Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : laboratoire URePSSS, Université de Lille
Durée : 6 mois
Contact : elsa.heyman@univ-lille.fr
Date limite de publication : 2024-01-31

Contexte :
In type 1 diabetes (T1D), physical activity is an essential component of the treatment plan because of its
recognised beneficial effects on numerous health parameters. Nevertheless, T1D individuals often have
a level of physical activity that falls short of international recommendations. The main obstacles to
physical activity are fear of hypoglycaemia and diabetes imbalance. Depending on the intensity, duration,
method and timing of the last insulin injection, physical activity can have a hypo- or hyperglycaemic
effect. Faced with this situation, it is difficult for T1D sports practitioners to anticipate appropriate
adaptations to their insulin and/or diet: at present, recommendations as to the adaptations to be made
according to the characteristics of the exercise remain very vague due to the lack of studies carried out
under real-life conditions (glucose monitoring sensors).
The overall aim of the project is to improve the accuracy of algorithms for predicting variations in blood
sugar levels as a function of physical activity, using data recorded by sensors worn in everyday life,
taking into account diet, insulin administered (e.g., from insulin pumps), etc.

Sujet :
The sensors (accelerometer, continuous glucose monitoring systems, insulin pumps, etc) worn by the
patients living with T1D generate a large amount of temporal data each day. This data needs to be
processed and analysed automatically to produce simple indicators that are useful to patients, and to
enable research teams to base their predictive models on it. Codes for calculating indices of glycaemic
excursions (e.g., time spent at different thresholds of hypoglycaemia or hyperglycaemia, glycaemic
variability, i.e., rapid variations towards high and low glycaemia levels, etc.) have already been developed
to process data from glucose sensors. These codes are also designed to create a formatted database for
each patient, enabling a number of simple indicators to be displayed and calculated. These codes were
then put into an intuitive web interface for researchers and doctors.
The main objective of the internship will be to explore machine learning methods in order to improve
the algorithms and statistical models for prediction of hypo and hyperglycaemic risk around physical
activity (considering their temporal dynamics).

Profil du candidat :
o In-depth knowledge of data science
o In-depth knowledge of the main supervised and unsupervised learning models
o Strong skills in R or, failing that, in Python for data science
o Fluent reading of English
o Autonomy, rigor, reliability
o Ability to listen and communicate with the scientific community
o Ability to present work orally and in writing

Formation et compétences requises :
master in data science.

Adresse d’emploi :
The internship will be located at the University of Lille, within the URePSSS laboratory
(Multidisciplinary Sport, Health and Society Research Unit, ULR 7369) under the supervision of Prof.
Elsa Heyman and Dr. Pierre Morel, in close collaboration with Prof. Philippe Preux of the CRISTAL
laboratory (UMR 9189, IT, Signal and Automation Research Centre). Regular videoconference meetings
will be organised with a collaborator from the University of Rennes (Joris Heyman) and Montreal (Rémi
Rabasa-Lhoret).

Document attaché : 202311281455_FicheStageURePSSS_Anglais_URePSSS_CRISTAL.pdf