Blood Glucose Level Prediction with Minimal Inputs Using Feedforward Neural Network for Diabetic Type 1 Patients

Abstract

Introduction: Diabetes mellitus is one of the rapidly increasing diseases throughout the world. Studies reveal that proper management of blood glucose levels can reduce the complications associated with diabetes type 1. Objective: We use only continuous blood glucose data and predicted future blood glucose level using the previous data. Method: In this research, we input Continuous Glucose Monitoring (CGM) data to train a feedforward neural network using window model, to get optimal neural network for each subject in predicting prior blood glucose values. We have investigated virtual CGM data of 10 subjects in order to depict the efficiency of the proposed method and to validate the ANN. These ten case studies have been compiled from AIDA i.e. the freeware mathematical diabetes simulator. Results: For BGL predictions, improved results have been shown for minimal inputs in the prediction horizon (PH) of 15 minutes. Results produced by experimentation reveal that our ANN is accurate, adaptive, and can be implemented in clinics. Moreover, this study targets to make life easier for T1D patients by minimizing human input to the system. Conclusion and Future work: We conclude that feedforward gives better results for minimal inputs while other methods have better results for multiple inputs. In the future, we intend to investigate a larger collection of AIDA scenarios, and for real patients.