[home] [Personal Program] [Help]

---

Session: MSK & Sweat sensing
Session starts: Friday 27 January, 11:30
Presentation starts: 12:15
Room: Room 530

---

Xiaoyu Yin (Eindhoven University of Technology)
Elisabetta Peri (Eindhoven University of Technology)
Eduard Pelssers (Royal Philips)
Jaap den Toonder (Eindhoven University of Technology)
Massimo Mischi (Eindhoven University of Technology)

Abstract:
Introduction: Monitoring blood glucose levels is of utmost importance to improve the health and quality of life of patients affected by diabetes. Sweat sensing could be a non-invasive alternative to conventional invasive blood sampling for glucose monitoring. To achieve clinical relevance, models that can predict glucose levels in blood from measured glucose levels in sweat are necessary. However, such models are under-investigated. In this paper, we present a novel method based on biophysical modeling of the transport mechanism of glucose through a sweat gland and we propose a strategy to predict blood glucose levels by sweat sensing. Methods: The strategy we propose is based on a forward model that builds on the work of La Count et al. [1], and it numerically simulates the transport process of glucose from the blood to sweat using COMSOL Multiphysics. The main innovations in our model are in the transport mechanism of glucose from the interstitial space into the secretory coil, where the influx of glucose and water are combined. For making a backward prediction that gives an estimate of blood glucose levels from known sweat values, we used a sequential quadratic programming optimization algorithm based on the forward model. We used five datasets from the literature [1], including experimental glucose levels in blood and sweat, for both the forward model and backward strategy and validated them in terms of root-mean-square-error (RMSE) and root-mean-square-percentage-error (RMSPE). Results and Discussion: The average RMSE and RMSPE obtained by our forward model was 10±13 μmol/L and 10%±6%, smaller than for the original model (RMSE=15±18 μmol/L, RMSPE=20%±8%). This suggests that the proposed modifications improve the prediction accuracy of the model. The average RMSE and RMSPE of the backward model were 0.55±0.50 mmol/L and 6%±4% respectively, showing satisfactory prediction accuracy. Altogether, our study enables a more precise forecast of blood glucose changes based on known sweat glucose levels, possibly contributing to an overall improvement in diabetes monitoring via non-invasive sweat sensing.