Internet des Objets pour le suivi et la modélisation de la glycémie de patients diabétiques
Début du financement: 01/10/2017
Axe DigiCosme & tâche: DataSense
Sujet : Internet des Objets pour le suivi et la modélisation de la glycémie de patients diabétiques
Directeurs de thèse :Medhi AMMI, LIMSI & Mounim El Yacoubi, Telecom SudParis
Institutions : LIMSI, Telecom SudParis
Laboratoire gestionnaire : LIMSI
Doctorant : Maxime DE BOIS
Productions scientifiques :
- Energy expenditure estimation through daily activity recognition using a smart-phone, Maxime DE BOIS, Hamdi AMROUN, Mehdi AMMI, WF-IoT, 2018
Diabetes Mellitus (DM) has become a paramount issue in the modern world. Globally, 422 millions of adults are estimated to be diabetic in 2014 and diabetes itself is directly imputed 1.5 million deaths in 2012 (World Health Organization, 2016).
Diabetes is a chronic disease that can be divided into three main categories: type-1 diabetes mellitus (T1DM), type-2 diabetes mellitus (T2DM) and gestational diabetes. While T1DM is characterized by the body’s inability to create insulin (which is an hormone that regulates blood glucose), T2DM is identified by its increasing resistance to insulin and gestational is distinguished by high blood glucose during pregnancy.
For diabetic people, the main problem is to maintain their BG within acceptable ranges throughout days and nights. If the BG falls to low (hypoglycemia), it can result in various short-term symptoms including clumsiness, trouble talking, loss of consciousness or even death depending on the severity of the hypoglycemia. In the other hand, high blood glucose (hyperglycemia) can lead to more long-term complications such as poor blood flow, cardiovascular diseases or blindness.
To help diabetic people cope with their disease, a lot of technical solutions and have been engineered. One of the most notable and recent progress in the Internet of Things area is the invention of Continuous Glucose Monitoring (CGM) devices. Diabetic people do not need to prick their skin with a lancet to obtain their current blood glucose level anymore. It can now be done by scanning a patch located at the back of his arm, like with the FreeStyle Libre CGM device. Besides, we are currently witnessing the rise of smart-phone coaching applications featuring diabetes such as the FDA-approved application called mySugr.
From a research perspective, a lot of advances have been done in regulating blood glucose. We can identify two main approaches to the problem. In on hand, there is the challenging project of building an Artificial Pancreas that can automatically deliver the appropriate insulin dosages to the body (Bequette et al., 2012). To do so, efforts are focused on building efficient closed-loop systems such as Model Predictive Controllers (Dassau et al., 2012). In the other hand, the endeavors of a lot of researchers have been toward the prediction of future BG values. For instance, the forecasting of future glucose values can then be used inside a MPC or even as is to alert the diabetic of an incoming hypoglycemia.
This is this context that my doctorate takes place. It is entitled Internet of Things for the monitoring and the modeling of diabetic patients’ glycaemia. My works aim at improving the prediction of blood glucose and at the same time at integrating those kind of predictive models into IoT systems that can be used by diabetic patients, in real life environments. To do so, there are quite a lot of issues that need to be addressed. First a glucose predictive model needs to be very close to reality and failure-proof or the patient may suffer severe consequences. Then, in order for those models to be useful for to diabetic patients, they need to be carefully integrated: they need to be easy to use, non-intrusive, portable and secure.
We are currently working with the diabetic association Revesdiab and the Sud-Esson eHospital Center (CHSE) to create data collection experiments with diabetic patients. The current experiment involves 2 physical activity monitoring wristbands (Actigraph wGT3X-BT and Fitbit Charge 2), the FreeStyle Libre continuous glucose monitoring device and the diabetic smartphone app mySugr. Once enough data have been collected, we will be able to process them in order to build accurate glucose predictive models.
Besides, we are currently working on those predictive models using data simulators such as the UVA-Padova T1DM Simulator. Models such as Support Vector Regression, Gaussian Process, Autoregressive and Recurrent Neural Networks are currently investigated.