Paper Submission & Registration
8th Dutch Bio-Medical Engineering Conference
Go-previous
15:20   Wearable
Chair: Chris Baten
15:20
15 mins
Decision-support system (DSS) that derives one's health condition from wearable sensors
Mohd Khalil Abu Hantash, Elisabeth Wilhelm, G.J. Verkerke, Ming Cao
Abstract: The increasing numbers of employees that experience medical conditions which affect their well-being and cause loss in productivity is becoming a real concern for companies in countries around the world. According to a study done on U.S. workers across 14 major occupations, the annual costs related to lost productivity totalled $84 billion [1]. These health conditions could be associated with risk factors such as an increase in work demands, unhealthy lifestyle, work environment and personal problems [2-7]. Novel sensors, communication technologies, and wearable devices can be used to monitor the user’s lifestyle and physiological parameters continuously without hampering his/her daily activities. Based on the information collected from sensors, interventions can be made in the form of advice to change the user’s lifestyle to improve the quality of life and productivity. Such a system can be referred to as a Decision Support Systems (DSS) [8]. Several studies have been performed on the development of DSSs for patients or people at risk of various chronic diseases [9-12].These systems collect data from users by means of questionnaires, sensors, or/and blood samples. The choice of adopted method and collected data depends on the addressed chronic disease/s, the associated risk factors, and the design requirements (e.g., accuracy) and restriction (e.g., invasive or non-invasive). Stress is a risk factor associated with several chronic diseases that can be assessed from physiological parameters such as heart rate, body temperature and blood pressure with an accuracy of up to 85% [13]. To our knowledge, there are no studies previously done that have developed a DSS that focus on the prevalent health conditions of employees. The main research question of this project is: Can a DSS that relies on data only from sensors and wearable devices influence one’s lifestyle and with that improve the quality of life and productivity of office employees? Within this project, the main objective will be to realise a DSS that relies on data collected from sensors and wearable devices to detect the most prevalent health conditions developed by office employees and provide the best possible interventions to improve the user’s quality of life and productivity.
15:35
15 mins
Data imputation framework to handle missing physiological data from wearable sensors
Carlijn Braem, Utku Yavuz, Peter Veltink
Abstract: Introduction: Data loss is inevitable in ambulatory monitoring of subjects using on-body sensing systems. Missing data influences statistical outcomes and interpretation, therefor it is critical for reconstructing data, with data imputation techniques. However, data loss and the use of data imputation are infrequently and inconsequently reported. A framework on how to handle missing data from wearable sensing is needed. Aim: To present and evaluate a data imputation framework for dealing with missing data from wearable sensors. Method: The data imputation framework consists of three parts; 1) characterization of time-series data, data loss and gaps, 2) data imputation models, and 3) performance analysis of data imputation models. Time-series characterization consists of the data histogram and autocorrelation. The percentage of data loss is evaluated in intervals, to determine trends in data loss. Data imputation models include, time-invariant models, such as mean substitution, interpolation and last observation carried forward. Time-variant models comprise of moving average models and dynamic time warping based imputation. ARIMA and structural model-based imputation will be implemented and evaluated before the conference. A procedure to simulate data loss was used to compare the performance of the data imputation models. Performance indicators include root mean square error (RMSE) and Bland-Altman analysis. All the methods are implemented in Python 3.7 (Python Software Foundation, https://www.python.org/ ). The data imputation framework is evaluated with data from the Diabetes and Lifestyle Cohort Twente (DIALECT)-2 cohort. For two weeks, wearable sensor data, comprised of continuous glucose (Freestyle Libre), step count and heart rate (Fitbit), were gathered in 73 free-living type 2 diabetics. Results: On average 14% [0-77%], 13% [1%-63%], and 19% [8-86%] of the glucose, heart rate and step count data are missing. In the step count, more data loss is present halfway the recording, due to data storage limitations. Preliminary RMSE results, indicate that cubic and linear interpolation outperforms mean and moving average imputation for glucose and heart rate data, respectively. Different data imputation techniques show minimal difference in RMSE for step count data. Conclusion: The presented data imputation framework can be implemented to evaluate and reconstruct missing physiological data from wearable sensors.
15:50
15 mins
Durability optimisation of powerful soft actuartors for wearables
Suraj Giri, Allan Veale, Herman van der Kooij
Abstract: Robotic exoskeletons can aid the rehabilitation of people with a stroke or spinal cord injury. They have actuators that assist the movements of the lower extremities. However, their weight and rigid structure limit the natural movement of the wearer, making them impractical for daily use. A radically different approach is the use of soft, elastic structures that do not hinder movement, offer safety, and adapt to complexly shaped bodies. These soft structures can actively assist movement, but are vulnerable to failing. This work presents the use of a semi-realistic Active Wearable Test (AWT) rig to optimise the safety and comfort of soft wearable actuators through extensive fatigue and failure testing. Based on the results, we suggest guidelines for durability testing and design of soft high power and torque actuators. An exosuit based on the pleated Pneumatic Interference Actuator (PPIA), which was powerful enough to provide torques required for all normal activities of daily living, was initially tested with the AWT to evaluate how the actuator design failed due to fatigue and failure testing. These results were used to further improve the actuator design for safety and performance such that it did not leak, remained aligned over many actuation cycles, and its dead volume was minimised for minimal hindrance. Additionally, as part of the guidelines for improved durability testing, the AWT’s motion was automated, friction degradable material was placed where the actuator contacts the AWT’s leg, and the AWT’s leg was inspected for permanent depressions after testing. These improvements should enhance testing reproducibility and enable characterisation of the interaction forces between the actuator tested and the AWT. The latter will be invaluable to designing safer wearable actuators.
16:05
15 mins
Feasibility of using smartwatches in clinical trials to observe drug induced changes in the heart rate
Willem O. Elzinga, Laura Borgmans, Vasileios Exadaktylos, Geert Jan Groeneveld, Robert -Jan Doll
Abstract: Introduction Electrocardiography (ECG) is of vital importance in early phase drug development. Changes in the heart rhythm are often studied in early phase clinical trials, either as a safety measure, or to study the impact on, for example, the heart rate (HR). While a Holter-ECG is the gold standard, the increasing quality of wearables able to capture the heart rate opens new possibilities for early phase drug development. For example, such devices yield additional HR measures while subjects are sent home. Here, we describe how a smartwatch can be used to observe drug induced changes in the HR. Methods Smartwatch data was collected during a multiple-dose, investigator-blinded, randomized, placebo-controlled, parallel-group design study including 12 subjects. Subjects wore a Withings Steel HR watch for 14 days. After the 7th day subjects either received a placebo or started with a daily ascending compound known to increase the heartrate. The watch captured the HR about every 10 minutes, and estimated sleep stages during the night. Holter-ECG was used to measure the HR on the first day and last day during a visit in the clinic. Results HR data was observed to be highly variable throughout the day, but less so during the night. Therefore, using the estimated sleep stages, only HR data collected while asleep were used for further analysis. Per study day, the 2.5, 50.0, and 97.5 percentiles of the HR were used as biomarkers. Differences between the placebo and active group were statistically significant, with the active group having an average increase in HR of about 9.4 beats per minute after treatment. The in-clinic Holter-ECG showed an increase of 13.3 beats per minute in the active arm in an awake state. Discussion Here, we demonstrated that a commercially available smartwatch can be used to complement in-clinic measurements. While the accuracy of HR and sleep state measures using a smartwatch can be debated, the addition of such a device is cheap and simple to manage. Therefore, we argue that smartwatch obtained HR has the potential as an informative biomarker to observe drug induced changes.


end %-->