Blood Glucose Prediction And Hypoglycemia Alarm Based On Canonical Correlation Analysis

With the development of the times and the advancement of science and technology,people's health consciousness is getting stronger and stronger.Diabetes,a disease with a very high prevalence,is getting more and more attention.Although diabetes itself does not endanger people's lives,patients with diabetes who have been in a state of the hyperglycemia for a long time will cause chronic damage to various tissues and organs,leading to many complications.In addition,patients with hypoglycemia are more likely to cause shock and even death.Therefore,self-monitoring and self-management of diabetic patients are crucial in the treatment process.This requires real-time monitoring of blood glucose changes,and even predicts blood glucose concentrations in advance to reduce the risk of disease.An effective prediction algorithm not only predict blood glucose concentration accurately,but also help identify hypoglycemic events and give early warnings,thereby reducing the incidence and maintaining health.The main contributions of this paper are as follows:(1)This paper first verifies the usability of the canonical correlation analysis(CCA)in predicting blood glucose concentration:First of all,auto-correlation and cross-correlation are calculated.After reconstructing the input(historical blood glucose value)and output(predicted blood glucose value)data,one gets a new comprehensive variable,so that the information contained in the data can be better used in the prediction process.Then,the correlation between the original data is reflected by finding the linear relationship between the comprehensive variables.Testing on clinical data shows the effectiveness of the method.Based on this,an error compensation strategy is proposed for the time lag phenomenon commonly found in time series prediction,which can effectively solve this problem and greatly reduce the prediction error.(2)Because the blood glucose concentration changes in the human body is a complex non-linear process.This paper proposes to introduce a kernel function on the basis of CCA,and map the original data to a high-dimensional space through the kernel function to reconstruct canonical correlation variables in high dimensions to find their correlation.Due to the adjustable parameters in the introduced kernel function,the accuracy of the prediction will be affected.Particle swarm optimization(PSO)is used to automatically optimize the kernel parameter.Finally,the PSO-KCCA is tested on real data and has obtained good prediction results.The obtained prediction results of PSO-KCCA were applied to early warning of hypoglycemia.In this process,the error of the prediction model was considered,and the previously fixed threshold of hypoglycemia was changed.Combine the prediction error of each patient with the hypoglycemia threshold to form a personalized threshold.The experimental data also come from measurements made by people with diabetes.Experimental results show that this method can not only make the false negative rate extremely low,but also ensure the sensitivity and specificity.