Research On Blood Glucose Monoitoring Method Based On ECG Signal

At present,diabetes has become the third most serious threat to human health inferior to cancer and cardiovascular and cerebrovascular diseases.Therefore,disease prevention and clinical treatment of diabetes have aroused strong attention in the medical field.According to the latest statistics from the International Diabetes Federation,537 million people have diabetes all around the world,the number of adults with the disease has more than doubled since 20 years ago,and the numbers are expected to continue to rise.Therefore,both healthy people and diabetic patients urgently need an accurate,non-invasive,efficient and convenient blood glucose monitoring instrument.On the current,most of the relevant instruments that can detect blood glucose in the market are invasive methods of blood collection through finger tips.Users will not only have trauma,pain and even the risk of infection in the process of use,but also the cost of blood glucose detection is high and the accuracy cannot be guaranteed.Hypoglycemia or hyperglycemia events in both healthy people and diabetic patients will cause irreversible harm to themselves.Frequent hyperglycemia will lead to decreased body resistance,impaired kidney function,various neurovascular complications and even diabetes.Hypoglycemia can cause symptoms such as hand shaking,dizziness,bad temper,induce cardiovascular,cerebrovascular diseases or sudden death in severe cases.Reliable non-invasive and high-precision blood glucose monitoring instruments are not yet widely available,and blood glucose monitoring technology is not yet mature,so human hypoglycemia or hyperglycemia events cannot be effectively monitored.Human blood glucose monitoring has also attracted great attention in many fields.Non-invasive blood glucose detection instruments are still the focus of academic research in the field of human health management in the domestic and overseas because of their favorable market application prospects,such as non-trauma,real-time monitoring of blood glucose change trend and wide application of people.Related studies have shown that blood glucose change will affect the change of physiological signals related to body,but because the research needs a large number of experimental data and theoretical support and data processing,data calculation,related research algorithm needs to be improved,and the influence of such factors as noninvasive blood glucose detection technology is still in continuous research and development,noninvasive blood glucose detection instrument has not been promoted and apply to the public.Based on the current blood glucose monitoring technology,this study proposes a high-precision blood glucose monitoring technology,which can monitor blood glucose level in real time and accurately based on the ECG signasl of human body.Dexcom G6 and BIOPAC continuous monitoring system are used to continuously collect human blood glucose and real-time ECG signals.After data pretreatment,two deep learning methods are used to achieve high accuracy prediction of blood glucose based on single ECG cycle signal and extracted single cycle ECG signal characteristics respectively.The glucose monitoring is more accurate and effective by adjusting the parameters of the machine learning model.The results show that CNN-LSTM has an accuracy of 80% and 88% for individual and population modeling based on the single ECG cycle signal.On this occasion,the population modeling perform better,and the blood glucose concentration was divided into ten categories with F1 score were 0.95,0.88,0.91,0.85,0.92,0.88,0.86,0.86,0.87 and0.86,respectively.The accuracy of individual modeling and population modeling using random forest based on extracted ECG signal features was 99.9% and 83.8%,respectively.Under the circumstances,the individual modeling is superior and the average F1 score is up to 0.9999.Therefore,this study shows that CNN-LSTM is used for population modeling based on single ECG cycle signals and random forest is used for individual modeling based on extracted features of single cycle ECG signals,both approaches are can provide theoretical support for real-time,accurate blood glucose monitoring and technical guidance.However,based on the extracted characteristics of ECG signals modeling and using random forest for individual blood glucose monitoring performance superiority,blood glucose monitoring accuracy exceed 99.9%,can be personalized noninvasive blood glucose monitoring technology production to provide significance support.