Detection of Hypoglycemia and Hyperglycemia Using Noninvasive Wearable Sensors: Electrocardiograms and Accelerometry

BACKGROUND

Monitoring glucose excursions is important in diabetes management. This can be achieved using continuous glucose monitors (CGMs). However, CGMs are expensive and invasive. Thus, alternative low-cost noninvasive wearable sensors capable of predicting glycemic excursions could be a game changer to manage diabetes.

METHODS

In this article, we explore two noninvasive sensor modalities, electrocardiograms (ECGs) and accelerometers, collected on five healthy participants over two weeks, to predict both hypoglycemic and hyperglycemic excursions. We extract 29 features encompassing heart rate variability features from the ECG, and time- and frequency-domain features from the accelerometer. We evaluated two machine learning approaches to predict glycemic excursions: a classification model and a regression model.

RESULTS

The best model for both hypoglycemia and hyperglycemia detection was the regression model based on ECG and accelerometer data, yielding 76% sensitivity and specificity for hypoglycemia and 79% sensitivity and specificity for hyperglycemia. This had an improvement of 5% in sensitivity and specificity for both hypoglycemia and hyperglycemia when compared with using ECG data alone.

CONCLUSIONS

Electrocardiogram is a promising alternative not only to detect hypoglycemia but also to predict hyperglycemia. Supplementing ECG data with contextual information from accelerometer data can improve glucose prediction.

Analyzing Cattle Activity Patterns with Ear Tag Accelerometer Data

In this study, we equip two breeds of cattle located in tropical and temperate climates with smart ear tags containing triaxial accelerometers to measure their activity levels across different time periods. We produce activity profiles when measured by each of four statistical features, the mean, median, standard deviation, and median absolute deviation of the Euclidean norm of either unfiltered or high-pass-filtered accelerometer readings over five-minute windows. We then aggregate the values from the 5 min windows into hourly or daily (24 h) totals to produce activity profiles for animals kept in each of the test environments. To gain a better understanding of the variation between the peak and nadir activity levels within a 24 h period, we divide each day into multiple equal-length intervals, which can range from 2 to 96 intervals. We then calculate a statistical measure, called daily differential activity (DDA), by computing the differences in feature values for each interval pair. Our findings demonstrate that patterns within the activity profile are more clearly visualised from readings that have been subject to high-pass filtering and that the median of the acceleration vector norm is the most reliable feature for characterising activity and calculating the DDA measure. The underlying causes for these differences remain elusive and is likely attributable to environmental factors, cattle breeds, or management practices. Activity profiles produced from the standard deviation (a feature routinely applied to the quantification of activity level) showed less uniformity between animals and larger variation in values overall. Assessing activity using ear tag accelerometers holds promise for monitoring animal health and welfare. However, optimal results may only be attainable when true diurnal patterns are detected and accounted for.

Dose-response association between 24-hour total movement activity and testosterone deficiency in adult males

Background and objectives

Previous studies on the relationship between physical activity and testosterone are limited and controversial. Hence we investigated whether high level of physical activity is associated with a low risk of testosterone deficiency (TD).

Methods

This cross-sectional analysis was conducted in a representative sample of US adult males who participated in the 2011-2014 cycle of the National Health and Nutrition Examination Survey (NHANES). We used the monitor independent movement summary (MIMS) to assess activity intensity, a novel physical activity metrics developed using raw data collected by accelerometers. Multivariable regression and smooth curve fitting was used to describe the relationships between physical activity and TD, and segmented regression model were used to analyze the threshold effect between them. Sensitivity analysis was conducted using interaction and stratified analysis.

Results

A U-shaped relationship between daily MIMS units and risk of TD was observed. The optimal value of daily MIMS units for the lowest risk of TD was 14.77 (×103), the risk of TD decreased by 5% in patients per unit increase of daily MIMS units when daily MIMS units <14.77 (×103) (adjusted OR = 0.95, 95%CI: 0.91, 0.99), but increased by 12% per unit increase of daily MIMS units when daily MIMS units ≥14.77 (×103) (adjusted OR = 1.12, 95%CI: 1.01, 1.23). In sensitivity analyses, the threshold effect was also similar according to baseline characteristics (P-interaction >0.05).

Conclusion

In a nationally representative sample of US adult males, light to moderate intensity physical activity is associated with a lower odds of TD, while high-intensity physical activity is associated with a higher risk of TD.

Explaining and Visualizing Embeddings of One-Dimensional Convolutional Models in Human Activity Recognition Tasks

Human Activity Recognition (HAR) is a complex problem in deep learning, and One-Dimensional Convolutional Neural Networks (1D CNNs) have emerged as a popular approach for addressing it. These networks efficiently learn features from data that can be utilized to classify human activities with high performance. However, understanding and explaining the features learned by these networks remains a challenge. This paper presents a novel eXplainable Artificial Intelligence (XAI) method for generating visual explanations of features learned by one-dimensional CNNs in its training process, utilizing t-Distributed Stochastic Neighbor Embedding (t-SNE). By applying this method, we provide insights into the decision-making process through visualizing the information obtained from the model's deepest layer before classification. Our results demonstrate that the learned features from one dataset can be applied to differentiate human activities in other datasets. Our trained networks achieved high performance on two public databases, with 0.98 accuracy on the SHO dataset and 0.93 accuracy on the HAPT dataset. The visualization method proposed in this work offers a powerful means to detect bias issues or explain incorrect predictions. This work introduces a new type of XAI application, enhancing the reliability and practicality of CNN models in real-world scenarios.

Using quantile regression and relative entropy to assess the period of anomalous behavior of marine mammals following tagging

Tagging of animals induces a variable stress response which following release will obscure natural behavior. It is of scientific relevance to establish methods that assess recovery from such behavioral perturbation and generalize well to a broad range of animals, while maintaining model transparency. We propose two methods that allow for subdivision of animals based on covariates, and illustrate their use onN = 20 narwhals (Monodon monoceros) andN = 4 bowhead whales (Balaena mysticetus), captured and instrumented with Acousonde™ behavioral tags, but with a framework that easily generalizes to other marine animals and sampling units. The narwhals were divided into two groups based on handling time, short (t < 58 min) and long (t ≥ 58 min), to measure the effect on recovery. Proxies for energy expenditure (VeDBA) and rapid movement (jerk) were derived from accelerometer data. Diving profiles were characterized using two metrics (target depth and dive duration) derived from depth data. For accelerometer data, recovery was estimated using quantile regression (QR) on the log-transformed response, whereas depth data were addressed using relative entropy (RE) between hourly distributions of dive duration (partitioned into three target depth ranges) and the long-term average distribution. Quantile regression was used to address location-based behavior to accommodate distributional shifts anticipated in aquatic locomotion. For all narwhals, we found fast recovery in the tail of the distribution (<3 h) compared with a variable recovery at the median (∼1-10 h) and with a significant difference between groups separated by handling time. Estimates of bowhead whale recovery times showed fast median recovery (<3 h) and slow recovery at the tail (>6 h), but were affected by substantial uncertainty. For the diving profiles, as characterized by the component pair (target depth, dive duration), the recovery was slower (narwhals-long:t < 16 h; narwhals-short:t < 10 h; bowhead whales: <9 h) and with a difference between narwhals with short vs long handling times. Using simple statistical concepts, we have presented two transparent and general methods for analyzing high-resolution time series data from marine animals, addressing energy expenditure, activity, and diving behavior, and which allows for comparison between groups of animals based on well-defined covariates.

Monitoring Internal and External Load During Volleyball Competition

PURPOSE

The aim of this study was to identify the relationships between continuously measured internal and external load variables during volleyball competition.

METHODS

A total of 8 male elite volleyball athletes (Belgian Liga A and Liga B) were monitored during official competition matches. In total, 63 individual measurements are included in this study. The authors used heart-rate (HR) data as internal load and accelerometer-based activity as external load. Data were recorded at a sampling frequency of 1 Hz using wearable technology during official competition. Workload during continuous game play and individual jumps performed while serving and spiking (selected by means of video analysis) were studied using correlation analysis and dynamic time-series modeling.

RESULTS

Significant linear correlations were observed between peak acceleration and maximal HR of individual serves (ρ = .62; P = 1.6e-5) and spikes (ρ = .49; P = 1.2e-3) that were performed during the warm-up. These same actions performed during the match did not show significant correlations. The correlation between the mean HR and mean activity throughout the entire match was also found to be significant (ρ = .67; P = 2.0e-9). With respect to the time-series models, the mean value for the goodness of fit (RT2) between HR and activity was equal to .83 and .67 for the individual actions and the entire matches, respectively.

CONCLUSIONS

The results show that there are strong relationships between internal and external load during volleyball competition. Second-order transfer function models are capable of explaining the main dynamics of HR (internal load) in response to accelerometer-based activity (external load). Time-series analysis of continuously measured workload is proposed for use in practice.

Deep Learning for Classifying Physical Activities from Accelerometer Data

Physical inactivity increases the risk of many adverse health conditions, including the world’s major non-communicable diseases, such as coronary heart disease, type 2 diabetes, and breast and colon cancers, shortening life expectancy. There are minimal medical care and personal trainers’ methods to monitor a patient’s actual physical activity types. To improve activity monitoring, we propose an artificial-intelligence-based approach to classify physical movement activity patterns. In more detail, we employ two deep learning (DL) methods, namely a deep feed-forward neural network (DNN) and a deep recurrent neural network (RNN) for this purpose. We evaluate the two models on two physical movement datasets collected from several volunteers who carried tri-axial accelerometer sensors. The first dataset is from the UCI machine learning repository, which contains 14 different activities-of-daily-life (ADL) and is collected from 16 volunteers who carried a single wrist-worn tri-axial accelerometer. The second dataset includes ten other ADLs and is gathered from eight volunteers who placed the sensors on their hips. Our experiment results show that the RNN model provides accurate performance compared to the state-of-the-art methods in classifying the fundamental movement patterns with an overall accuracy of 84.89% and an overall F1-score of 82.56%. The results indicate that our method provides the medical doctors and trainers a promising way to track and understand a patient’s physical activities precisely for better treatment.

Hidden Markov models for monitoring circadian rhythmicity in telemetric activity data

Wearable computing devices allow collection of densely sampled real-time information on movement enabling researchers and medical experts to obtain objective and non-obtrusive records of actual activity of a subject in the real world over many days. Our interest here is motivated by the use of activity data for evaluating and monitoring the circadian rhythmicity of subjects for research in chronobiology and chronotherapeutic healthcare. In order to translate the information from such high-volume data arising we propose the use of a Markov modelling approach which (i) naturally captures the notable square wave form observed in activity data along with heterogeneous ultradian variances over the circadian cycle of human activity, (ii) thresholds activity into different states in a probabilistic way while respecting time dependence and (iii) gives rise to circadian rhythm parameter estimates, based on probabilities of transitions between rest and activity, that are interpretable and of interest to circadian research.

Visualising Combined Time Use Patterns of Children's Activities and Their Association with Weight Status and Neighbourhood Context

Compositional data techniques are an emerging method in physical activity research. These techniques account for the complexities of, and interrelationships between, behaviours that occur throughout a day (e.g., physical activity, sitting, and sleep). The field of health geography research is also developing rapidly. Novel spatial techniques and data visualisation approaches are increasingly being recognised for their utility in understanding health from a socio-ecological perspective. Linking compositional data approaches with geospatial datasets can yield insights into the role of environments in promoting or hindering the health implications of the daily time-use composition of behaviours. The 7-day behaviour data used in this study were derived from accelerometer data for 882 Auckland school children and linked to weight status and neighbourhood deprivation. We developed novel geospatial visualisation techniques to explore activity composition over a day and generated new insights into links between environments and child health behaviours and outcomes. Visualisation strategies that integrate compositional activities, time of day, weight status, and neighbourhood deprivation information were devised. They include a ringmap overview, small-multiple ringmaps, and individual and aggregated time–activity diagrams. Simultaneous visualisation of geospatial and compositional behaviour data can be useful for triangulating data from diverse disciplines, making sense of complex issues, and for effective knowledge translation.

Multiple imputation of completely missing repeated measures data within person from a complex sample: application to accelerometer data in the National Health and Nutrition Examination Survey

The Physical Activity Monitor component was introduced into the 2003-2004 National Health and Nutrition Examination Survey (NHANES) to collect objective information on physical activity including both movement intensity counts and ambulatory steps. Because of an error in the accelerometer device initialization process, the steps data were missing for all participants in several primary sampling units, typically a single county or group of contiguous counties, who had intensity count data from their accelerometers. To avoid potential bias and loss in efficiency in estimation and inference involving the steps data, we considered methods to accurately impute the missing values for steps collected in the 2003-2004 NHANES. The objective was to come up with an efficient imputation method that minimized model-based assumptions. We adopted a multiple imputation approach based on additive regression, bootstrapping and predictive mean matching methods. This method fits alternative conditional expectation (ace) models, which use an automated procedure to estimate optimal transformations for both the predictor and response variables. This paper describes the approaches used in this imputation and evaluates the methods by comparing the distributions of the original and the imputed data. A simulation study using the observed data is also conducted as part of the model diagnostics. Finally, some real data analyses are performed to compare the before and after imputation results. Published 2016. This article is a U.S. Government work and is in the public domain in the USA.

Optical wireless connected objects for healthcare

In this Letter the authors explore the communication capabilities of optical wireless technology for a wearable device dedicated to healthcare application. In an indoor environment sensible to electromagnetic perturbations such as a hospital, the use of optical wireless links can permit reducing the amount of radio frequencies in the patient environment. Moreover, this technology presents the advantage to be secure, low-cost and easy to deploy. On the basis of commercially available components, a custom-made wearable device is presented, which allows optical wireless transmission of accelerometer data in the context of physical activity supervision of post-stroke patients in hospital. Considering patient mobility, the experimental performance is established in terms of packet loss as a function of the number of receivers fixed to the ceiling. The results permit to conclude that optical wireless links can be used to perform such mobile remote monitoring applications. Moreover, based on the measurements obtained with one receiver, it is possible to theoretically determine the performance according to the number of receivers to be deployed.