Diabetes patients have to constantly adjust their glucose level by injecting two types of insulin. The insulin dose depends on a variety of parameters most prominently including the current glucose level, the bread units of actual food intake, and recent insulin doses. Electronic devices enable the patients to accurately measure their blood glucose level and to record all relevant data. Although, most patients get very experienced in choosing appropriate insulin doses, they can never be certain that the glucose level will stay within non-critical bounds. Specifically, a sudden drop of the glucose level (Hypoglucose) can induce a life-threatening malfunction of inner organs.
A reliable prediction of the patient's blood glucose level within a time window of only some hours would greatly improve life quality and possibly prevent the patient from critical glucose level states. A main problem in this prediction task is the choice of relevant parameters (feature selection). In order to assist a human expert in selecting the features for experiments with blood glucose predictions using Artificial Neural Networks (ANNs), we are developing GlucoPass. This system keeps a consistent record of each patient's data set and allows flexible data preprocessing. The filtered data is conveniently prepared to be used for ANN training. Visualization of ANN prediction and comparison to standard time series prediction methods are included. Furthermore, a WWW interface allows patients to update their measurement records via Internet which will allow validation of ANN prediction accuracy.
Screenshots of current prototype to be included here