Nonlinear block-oriented identification for insulin-glucose models

The application of advanced identification techniques to model
insulin-glucose systems represents a crucial step towards the development of
the artificial pancreas for diabetes patients. Type 1 diabetes mellitus
(T1DM) is a disease characterized by the fact that the pancreas is not able
to produce a sufficient amount of insulin. Therefore, when treating patients
with exogenous insulin delivery, the level of glucose in the blood needs to
be carefully regulated to avoid severe problems such as hypoglycemia,
retinopathy or cardiovascular diseases.
Several mathematical descriptions (mainly first principle models) have been
considered to represent the diabetic patient, and automated closed-loop
control systems based on these models are currently under study. The main
difficulties associated to the existing models are related to the fact that
the tuning of parameters differs for each patient, and that the model
parameters cannot be identified in practice.
The objective of this work is the identification of models to describe the
glucoregulatory system, based on input-output data. In particular, nonlinear
system identification methods for block structures are combined with the use
of nonlinear functions from statistical learning, e.g. Neural Networks
(NNs).