Research On Noninvasive Blood Glucose Detection Technology And Its Application In Medical Service

With the development of economic,there are great changes in the diet structure of people.Excessive intake of high-sugar and high-calorie food has made the incidence of diabetes higher than before in global and the patients has begun to cover younger people.As a key point,blood glucose detection plays an important role in the prevention and treatment of diabetes.However,the current blood glucose detection technology has the disadvantages of trauma,easy infection,long time,and high cost,which is not conducive to the daily management of patients and the regulation of their own blood sugar.Therefore,an algorithm for non-invasive blood glucose detection using electrocardiogram(ECG)and photoplethysmography(PPG)is proposed in this paper.Based this an APP for the actual needs of diabetic patients is designed.Firstly,the MAX86150 evaluation kit is used to collect ECG and PPG signals.Due to disturbance from the collection environment,collection equipment,and human body conditions,there are high-frequency noise and baseline wandering noise in the signal.An improved median filter algorithm to remove high-frequency noise in the signal is proposed in this paper.For the baseline wandering noise in the signal,the denoising effects of the wavelet transform and empirical mode decomposition(EMD)algorithms is compared.The result shows that the two algorithms are both good for filtering.The filtering effect of EMD is worse than wavelet transform,which will cause the loss of part features in the ECG signal.Therefore,the wavelet transform method is used to remove the baseline wandering noise in the signal.Secondly,in order to identify the peak point of R wave of the ECG signal and the rising branch point of the PPG signal to obtain the pulse wave transit time(PWTT),the differential threshold method for feature point identification is proposed in this paper.Meanwhile,Kaiser-Teager features,heart rate features,spectral entropy features,and spectral energy logarithmic features are extracted from the two signals.The human body information such as height and weight,room temperature and detection time are combined as machine learning input.Then three types of machine learning algorithms including XGBoost,Light GBM and BP Neural Network are used to predict human blood glucose.The blood glucose reference value of this paper was detected by invasive blood glucose meter.The number of samples was 10 people,and the total number of samples was 1680.The data was collected for two weeks before and after meals and exercises.The final experiment results show that in the single-person model,the proportion of the three algorithms in the A area of clark error network is more than 83%,and Light GBM has the highest accuracy with 89%.The accuracy of the multiple-person model is lower than the single-person model.And the proportion in A area of Clark error network of XGBoost is the best,which reaches 75%.Finally,based on the blood glucose detection algorithm of the single-person model,a noninvasive blood glucose management APP is designed in this paper.Patients can check blood glucose,view historical blood glucose values and upload data through their mobile phones at any time.The first two of them are convenient for patients to monitor and regulate their blood glucose levels daily.The latter can upload the data on time to the server,which can make professional medical staff easily find abnormalities and give professional advice.