Noninvasive blood glucose detection device and method

Abstract

The invention discloses a noninvasive blood glucose detection device and method. The noninvasive blood glucose detection device comprises a radar transmitter, a radar receiver, a data acquisition module, a blind source separation module, a feature extraction module and a blood glucose evaluation module, a broadband radio frequency modulation signal emitted by the radar transmitter irradiates a certain human tissue, a reflected radio frequency signal is received by the radar receiver and then is converted into an analog baseband signal, and the analog baseband signal is sampled by the data acquisition module and then is converted into a digital signal. After the electromagnetic waves are transmitted to the blind source separation module for noise filtering, interference removal and data enhancement processing, feature information of the electromagnetic waves in the subcutaneous tissue blood transmission process is extracted, and then blood glucose concentration information is obtained through machine learning according to the feature information. When the device is used for measuring blood glucose, test paper or a reagent is not needed, and low detection cost, zero pollution, comfort and no feeling are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical devices, and particularly relates to a non-invasive blood glucose detection device and method. BACKGROUND

[0002] The current blood glucose measurement methods are mainly biochemical blood glucose measurement method (clinical diagnosis) and invasive blood glucose measurement method (daily monitoring), and the two technologies are relatively mature and are the main methods for blood glucose measurement at present. However, both the two methods need blood sampling for detection, and blood sampling or blood taking by finger pricking needle will cause pain and risk of infection, and also cause psychological fear and resistance. This limits the frequency of blood glucose measurement, and makes the diabetic patients unable to obtain satisfactory blood glucose monitoring, so it is urgent to adopt non-invasive blood glucose measurement technology to overcome the shortcomings of invasive blood sampling. The non-invasive blood glucose measurement technology has more obvious advantages: no pain and trauma to the human body, no risk of infection, no need for consumables such as test paper, long-term use, low additional cost, and continuous monitoring, which is conducive to better control of the condition of diabetic patients. Therefore, in recent years, the non-invasive blood glucose measurement technology has become a research hotspot in the field of blood glucose measurement. According to the detection principle, the non-invasive blood glucose measurement technology is mainly divided into optical method (near-infrared spectroscopy, Raman spectroscopy and fluorescence method), microwave method, electrochemical method and metabolic heat integration method.

[0003] Near-infrared spectroscopy - near-infrared light has relatively strong ability to penetrate biological fluids and soft tissues, and the glucose absorption peak on the absorption spectrum of transmitted light and reflected light can quantitatively infer the glucose concentration. However, the infrared sensor receives a weak spectrum signal intensity related to blood glucose, and there are many other interference factors in the tissue, which will produce similar near-infrared spectrum of glucose, resulting in inaccurate measurement.

[0004] Raman spectroscopy - when a laser of a certain frequency hits the surface of a sample, the molecules in the material will absorb part of the energy to vibrate in different ways and degrees, and scatter light of different frequencies. This process is Raman scattering. Different molecules will produce scattering light of a specific frequency, and the glucose concentration can be quantitatively analyzed based on this scattering light. The disadvantage is that the Raman spectrum of low-concentration glucose is easily covered by the strong background noise of the surrounding environment, resulting in inaccurate detection.

[0005] Fluorescence method - the process of decay of a molecule from a high-energy excited state back to a ground state and emission of a photon is called radiative relaxation phenomenon. Radiative relaxation produces fluorescence and phosphorescence, and there are a large number of fluorescent groups in human blood which can emit fluorescence when they are in an excited state. The fluorescence sensor characterizes the amount of glucose by detecting the degree of fluorescence quenching. This method can only measure body fluids such as tears and sweat, and the correlation of these body fluids with blood glucose is poor, and there are also problems of low signal-to-noise ratio, detection lag and inability to quantify.

[0006] Microwave method - using the strong penetration and low diffraction interference characteristics of microwaves, based on the principle that the reflection, transmission or absorption of electromagnetic waves is closely related to the dielectric properties of the tissue. The dielectric constant of blood changes with the fluctuation of glucose concentration therein. The poor sensitivity and selectivity of conventional microwave sensors limit their detection accuracy. In addition to the sensitivity problem, it is also susceptible to noise and interference. Similarly, selectivity is also a problem, because the radio frequency signal wave must penetrate the skin, fat and blood vessel layer before reaching the blood - all of which have different dielectric properties, so complex signal post-processing is required to reduce the interference introduced by these other tissues.

[0007] Electrochemical method - by applying a weak current to the skin, glucose is extracted from under the skin for detection. The back of this blood glucose meter is in contact with the human body through a layer of gel pad. There are two electrodes in the gel, which are connected by an electric circuit when in use, generating a micro-current through the skin of the human body. The charged ions in the skin move to the positive and negative electrodes under the action of the current, while the glucose molecules in the interstitial fluid are "wrapped" together with the charged ions and enter the gel. By measuring the reaction degree of glucose molecules and an enzyme (glucose oxidase) in the gel, the current blood glucose level can be calculated. The defects of this technology are: the current intensity for extracting blood glucose is sufficient to cause damage to the skin; it is very difficult to extract blood glucose from under the skin, and the extraction of glucose itself causes related chain reactions in the body, making the measured glucose content inaccurate.

[0008] Metabolic heat integration method - blood glucose can be calculated by testing physiological parameters related to human metabolism and basic physiological information of the human body. The testable parameters include environmental temperature, environmental humidity, human body surface temperature and humidity, blood flow rate, and blood oxygen saturation. Based on the principle of metabolic heat integration, multi-sensor integration technology is used to realize non-invasive measurement of blood glucose. The defect is that the blood glucose detection is indirectly calculated from the human metabolism related parameters, and the detection accuracy is limited. SUMMARY

[0009] Invention purpose: The purpose of the present application is to provide a non-invasive blood glucose detection device and method, which can realize continuous real-time monitoring of blood glucose level, and can send the detection report to the mobile phone or cloud platform through Wifi and 5G network.

[0010] Technical scheme: A non-invasive blood glucose monitoring device, comprising a radar transmitter, a radar receiver, a data acquisition module, a blind source separation module, a feature extraction module and a blood glucose evaluation module. The radar transmitter is used to provide radar radio frequency transmission signals, which are radiated to space through a feed line and an antenna. The radar receiver is used for pre-selection, amplification, frequency conversion, filtering, demodulation and digitization of weak signals received by the radar antenna, while suppressing external interference, clutter and internal noise. The data acquisition module is used for converting the analog baseband signal of the radar receiver into a digital signal convenient for the processor to process. The blind source separation module is used for separating each independent signal from the mixed signal of echo signals of multiple radiation sources containing targets and interference, noise and multipath reflection signals. The feature extraction module is used for extracting reflection and transmission information of electromagnetic waves in the human body from the output signal of the blind source separation module, and calculating blood dielectric constant and viscosity according to the reflection and transmission information. The blood glucose evaluation module is used for obtaining the transmission and reflection coefficients of electromagnetic wave signals in the blood transmission process according to the reflection and transmission information obtained by the feature extraction module and the measured glucose solution, and realizing the classification and evaluation of human blood glucose concentration by machine learning. The radar transmitter emits a wideband radio frequency modulated signal to irradiate a certain human tissue, and the reflected radio frequency signal is received by the radar receiver and is down-converted into an analog baseband signal. The analog baseband signal is sampled by the data acquisition module and converted into a digital signal, which is transmitted to the blind source separation module for filtering noise, removing interference and data enhancement processing, and then the feature information of electromagnetic waves in the subcutaneous tissue blood transmission process is extracted. Then, the blood glucose concentration information is obtained by machine learning according to the feature information.

[0011] Further, the electromagnetic wave signal transmission path information carrying the feature information of the dielectric properties of human blood includes transmission waves and reflection waves.

[0012] Further, the high-frequency electromagnetic wave emitted by the radar transmitter irradiates the human tissue, penetrates the epidermis layer and the dermis layer in turn, and is reflected out after entering the subcutaneous tissue layer.

[0013] A non-invasive blood glucose detection method based on any one of the above non-invasive blood glucose monitoring devices, comprising the following steps: S1, the radar receiver down-converts the received radio frequency signal into an analog baseband signal; S2, the data acquisition module converts the analog baseband signal into a digital signal and transmits it to the blind source separation module; S3, the blind source separation module extracts the feature information of electromagnetic waves in the subcutaneous tissue fluid transmission process; S4, input the extracted feature information into the teacher-student network model to obtain the blood glucose concentration information.

[0014] Further, the local oscillator signal of the radar receiver is the same as that of the radar transmitter, and the baseband signal generated by the radar signal transmission to the detection object and returned to the receiver is a fixed intermediate frequency signal.

[0015] Further, in the teacher-student network model, the top pipeline is a teacher network, which measures based on different concentrations of glucose solution instead of blood glucose concentration, adopts a convolutional neural network algorithm for classification machine learning training, and fits the relationship between the corrected dielectric constant and the glucose concentration; the bottom pipeline is a student network, which performs blood glucose concentration estimation based on a wideband radar; The teacher-student network model is trained based on a machine learning algorithm for glucose concentrations in the range of 2.2-18 mmol / L, and the best teacher-student network model is obtained for estimating the glucose concentration.

[0016] Compared with the prior art, the present application has the following significant effects: 1. The device of the present application uses a radar sensor to measure blood glucose in close proximity without contact, which is non-invasive measurement, especially for people with damaged skin or pain sensitivity. 2. The device of the present application calculates the thickness of each layer of skin medium and the dielectric constant of blood based on the energy emitted by the radar transmitter and the energy received by the receiver, and further obtains the blood glucose concentration information. Since no test paper or reagent is required for measuring blood glucose, the detection cost is low, zero pollution and comfortable. 3. The method of the present application is based on high-frequency wideband electromagnetic wave signal detection, so the measurement speed is fast and the accuracy is high. Real-time measurement data can be sent to a designated mobile phone or cloud platform through WiFi or 4 / 5G network, which is easy for remote diagnosis and treatment or health monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 Figure 1 is a structural schematic diagram of the detection device of the present application; Figure 2 Figure 2 is a human tissue skin structure diagram; Figure 3 Figure 3 is a schematic diagram of an electromagnetic wave detection human blood glucose model; Figure 4 Figure 4 is a schematic diagram of the radar transmitter output signal waveform; Figure 5 Figure 5 is a schematic diagram of the teacher-student network model for blood glucose detection; Figure 6 Figure 6 is a machine learning model flowchart of the detection algorithm. DETAILED DESCRIPTION

[0018] The present application will be further described in detail below in conjunction with the drawings and specific embodiments of the specification.

[0019] As Figure 1The non-invasive blood glucose monitoring device is shown in the schematic diagram, including a radar transmitter, a radar receiver, a data acquisition module, a blind source separation module, a feature extraction module and a blood glucose evaluation module. When the wideband radio frequency modulation signal emitted by the radar transmitter irradiates a certain human tissue (such as a wrist), the reflected radio frequency signal is received by the radar receiver and is down-converted into an analog baseband signal. The analog baseband signal is sampled by the data acquisition module and is converted into a digital signal, which is transmitted to a data processing unit for noise filtering, interference removal, data enhancement and other processing. Then, the characteristic information of the electromagnetic wave in the subcutaneous tissue fluid transmission process is extracted, and the blood glucose concentration information is obtained through machine learning according to the characteristic information.

[0020] The radar transmitter is an important part of the radar system, which is used to provide a radar radio frequency transmission signal meeting a certain specification, which is radiated to space through a feeder and an antenna.

[0021] The radar receiver is also an important part of the radar system, which is used to pre-select, amplify, convert, filter, demodulate and digitize the weak signal received by the radar antenna, while suppressing external interference, clutter and internal noise, so that the echo signal can maintain the characteristic information of the target as much as possible.

[0022] The data acquisition module is an analog-to-digital converter (ADC module), which is used to convert the analog baseband signal of the radar receiver into a digital signal convenient for processor processing, and its performance directly determines the precision and fidelity of the information that the detection system can obtain.

[0023] The blind source separation module is used to separate each independent signal from the mixed signal of the echo signal of multiple radiation sources containing targets and interference, noise and multi-path reflection signal. The single-channel or multi-channel radar array received signal is first pre-processed by centering and whitening, different algorithms are selected according to the signal characteristics and prior knowledge for iterative optimization, and after the output separation signal is corrected and screened, the target signal can be retained and the interference signal can be suppressed.

[0024] The feature extraction module is used to extract the reflection and transmission information of electromagnetic waves in the human body from the output signal of the blind source separation module, calculate the blood dielectric constant, viscosity and other characteristic information according to the reflection and transmission information, and thus infer the blood composition and proportion.

[0025] The blood glucose evaluation module is used to obtain the transmission and reflection coefficient characteristic information of the electromagnetic wave signal in the blood medium according to the reflection and transmission information obtained by the feature extraction module and the measured glucose solution (or blood), and to realize the classification and evaluation of the human blood glucose concentration by using supervised learning and deep learning machine learning.

[0026] The detection device of the present application uses a wideband radar transceiver to non-contact close-range sense the blood glucose level of the human body, such as Figure 1As shown, the radar transmitter emits a wideband radio frequency modulated signal to irradiate a certain human tissue (such as a wrist), and the reflected radio frequency signal is converted into an analog baseband signal after being received by the radar receiver, the analog baseband signal is sampled by the data acquisition module and converted into a digital signal, and then transmitted to the blind source separation module for noise filtering, interference removal, data enhancement and other processing, and then the transmission and reflection of electromagnetic waves in the subcutaneous tissue fluid / blood are extracted, and then the blood glucose concentration information is obtained through machine learning according to the characteristic information.

[0027] As shown in the human tissue skin structure such as Figure 2 As shown, it is divided into epidermis, dermis and subcutaneous tissue layer, and there are a large number of blood vessels and capillaries in the subcutaneous tissue layer, and the blood vessels and interstitial fluid of the tissue layer contain a large number of glucose molecules, so the subcutaneous tissue fluid and blood contain information characteristics for detecting blood glucose concentration. The high-frequency electromagnetic wave emitted by the radar transmitter irradiates the human tissue, penetrates the epidermis, dermis and subcutaneous tissue layer in turn, and is reflected out again, and the electromagnetic wave detection human blood glucose model can be simplified as Figure 3 As shown. According to the energy emitted by the radar transmitter and the energy received by the receiver, the thickness d x and dielectric constant ε x of each layer of skin medium are calculated, wherein x=1 is the epidermis, x=2 is the dermis, and x=3 is the subcutaneous layer.

[0028] The blood glucose detection method of the application comprises the following steps: Step 1, the radar transmitter emits a specific waveform signal to the detection position at a close range, and the radar receiver converts the received radio frequency signal into an analog baseband signal; The working frequency of the radar transmitter is Fstart to Fstop, and the output waveform adopts a frequency hopping auxiliary frequency modulation continuous wave (FH-FMCW) modulation technology, which can effectively reduce interference. The receiver local oscillator signal is the same as the transmitter, and the baseband signal generated by the radar signal transmitted to the detection object to the receiver is a fixed intermediate frequency signal.

[0029] Step 2, the data acquisition module converts the analog baseband signal into a digital signal and transmits it to the data processing unit; The data acquisition module collects 256 frames of waveforms per channel each time, each frame of data contains 128 chirp signals, and 64 data are collected for each chirp signal, Figure 4 It is a frame waveform schematic diagram.

[0030] Step 3, the data processing unit extracts the characteristic information of electromagnetic waves in the subcutaneous tissue fluid or blood transmission process; The data processing unit is implemented by FPGA or DSP, and includes a blind source separation module and a feature extraction module; the blind source separation module is based on single-channel or multi-channel radar array receiving signals, which are first subjected to centering and whitening preprocessing, and then subjected to iterative optimization processing according to signal characteristics and prior knowledge by using independent component analysis (based on high-order statistics), second-order statistics-based algorithm or hybrid optimization algorithm, so as to suppress interference signals and output pure target signals. The feature extraction module extracts feature information of the transmission process of electromagnetic waves in the subcutaneous tissue fluid or blood of the human body according to the output signals of the blind source separation module, and the information carries the information of the dielectric constant of the human tissue fluid or blood.

[0031] Step 4: input the extracted feature information into the teacher-student network model to obtain blood glucose concentration information; As shown in Figure 5 , it is a schematic diagram of the teacher-student network model. The top pipeline is the teacher network, which is based on different concentrations of glucose solution instead of blood with different blood glucose concentrations for measurement, uses convolutional neural network algorithm (CNN) for classification machine learning training, and fits the relationship between dielectric constant and glucose concentration. It provides supervision for human blood glucose measurement. The bottom pipeline is the student network, which performs broadband radar-based blood glucose concentration estimation.

[0032]

[0033] In the above formula, is the dielectric constant of the liquid to be measured, ε s is the static dielectric constant, ε ∞ is the high-frequency dielectric constant, is the vacuum dielectric constant, τ is the relaxation time, and σ is the ion conductivity. The static dielectric constant, the high-frequency dielectric constant and the relaxation time are all empirical constants related to the purity of the liquid to be measured.

[0034] The data collected by the teacher network is derived from glucose solution, and the device of the present application is used to measure multiple glucose solution samples with different concentrations at close range and non-contact. The main goal is to train the teacher-student network model based on machine learning algorithm for 64 glucose concentrations in the range of 2.2-18 mmol / L, i.e. the resolution is 0.25 mmol / L, to obtain the best teacher-student network model parameters for estimating glucose concentration. The machine learning model flowchart of the detection algorithm is shown in Figure 6 . The data collected by the student network is derived from the subcutaneous tissue fluid or blood of the human body to be detected. After using the teacher network training model to supervise the learning of the related data of the real human blood collected by the device of the present application, the human blood glucose concentration is inferred.

Claims

1. A non-invasive blood glucose monitoring device, characterized in that, It includes a radar transmitter, a radar receiver, a data acquisition module, a blind source separation module, a feature extraction module, and a blood glucose assessment module; The radar transmitter is used to provide radar radio frequency transmission signals, which are radiated into space via feeders and antennas; The radar receiver is used to pre-select, amplify, frequency convert, filter, demodulate and digitize the weak signal received by the radar antenna, while suppressing external interference, clutter and internal noise. The data acquisition module is used to convert the analog baseband signal of the radar receiver into a digital signal that is easy for the processor to process; The blind source separation module is used to separate individual signals from the mixed signal of echo signals, noise and multipath reflection signals of multiple radiation sources containing targets and interference. The feature extraction module is used to extract the real information of electromagnetic wave reflection and transmission in the human body from the output signal of the blind source separation module, and calculate the blood dielectric constant and viscosity based on the reflection and transmission information. The blood glucose assessment module is used to obtain the transmission and reflection coefficients of electromagnetic wave signals during blood transmission based on the reflection and transmission information obtained by the feature extraction module and the measured glucose solution, and to use machine learning to classify and assess human blood glucose concentration. The wideband radio frequency modulation signal emitted by the radar transmitter illuminates a certain human tissue. The reflected radio frequency signal is received by the radar receiver and down-converted into an analog baseband signal. This analog baseband signal is sampled by the data acquisition module and converted into a digital signal. It is then transmitted to the blind source separation module for noise filtering, interference removal, and data enhancement processing. The feature information of the electromagnetic wave transmission process in the subcutaneous tissue blood is extracted, and then the blood glucose concentration information is obtained through machine learning based on this feature information.

2. The non-invasive blood glucose monitoring device according to claim 1, characterized in that, The transmission path information of electromagnetic wave signals carrying information about the dielectric properties of human blood includes transmitted waves and reflected waves.

3. The non-invasive blood glucose monitoring device according to claim 1, characterized in that, The high-frequency electromagnetic waves emitted by the radar transmitter irradiate human tissue, penetrating the epidermis and dermis in sequence, entering the subcutaneous tissue layer, and then being reflected back out.

4. A non-invasive blood glucose detection method, based on the non-invasive blood glucose monitoring device as described in any one of claims 1-3, characterized in that, The steps include the following: S1, the radar receiver down-converts the received radio frequency signal into an analog baseband signal; S2, the data acquisition module converts the analog baseband signal into a digital signal and transmits it to the blind source separation module; S3, the blind source separation module extracts the characteristic information of electromagnetic wave transmission process in subcutaneous tissue fluid; S4. Input the extracted feature information into the teacher-student network model to obtain blood glucose concentration information.

5. The non-invasive blood glucose detection method according to claim 4, characterized in that, The local oscillator signal of the radar receiver is the same as that of the radar transmitter. The baseband signal generated by the radar signal transmitted to the detected object and returned to the receiver is a fixed intermediate frequency signal.

6. The non-invasive blood glucose detection method according to claim 4, characterized in that, In the teacher-student network model, the top pipeline is the teacher network. It uses glucose solutions of different concentrations to measure blood glucose concentration instead of blood glucose concentration. It employs a convolutional neural network algorithm for classification machine learning training to fit and correct the relationship between dielectric constant and glucose concentration. The bottom pipe is the student network, which performs broadband radar-based blood glucose concentration estimation; Based on machine learning algorithms, a teacher-student network model was trained on glucose concentrations in the range of 2.2–18 mmol / L to obtain the optimal teacher-student network model for estimating glucose concentration.

Images