CN110338813B

CN110338813B - Noninvasive blood glucose detection method based on spectrum analysis

Info

Publication number: CN110338813B
Application number: CN201910480597.7A
Authority: CN
Inventors: 陈剑虹; 何菲; 雷苏力; 刘泽晨
Original assignee: Xian University of Technology
Current assignee: Xian University of Technology
Priority date: 2019-06-04
Filing date: 2019-06-04
Publication date: 2022-04-12
Anticipated expiration: 2039-06-04
Also published as: CN110338813A

Abstract

A noninvasive blood glucose detection method based on spectral analysis comprises the following steps: step 1, collecting a PPG signal of a finger tip by using a photoelectric detector based on the principle of near infrared spectrum, and sending the PPG signal to a computer after signal conditioning; step 2, filtering high-frequency noise of the collected PPG signal; step 3, acquiring a valley point of the denoised PPG signal; step 4, fitting baseline drift of the PPG signal by taking the valley point as a base point, and obtaining a pure PPG signal by removing the baseline drift; step 5, decomposing the pure PPG signal, performing FFT on each decomposed layer of signal, and acquiring PPG signal frequency component information related to blood sugar value change on the premise of knowing the blood sugar value change; step 6, performing an OGTT experiment, and storing a blood glucose reference value and PPG signal data; step 7, establishing a blood glucose model through partial least squares regression; step 8, predicting the blood sugar value by using the established blood sugar model, and carrying out error analysis; the method has the characteristics of no wound and accurate detection result.

Description

Noninvasive blood glucose detection method based on spectrum analysis

Technical Field

The invention belongs to the technical field of biological signal processing, and particularly relates to a noninvasive blood glucose detection method based on spectrum analysis.

Background

Near infrared light is an electromagnetic wave, has a wavelength of about 780nm to 2500nm, and is suitable for human body detection. The near infrared spectrum technology is simple to operate, does not damage a sample when being used for detection research, does not need a reagent, does not have hidden danger in the aspect of environmental pollution, and is an ideal detection technology. The near infrared light is used to detect blood sugar of a human body, and is generally used to detect skin, mucosa or other body parts containing body fluids, such as fingers, wrists, ears, etc. When the near infrared light irradiates the human body detection part, skin tissues, bones and blood of the detection part can absorb the light to a certain extent. The optical signal transmitted or reflected back to the detection site is converted into an electrical signal, i.e. a photoplethysmographic signal (PPG signal), using a photosensor. When the glucose concentration in human blood increases, the absorption of light by the glucose in the blood changes, which is reflected in a weak peak change in the PPG signal, but since this change is very small and is difficult to accurately determine from the time domain, it is proposed to obtain frequency component information of the PPG signal, which is related to the change in blood glucose level, from the frequency domain.

The traditional blood sampling type blood sugar detection method not only can bring physiological and psychological pressure to patients, but also can cause cross infection of blood, and continuous monitoring cannot be realized. The household glucometer needs to replace the test paper when detecting blood sugar every time, and the medical expenditure of a family is increased. Compared to these invasive blood glucose test methods. The noninvasive blood glucose detection method provided by the invention can realize noninvasive detection and continuous blood glucose monitoring, is favorable for integration of medical monitoring equipment, and is suitable for daily monitoring of hospitals and families.

With the increasing demand of people for medical detection equipment, many experts and scholars try to perform blood glucose detection by using different noninvasive methods, including a bio-impedance method, a fluorescence technology, an optical method and the like. The biological impedance method is greatly influenced by moisture and temperature, the fluorescence technology has the defect of short service life, and the optical method has the characteristics of no wound, high speed, multi-dimensional information and the like and is widely applied to the research of non-invasive blood sugar detection. Near infrared light is the most commonly used technique in optical non-invasive blood glucose measurements due to its good penetration into body fluids and soft tissues. In current methods of detecting blood glucose by combining near infrared light with PPG technology, time domain analysis is usually performed on PPG signals to obtain information related to blood glucose changes in the signals. However, since the change in blood glucose is very weak in the waveform of the PPG signal, it is difficult to accurately detect the blood glucose level from the detection result.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention aims to provide a noninvasive blood glucose detection method based on spectral analysis, so that the invasiveness of blood sampling type blood glucose detection and possible blood cross infection are avoided, and the frequency component information of a PPG signal related to the blood glucose value change is extracted from a frequency domain aiming at the problem that the PPG signal is difficult to accurately reflect the blood glucose change from a time domain, so that a blood glucose model is established, and the blood glucose value is predicted; the method has the characteristics of no wound and accurate detection result.

In order to achieve the purpose, the invention adopts the technical scheme that: a noninvasive blood glucose detection method based on spectral analysis comprises the following steps:

step 1, collecting a PPG signal of a finger tip by using a photoelectric detector based on the principle of near infrared spectroscopy, wherein the original PPG signal directly collected is very weak and contains noise interference, so that the original PPG signal is subjected to circuit conditioning and then is subjected to A/D collection and then is sent to a computer by a serial port;

step 2, in order to better analyze the frequency components related to the change of the PPG signal and the blood sugar value from the frequency domain, filtering the high-frequency noise of the collected PPG signal by using software filtering to obtain a denoised PPG signal;

step 3, setting a threshold value after twice differentiation on the denoised PPG signal to obtain a valley point of the PPG signal;

step 4, fitting a base line drift with an envelope line as a PPG signal by using a cubic spline interpolation method with a valley point as a base point, and removing the base line drift by using the difference between the original PPG signal and the original PPG signal to obtain a pure PPG signal;

step 5, decomposing the pure PPG signal by using continuous wavelet transform and discrete wavelet transform, performing fast Fourier transform on the decomposed wavelet coefficient to obtain a corresponding frequency spectrum, and acquiring PPG signal frequency component information related to blood sugar value change through a large amount of experimental data on the premise of knowing the blood sugar value change;

step 6, performing an OGTT experiment, performing PPG signal acquisition and real blood glucose value detection in the step 2 every 5 minutes, storing data, and performing the experiment for 90-120 minutes;

step 7, after denoising and baseline drift removing are carried out on the PPG signal acquired in the step 6, frequency component amplitude information of the PPG signal related to the blood sugar value is acquired through the method in the step 5, and a blood sugar model is established by utilizing partial least square regression between the amplitude information and the real blood sugar value;

and 8, predicting the blood sugar value by using the established blood sugar model, and performing error analysis by using the prediction error and the Clark grid error.

Compared with the prior art, the invention has the following beneficial characteristics:

the invention combines the PPG technology and the near infrared technology to carry out non-invasive blood sugar detection, has simple equipment, avoids the complexity of the traditional large-scale equipment for carrying out blood sugar detection based on the near infrared spectrum, and also solves the defect that the waveform change related to the blood sugar change is difficult to be accurately analyzed from the time domain when the traditional PPG technology is used for carrying out blood sugar detection. The blood sugar detection method provided by the invention not only utilizes simple detection equipment, but also has smaller predicted blood sugar value error and accords with the national standard.

Drawings

FIG. 1 is an overall flow chart of the present invention.

Fig. 2 is a schematic block diagram of the PPG signal acquisition system of the present invention.

Fig. 3(a) is a diagram of the raw PPG signal acquired by the present invention.

Fig. 3(b) is a spectral diagram of the original PPG signal of the present invention.

FIG. 3(c) is a PPG signal graph after denoising based on the wavelet threshold method.

Fig. 3(d) is a spectrogram of a PPG signal denoised based on a wavelet thresholding method according to the present invention.

Fig. 4 is a PPG signal valley point identification diagram based on a differential threshold method.

FIG. 5 is a graph of baseline wander fitted using cubic spline interpolation based on valley points in accordance with the present invention.

Figure 6 is a graph of PPG signal after removal of baseline drift.

Fig. 7 is a frequency spectrum diagram of wavelet coefficients of the PPG signal at various scales after continuous wavelet transform.

Fig. 8 is a spectrum diagram of each layer decomposition signal of the PPG signal after discrete wavelet transform.

Fig. 9(a) is a graph showing the trend of the actual blood glucose level.

Fig. 9(b) is a graph of the trend of the amplitude change of the frequency components obtained based on the continuous wavelet transform.

Fig. 9(c) is a graph of the trend of the amplitude change of the frequency components obtained based on the discrete wavelet transform.

Fig. 10(a) is a clark grid error diagram of a single model blood glucose value prediction result based on the non-invasive blood glucose detection method of the present invention.

Fig. 10(b) is a clark grid error diagram showing the result of blood glucose level prediction in a multi-person model based on the noninvasive blood glucose monitoring method according to the present invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

The overall design scheme of the noninvasive blood glucose detection method based on the spectrum analysis is shown as the figure 1, and the specific steps are as follows:

step 1, based on the principle of near infrared spectrum and photoplethysmography, adopting a transmission mode (the transmission mode means that a light source and a receiver are placed at two ends of a detected object, when the light source emits light with a specific wavelength, a photoelectric detector receives the light transmitted through the detected object), converting optical signals transmitted through finger tips into electric signals which are PPG signals by using the photoelectric detector, wherein the signal acquisition device adopts a finger-clipped near infrared pulse wave sensor, and the wavelength of the detection light source is one of sensitive absorption peaks of glucose-1600 nm; the collected PPG signal is processed by a circuit for A/D conversion after being amplified and filtered, and then is sent to an upper computer by a serial port; at the moment, the frequency of the PPG signal sent to the upper computer is generally between 0.05 Hz and 100Hz, main characteristic information is concentrated in frequency information below 20Hz, in order to avoid signal redundancy, the sampling frequency is set to be 200Hz, and a PPG signal acquisition system is shown in figure 2;

step 2, PPG signals directly sent by the serial port not only contain high-frequency noise, but also contain low-frequency baseline drift, and if the low-frequency baseline drift is not removed, the influence on frequency domain analysis of subsequent signals is caused; decomposing an original PPG signal into 8 layers by using a wavelet threshold denoising method through discrete wavelet transform; then, threshold acquisition is carried out on each decomposed wavelet coefficient by utilizing a maximum and minimum threshold principle, and then threshold quantization processing is carried out by combining a soft threshold denoising principle; finally, wavelet reconstruction is carried out on the signals to obtain PPG signals with noise removed based on a wavelet threshold method, FFT is carried out on the original PPG signals and the de-noised signals, the frequency domain shows that the wavelet threshold method de-noising not only effectively removes the high-frequency noise of the signals, but also retains the frequency information of the original signals to the maximum extent, and the de-noising effect is shown in figures 3(a) - (d);

step 3, setting a threshold value after differentiating the denoised PPG signal twice to obtain a valley point of the PPG signal, and referring to fig. 4;

step 4, fitting an envelope curve which is a base line drift of the PPG signal by using a cubic spline interpolation method with a valley point as a base point, and referring to fig. 5; the baseline drift is removed by subtracting the original PPG signal from it, resulting in a pure PPG signal, see fig. 6;

step 5, decomposing the preprocessed pure PPG signal by using continuous wavelet transform and discrete wavelet transform, performing fast Fourier transform on the decomposed wavelet coefficient to obtain a corresponding frequency spectrum, and acquiring PPG signal frequency component information related to blood sugar value change through a large amount of experimental data on the premise of knowing the blood sugar value change; through experimental study, the frequency component amplitude information obtained corresponding to the wavelet coefficient with the scale of 128 after the continuous wavelet transform is related to the blood sugar value change, see fig. 7; the frequency component amplitude information obtained for the detail component of layer 7 after the discrete wavelet transform is based on the blood glucose value variation, see fig. 8;

step 6, performing an OGTT experiment, taking 250ml of the prepared water solution containing 75g of glucose within 5 minutes in a fasting state by the volunteer, performing PPG signal acquisition and real blood glucose value detection in the step 1 every 5 minutes, storing data, and performing the experiment for 90-120 minutes;

step 8, comparing and analyzing the obtained PPG signal frequency component amplitude change information (obtained based on continuous wavelet transform and discrete wavelet transform) related to blood glucose value change and the corresponding change trend of blood glucose value through an OGTT experiment, and finding that the two have certain correlation, see fig. 9(a) - (c); the PPG signal frequency component amplitude value obtained by the algorithm is used as an independent variable, the real blood sugar value is used as a dependent variable, and a single blood sugar model and a multi-person blood sugar model are respectively established for the independent variable and the real blood sugar value by partial least squares regression. According to the detection error requirement of the national quality supervision bureau on the domestic blood glucose meter, when the detected blood glucose concentration value is more than 75mg/dL, the measurement error between the detection result of the domestic blood glucose meter and the actual blood glucose concentration value is allowed to be within +/-20%; when the detected blood glucose concentration value is less than or equal to 75mg/dL, the allowable error range is +/-15 mg/dL. The overall measurement result consistency rate, that is, the detection result satisfying the error tolerance range, should be greater than or equal to 95%. The single model blood glucose value consistency rate predicted by the noninvasive blood glucose detection method is 97.34, the multi-person model consistency rate is 96%, and the two are both more than 95%, and the method meets the national standard. Finally, Clark grid error analysis is performed on the blood glucose values predicted by the noninvasive blood glucose detection method based on the new spectrum analysis, and referring to fig. 10(a) to (b), it can be seen that the prediction results are all in the A, B region, that is, the prediction results have the test precision for guiding clinical reference.

Claims

1. A noninvasive blood glucose detection method based on spectral analysis comprises the following steps:

step 6, performing an OGTT experiment, performing PPG signal acquisition and real blood glucose value detection in the step 1 every 5 minutes, storing data, and performing the experiment for 90-120 minutes;

step 7, after the PPG signals acquired in the step 6 are denoised and baseline drift removed by the methods in the steps 2 to 4, frequency component amplitude information of the PPG signals related to the blood glucose value is acquired by the method in the step 5, and the amplitude information and the real blood glucose value are subjected to partial least squares regression to establish a blood glucose model;