Predicting Glycemia by Using RNNs and Heart Rate Patient Data

Type 1 diabetes mellitus (T1DM) is a chronic disease caused by the destruction of the pancreatic beta cells resulting in an insufficient insulin production. This generates high blood glucose levels which causes physical and cardiovascular problems [Guzzi et al.(2023)]. Currently, the commonly available therapy regards the intake of insulin to control glycemia. The level of glycemia varies on a daily basis and it is influenced by the glucose intake. The correct prediction of glycemia variability may suggest a correct dosage of insulin, therefore the optimal control strategy. There exist some physiological parameters which can be used for prediction of glycemia. Recurrent Neural Networks (RNNs) have been largely used for prediction of a continuous output from a similar input. Heart rate can be used as input for an RNN and its output used as glycemia values predictor. We report about an experimet performed at University Hospital of Catanzaro on a sampled dataset. We report about results in using an RNN for predicting blood glucose levels from heart rate signal.