Research On Non-invasive Blood Glucose Detection Based On PPG And ECG Fusion Signal

Diabetes is one of the three major chronic diseases in the world.According to the data from authoritative organizations such as the World Health Organization,there are 422 million adults worldwide suffering from diabetes.China has largest number of diabetic patients in the world.Diabetics need to check their blood glucose levels frequently and maintain blood glucose stable by injecting insulin,having hypoglycemic agents,and improving diets.The existing method of detecting blood glucose is mostly measured by collecting fingertip blood.Although the accuracy is high,the disadvantage is that the risk of infection is also high and it cannot be detected frequently,what’s more,it is difficult to observe the trend of blood glucose changes.There are semi-invasive measurement products on the market.The needles or laser pins are used to pierce the fingers or other parts of the body to obtain interstitial fluid(tissue fluid,etc.)for blood glucose measurement.The semi-invasive detection is less painful,however,detect blood glucose by interstitial fluid are the delayed results.It would be dangerous especially at times of hypoglycemia and hyperglycemia,and it still have the risk of infection.Non-invasive blood glucose detection has been a research hotspot in academia due to its no pain,continuously detection,and observable blood glucose trends.The global R&D team has been studied for non-invasive methods for detecting blood glucose in the past 20 years.Up to now,various non-invasive detection methods and technologies have emerged overseas.However,the difficulty of non-invasive blood glucose involves various limitations such as data acquisition,data processing,and calculation metho ds haven’t been solved.There is not a single technique can measure blood glucose precisely and could use for daily life with high accuracy.This article is based on previous research,Collecting Photoplethysmography(PPG)and Electrocardiogram(ECG)signals by using the prototype provided by Shu Tang Information Technology(Shenzhen)Co.,Ltd.;Improving the signal quality by empirical mode decomposition methods and intrinsic mode function based singular spectrum analysis method,making it easier to calculate the pulse width transmit time(PWTT).The optimized method has been used to select the feature vectors,and Extreme learning machine,convolution neural network and fractional order system have been approached to detect blood glucose.The optimal parameters selected methods are proposed.A total of 436 sets of data were collected from 17 test subjects(including 14 diabetic patients),and the data were divided into 6 levels(C1-C6)according to the blood glucose concentration.Through mixed modeling,individual modeling,selection of different data amount modeling,etc.Three groups of experiments were designed.The experimental results show that the accuracy of individual modeling is greatly superior to the accuracy of mixed modeling.Using more data(more than 10 sets)for modeling can improve the accuracy of detection result.The methods all achieved relatively accurate laboratory calculations(accuracy rate of over 85%),and the calculation results based on the extreme learning machine reached an accuracy of 91%,providing a choice for the non-invasive blood glucose detection technology productization.