels were time aligned at 1-second time units and segmented at window lengths of 32 seconds, as best-performing parameters. We developed deep learning predictive models, assessed the channels' individual contribution using permutation feature importance analysis, and computed physiological data to psychometric scales' items normalized mutual information (NMI). We present a novel, fully automated method for the preprocessing and analysis of physiological data from a research-grade wearable device, including a viable supervised learning pipeline for time-series analyses.

RESULTS

Overall, 35 sessions (1512 hours) from 12 patients (manic, depressed, mixed, and euthymic) and 7 healthy controls (mean age 39.7, SD 12.6 years; 6/19, 32% female) were analyzed. The severity of mood episodes was predicted with moderate (62%-85%) accuracies (aim 1), and their polarity with moderate (70%) accuracy (aim 2). The most relevant features for the former tasks were ACC, EDA, and HR. There was a fair agreement in feature importance across classification tasks (Kendall W=0.383). Generalization of the former models on unseen patients was of overall low accuracy, except for the intra-individual models. ACC was associated with "increased motor activity" (NMI>0.55), "insomnia" (NMI=0.6), and "motor inhibition" (NMI=0.75). EDA was associated with "aggressive behavior" (NMI=1.0) and "psychic anxiety" (NMI=0.52).

CONCLUSIONS

Physiological data from wearables show potential to identify mood episodes and specific symptoms of mania and depression quantitatively, both in BD and MDD. Motor activity and stress-related physiological data (EDA and HR) stand out as potential digital biomarkers for predicting mania and depression, respectively. These findings represent a promising pathway toward personalized psychiatry, in which physiological wearable data could allow the early identification and intervention of mood episodes.

A Systematic Review of Time Series Classification Techniques Used in Biomedical Applications

Background: Digital clinical measures collected via various digital sensing technologies such as smartphones, smartwatches, wearables, and ingestible and implantable sensors are increasingly used by individuals and clinicians to capture the health outcomes or behavioral and physiological characteristics of individuals. Time series classification (TSC) is very commonly used for modeling digital clinical measures. While deep learning models for TSC are very common and powerful, there exist some fundamental challenges. This review presents the non-deep learning models that are commonly used for time series classification in biomedical applications that can achieve high performance. Objective: We performed a systematic review to characterize the techniques that are used in time series classification of digital clinical measures throughout all the stages of data processing and model building. Methods: We conducted a literature search on PubMed, as well as the Institute of Electrical and Electronics Engineers (IEEE), Web of Science, and SCOPUS databases using a range of search terms to retrieve peer-reviewed articles that report on the academic research about digital clinical measures from a five-year period between June 2016 and June 2021. We identified and categorized the research studies based on the types of classification algorithms and sensor input types. Results: We found 452 papers in total from four different databases: PubMed, IEEE, Web of Science Database, and SCOPUS. After removing duplicates and irrelevant papers, 135 articles remained for detailed review and data extraction. Among these, engineered features using time series methods that were subsequently fed into widely used machine learning classifiers were the most commonly used technique, and also most frequently achieved the best performance metrics (77 out of 135 articles). Statistical modeling (24 out of 135 articles) algorithms were the second most common and also the second-best classification technique. Conclusions: In this review paper, summaries of the time series classification models and interpretation methods for biomedical applications are summarized and categorized. While high time series classification performance has been achieved in digital clinical, physiological, or biomedical measures, no standard benchmark datasets, modeling methods, or reporting methodology exist. There is no single widely used method for time series model development or feature interpretation, however many different methods have proven successful.

An Overview of the Sensors for Heart Rate Monitoring Used in Extramural Applications

This work presents an overview of the main strategies that have been proposed for non-invasive monitoring of heart rate (HR) in extramural and home settings. We discuss three categories of sensing according to what physiological effect is used to measure the pulsatile activity of the heart, and we focus on an illustrative sensing modality for each of them. Therefore, electrocardiography, photoplethysmography, and mechanocardiography are presented as illustrative modalities to sense electrical activity, mechanical activity, and the peripheral effect of heart activity. In this paper, we describe the physical principles underlying the three categories and the characteristics of the different types of sensors that belong to each class, and we touch upon the most used software strategies that are currently adopted to effectively and reliably extract HR. In addition, we investigate the strengths and weaknesses of each category linked to the different applications in order to provide the reader with guidelines for selecting the most suitable solution according to the requirements and constraints of the application.

Synthesis of evidence on the use of ecological momentary assessments to monitor health outcomes after traumatic injury: rapid systematic review

BACKGROUND

With the increasing use of mobile technology, ecological momentary assessments (EMAs) may enable routine monitoring of patient health outcomes and patient experiences of care by health agencies. This rapid review aims to synthesise the evidence on the use of EMAs to monitor health outcomes after traumatic unintentional injury.

METHOD

A rapid systematic review of nine databases (MEDLINE, Web of Science, Embase, CINAHL, Academic Search Premier, PsychINFO, Psychology and Behavioural Sciences Collection, Scopus, SportDiscus) for English-language articles from January 2010-September 2021 was conducted. Abstracts and full-text were screened by two reviewers and each article critically appraised. Key information was extracted by population characteristics, age and sample size, follow-up time period(s), type of EMA tools, physical health or pain outcome(s), psychological health outcome(s), general health or social outcome(s), and facilitators or barriers of EMA methods. Narrative synthesis was undertaken to identify key EMA facilitator and barrier themes.

RESULTS

There were 29 articles using data from 25 unique studies. Almost all (84.0%) were prospective cohort studies and 11 (44.0%) were EMA feasibility trials with an injured cohort. Traumatic and acquired brain injuries and concussion (64.0%) were the most common injuries examined. The most common EMA type was interval (40.0%). There were 10 key facilitator themes (e.g. feasibility, ecological validity, compliance) and 10 key barrier themes (e.g. complex technology, response consistency, ability to capture a participant's full experience, compliance decline) identified in studies using EMA to examine health outcomes post-injury.

CONCLUSIONS

This review highlighted the usefulness of EMA to capture ecologically valid participant responses of their experiences post-injury. EMAs have the potential to assist in routine follow-up of the health outcomes of patients post-injury and their use should be further explored.

Quantification Analysis of Sleep Based on Smartwatch Sensors for Parkinson's Disease

Rapid eye movement (REM) sleep behavior disorder (RBD) is associated with Parkinson's disease (PD). In this study, a smartwatch-based sensor is utilized as a convenient tool to detect the abnormal RBD phenomenon in PD patients. Instead, a questionnaire with sleep quality assessment and sleep physiological indices, such as sleep stage, activity level, and heart rate, were measured in the smartwatch sensors. Therefore, this device can record comprehensive sleep physiological data, offering several advantages such as ubiquity, long-term monitoring, and wearable convenience. In addition, it can provide the clinical doctor with sufficient information on the patient's sleeping patterns with individualized treatment. In this study, a three-stage sleep staging method (i.e., comprising sleep/awake detection, sleep-stage detection, and REM-stage detection) based on an accelerometer and heart-rate data is implemented using machine learning (ML) techniques. The ML-based algorithms used here for sleep/awake detection, sleep-stage detection, and REM-stage detection were a Cole-Kripke algorithm, a stepwise clustering algorithm, and a k-means clustering algorithm with predefined criteria, respectively. The sleep staging method was validated in a clinical trial. The results showed a statistically significant difference in the percentage of abnormal REM between the control group (1.6 ± 1.3; n = 18) and the PD group (3.8 ± 5.0; n = 20) (p = 0.04). The percentage of deep sleep stage in our results presented a significant difference between the control group (38.1 ± 24.3; n = 18) and PD group (22.0 ± 15.0, n = 20) (p = 0.011) as well. Further, our results suggested that the smartwatch-based sensor was able to detect the difference of an abnormal REM percentage in the control group (1.6 ± 1.3; n = 18), PD patient with clonazepam (2.0 ± 1.7; n = 10), and without clonazepam (5.7 ± 7.1; n = 10) (p = 0.007). Our results confirmed the effectiveness of our sensor in investigating the sleep stage in PD patients. The sensor also successfully determined the effect of clonazepam on reducing abnormal REM in PD patients. In conclusion, our smartwatch sensor is a convenient and effective tool for sleep quantification analysis in PD patients.

Anomaly Detection Framework for Wearables Data: A Perspective Review on Data Concepts, Data Analysis Algorithms and Prospects

Wearable devices use sensors to evaluate physiological parameters, such as the heart rate, pulse rate, number of steps taken, body fat and diet. The continuous monitoring of physiological parameters offers a potential solution to assess personal healthcare. Identifying outliers or anomalies in heart rates and other features can help identify patterns that can play a significant role in understanding the underlying cause of disease states. Since anomalies are present within the vast amount of data generated by wearable device sensors, identifying anomalies requires accurate automated techniques. Given the clinical significance of anomalies and their impact on diagnosis and treatment, a wide range of detection methods have been proposed to detect anomalies. Much of what is reported herein is based on previously published literature. Clinical studies employing wearable devices are also increasing. In this article, we review the nature of the wearables-associated data and the downstream processing methods for detecting anomalies. In addition, we also review supervised and un-supervised techniques as well as semi-supervised methods that overcome the challenges of missing and un-annotated healthcare data.

Sleep Detection for Younger Adults, Healthy Older Adults, and Older Adults Living With Dementia Using Wrist Temperature and Actigraphy: Prototype Testing and Case Study Analysis

BACKGROUND

Sleep is essential for one's health and quality of life. Wearable technologies that use motion and temperature sensors have made it possible to self-monitor sleep. Although there is a growing body of research on sleep monitoring using wearable devices for healthy young-to-middle-aged adults, few studies have focused on older adults, including those living with dementia.

OBJECTIVE

This study aims to investigate the impact of age and dementia on sleep detection through movement and wrist temperature.

METHODS

A total of 10 younger adults, 10 healthy older adults, and 8 older adults living with dementia (OAWD) were recruited. Each participant wore a Mi Band 2 (accemetry-based sleep detection) and our custom-built wristband (actigraphy and wrist temperature) 24 hours a day for 2 weeks and was asked to keep a daily sleep journal. Sleep parameters detected by the Mi Band 2 were compared with sleep journals, and visual analysis of actigraphy and temperature data was performed.

RESULTS

The absolute differences in sleep onset and offset between the sleep journals and Mi Band 2 were 39 (SD 51) minutes and 31 (SD 52) minutes for younger adults, 49 (SD 58) minutes and 33 (SD 58) minutes for older adults, and 253 (SD 104) minutes and 161 (SD 94) minutes for OAWD. The Mi Band 2 was unable to accurately detect sleep in 3 healthy older adults and all OAWDs. The average sleep and wake temperature difference of OAWD (1.26 °C, SD 0.82 °C) was significantly lower than that of healthy older adults (2.04 °C, SD 0.70 °C) and healthy younger adults (2.48 °C, SD 0.88 °C). Actigraphy data showed that older adults had more movement during sleep compared with younger adults and that this trend appears to increase for those with dementia.

CONCLUSIONS

The Mi Band 2 did not accurately detect sleep in older adults who had greater levels of nighttime movement. As more nighttime movement appears to be a phenomenon that increases in prevalence with age and even more so with dementia, further research needs to be conducted with a larger sample size and greater diversity of commercially available wearable devices to explore these trends more conclusively. All participants, including older adults and OAWD, had a distinct sleep and wake wrist temperature contrast, which suggests that wrist temperature could be leveraged to create more robust and broadly applicable sleep detection algorithms.