A Multi-Network Fusion Prediction Method for Patient Blood Glucose Concentration Prediction and Hyper/Hypoglycemia Blood Glucose Warning

Abstract

To cope with the inconvenience as well as potential dangers of glycemic management in diabetic patients, we propose a fully automatic blood glucose prediction method based on neural networks. This method adopts a voting strategy to merge the advantages of regression prediction and classification prediction to construct a multi-network fusion prediction model, which can achieve the following dual objectives: (1) hyperglycemia and hypoglycemia warning with a time interval of 30 minutes, while the warning results have high reliability; (2) real-time blood glucose concentration prediction. To verify the effectiveness of our method, we tested 19 simulated patients in UVA/Padova T1DMS. The results show that our method can achieve the objectives well, and it also has better performance than the commonly used single neural network method.