Sensor-rich physical activity recognition for blood glucose prediction in people with type 1 diabetes

点击翻译按钮获取中文摘要

Hybrid closed-loop artificial pancreas systems are the current state of the art of diabetes technology [1]. These systems reduce the burden placed on those with diabetes by performing some of the processing required in diabetes management. However, to reach the next goal of diabetes technology, the fully closed-loop system, there are still challenges and a significant one of these is around management during physical activity. The impact of physical activity on blood glucose levels is not fully understood, although it is thought to vary noticeably between individuals and activity type [2]. This project aims to explore the potential of using additional sensor data, such as heart rate or accelerometer readings, during physical activity to improve the prediction of future blood glucose. This process will begin with a mixed method user study in which participants with type 1 diabetes will be recruited and questioned to understand: - Which medical devices and strategies they use to manage their diabetes. - How they currently manage physical activity. - What additional sensing equipment they would find acceptable to use during activity.This is to ensure that the algorithm developed is usable and useful to people and to gain an understanding of the nuances of blood glucose management that people with diabetes experience during physical activity. The link between physical activity and blood glucose levels will be explored through the user study, existing literature and available datasets. As part of this current glucose prediction algorithms will be investigated to inform the approach taken next.The next stage will be then developing an algorithm that takes the additional data and predicts blood glucose in the future. Machine learning techniques will be utilised to build this algorithm, with consideration taken into how it can be personalised to individuals. The OhioT1DM Dataset appears to contain potentially useful data and may form a starting point for this algorithm [3]. However, building a dataset of blood glucose, insulin dosage and activity sensor data may be necessary for the project and will form a useful contribution to the wider research topic. This will be especially so if data can be collected of multiple people performing the same exercise to compare their blood glucose response.Further stages of the project will be driven by areas of interest that arise during the earlier stages and directly by the users involved in this project.The link between physical activity and blood glucose levels is not fully understood and the work here aims to contribute its understanding. There is also currently limited use of additional sensor data in closed-loop control algorithms, bringing novelty to the research. However, several papers do mention similar ideas, suggesting that there are other researchers currently looking into it. Additionally, having a user-driven approach to the design of closed-loop algorithms will produce novel HCI contributions.References[1] Weaver KW, Hirsch IB. The Hybrid Closed-Loop System: Evolution and Practical Applications. Diabetes Technol Ther. 2018;20(S2):S216-S223. doi:10.1089/dia.2018.0091[2] Colberg SR, Hernandez MJ. The big blue test: effects of 14 minutes of physical activity on blood glucose levels. Diabetes Care. 2013;36(2):e21. doi:10.2337/dc12-1671[3] Marling C, Bunescu R. The OhioT1DM Dataset for Blood Glucose Level Prediction: Update 2020. CEUR Workshop Proc. 2020;2675:71-74.