Research On Non-invansive Blood Glucose Detection Method Based On Spectrum Analysis

Diabetes is a disease of abnormal glucose metabolism that manifests as hyperglycemia.Although it is a non-infectious chronic disease,it is extremely vulnerable to many kinds of complications,which seriously endangers human life and health.So far,there is no medical method to completely cure diabetes.Most patients use blood glucose meter to monitor and prevent blood glucose,which will not only bring psychological and physiological pressure to patients,but also may lead to blood crossover infection.Therefore,it is especially important to develop a non-invasive blood glucose detection system.Based on the near-infrared detection,an exploratory method to obtain the components related to the change of blood glucose from the frequency domain based on the photoelectric volume pulse wave signal(PPG signal)is proposed in this subject,and a blood glucose model to predict the blood glucose concentration value is established.The main research contents are as follows:1.The PPG signal detection scheme based on near-infrared light is studied,and the PPG signal of the human finger end is detected by the transmission detection method.The nearinfrared absorption peak of glucose was obtained by experiment,and the near-infrared LED with a wavelength of 1600 nm was selected as the detection light source.The amplitude information of the frequency component related to the blood glucose change was obtained from the spectrum of the PPG signal after wavelet decomposition.2.The acquisition system of PPG signal was built.The signal received by the sensor was sent to the computer through the A/D conversion by the corresponding conditioning circuit.The three denoising methods of Butterworth low-pass filtering,smoothing filtering and wavelet threshold were compared,and the denoising effect was evaluated in the frequency domain.The w avelet transform and cubic spline interpolation methods were used to remove the baseline drift.The wavelet threshold was selected for denoising,and the cubic spline interpolation method was used to remove the baseline drift.3.Several time-frequency analysis methods were researched and analysed,such as ShortTime Fourier Transform(STFT),Local Mean Decomposition(LMD),Wigner-Ville Distribution(WVD),and wavelet transform.Continuous wavelet transform and discrete wavelet transform were selected as the PPG signal processing algorithm in this thesis.Then FFT on the PPG signal when finished wavelet transform.Finally,the frequency component amplitude information related to blood glucose changes was obtained from the spectrum.4.Through the study of several regression modeling,the Partial Least Square Regression was choosed to establish the blood glucose model.The OGTT experiment was performed to compare the trend between the PPG signal’s frequency component extracted from the frequency domain and the blood glucose concentration value.The established blood glucose model was evaluated,and the correlation coefficient,RMSEP and RMSEC,and Clarke Error Grid were analyzed to verify the accuracy of the system.