Noninvasive blood glucose detection device based on multi-signal feature fusion

Abstract

The invention discloses a noninvasive blood glucose detection device based on multi-signal feature fusion, which comprises a host and an acquisition probe, and multiple signals comprise a photoacoustic signal and a photoelectric plethysmography (PPG) signal. By performing quality evaluation on the collected PPG signal, the optimal action position of the incident laser on the human body is positioned, so that the laser energy is concentrated in a blood vessel area, and the stability and repeatability of photoacoustic detection are improved. Meanwhile, feature extraction and fusion are performed on the photoacoustic signals and the PPG signals, and a blood glucose prediction regression model is established by utilizing a machine learning algorithm, so that the precision and reliability of noninvasive blood glucose detection are improved.

Description

Technical Field

[0001] This invention belongs to the field of non-invasive blood glucose detection and relates to the field of machine learning. Specifically, it provides a non-invasive blood glucose detection device based on multi-signal feature fusion. Background Technology

[0002] Diabetes is one of the four major chronic diseases threatening human health. Accurate blood glucose testing plays a crucial role in the screening and treatment of diabetes. Traditional blood glucose testing methods use needle pricks, which are not only painful but also carry the risk of infection. Therefore, non-invasive blood glucose testing methods are one of the important future development trends in blood glucose monitoring.

[0003] Currently, some researchers have utilized photoacoustic technology to achieve non-invasive blood glucose detection. For example, patent document CN113876321A (A Non-invasive Blood Glucose Detection Method Based on Photoacoustic Effect) calculates blood glucose concentration based on the difference in photoacoustic signal intensity between different measurement points. However, most in vivo photoacoustic blood glucose detection studies only measure at arbitrary locations within a certain part of the body, and in most cases, only the photoacoustic signal of glucose in tissue fluid is collected. Studies have shown that there are certain differences between blood glucose levels in tissue fluid and blood glucose levels, and there are also differences between different populations and different parts of the same individual, which can affect the accuracy of blood glucose detection. Therefore, when using photoacoustic technology to detect blood glucose, the location of the photoacoustic source generated by the laser within the human body is crucial for the rationality and accuracy of blood glucose measurement.

[0004] To accurately position the laser at human blood vessels, some studies, such as patent documents CN107157491A (A photoacoustic blood glucose detection device and method for automatic blood vessel positioning) and CN110037711A (A photoacoustic precise positioning detection device and method for blood glucose), use scanning to determine whether the incident laser has accurately entered the blood vessel. However, this method requires the use of a scanning platform, and the entire acquisition device is complex in structure and takes a long time to acquire data.

[0005] Photoplethysmography (PPG) is a technique used to measure changes in blood volume at a specific point in the body. A PPG device typically consists of a light source and a pulse wave sensor. Fluctuations in blood volume within blood vessels can be detected by the pulse wave sensor. The quality of the PPG signal acquired by the pulse wave sensor is best when the light source is directly illuminating the blood vessel. Therefore, PPG technology can be used to help locate the light source within the body. Furthermore, since PPG signals reflect information about cardiovascular function, including blood glucose levels, blood glucose can also be detected by extracting relevant features from the PPG signal. However, since human activity levels, skin color, and other factors can interfere with PPG signals and thus affect the accuracy of blood glucose detection, some researchers, when using PPG signals for non-invasive blood glucose detection, usually consider other factors that may be related to PPG signals and blood glucose, in addition to collecting human PPG signals. For example, the patent document with publication number CN117084677A (a multi-dimensional PPG blood glucose value prediction method) uses photoelectric measurement sensors, temperature sensors, and near-infrared light sensors to collect human PPG signals and body temperature data, construct a dataset, and establish a model to predict blood glucose concentration; the patent document with publication number CN113729698A (a non-invasive blood glucose detection method and system) simultaneously collects PPG signals and ECG signals, and uses the extracted PPG signal features and ECG features to complete blood glucose concentration prediction.

[0006] However, both body temperature data and ECG signals show weak correlation with blood glucose concentration, limiting their effectiveness in improving the accuracy of PPG signal-based blood glucose detection. However, when the laser wavelength used is the characteristic absorption wavelength of glucose, we can consider the photoacoustic signal to be strongly correlated with blood glucose concentration. Therefore, by combining photoacoustic and PPG signals to establish a model for predicting blood glucose concentration, the accuracy of non-invasive blood glucose concentration detection can be further improved.

[0007] Based on this, the present invention provides a non-invasive blood glucose detection device based on multi-signal feature fusion. Summary of the Invention

[0008] This invention provides a non-invasive blood glucose detection device based on multi-signal feature fusion, aiming to improve the reliability and accuracy of non-invasive blood glucose detection. This invention simultaneously acquires photoacoustic and PPG signals. On the one hand, by evaluating the quality of the PPG signal, it determines whether the laser source accurately falls into the human blood vessel area, improving the reliability of blood glucose measurement using photoacoustic technology. On the other hand, by fusing the features of photoacoustic and PPG signals, it fully extracts the characteristic information reflecting changes in blood glucose concentration from both signals, improving measurement accuracy.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] A non-invasive blood glucose detection device based on multi-signal feature fusion includes a main unit and a acquisition probe; characterized in that the main unit includes: a laser, a signal generator, a lock-in amplifier, a control module, a data processing module, and an embedded system circuit board; the acquisition probe is used to simultaneously acquire photoacoustic signals and PPG signals.

[0011] Furthermore, the laser is a power-stable semiconductor continuous wave laser, and it can be square wave intensity modulated to output lasers with different modulated light intensities; the wavelength of the laser is the characteristic absorption wavelength of glucose.

[0012] Furthermore, the control module is used to control the connection between the signal generator and the lock-in amplifier and the embedded system circuit board; and to drive the signal generator to output two different frequency square wave signals with a duty cycle of 50% in succession, wherein the low frequency signal has a frequency range of 50Hz to 400Hz and the high frequency signal has a frequency range of 350kHz to 400kHz.

[0013] Furthermore, the acquisition probe includes: a pulse wave sensor, a laser collimator, a focusing lens, and an ultrasonic transducer. The pulse wave sensor and the ultrasonic transducer are distributed on both sides of the tissue to be tested, and the pulse wave sensor and the laser beam are located on the same side of the tissue to be tested.

[0014] The present invention also provides a blood glucose detection method based on the aforementioned non-invasive blood glucose detection device, comprising the following steps:

[0015] Step 1: Fix the acquisition probe to the tissue to be tested, connect the control module to the signal generator and the embedded system circuit board, and the embedded system circuit board starts acquiring PPG signals;

[0016] Step 2: The data processing module evaluates the PPG signal quality, moves the acquisition probe according to the quality evaluation index, adjusts the laser incident position, determines the optimal acquisition area, and the data processing module begins recording the PPG signal.

[0017] Step 3: The control module disconnects the signal generator from the embedded system circuit board, and then connects the signal generator to the lock-in amplifier to start acquiring photoacoustic signals;

[0018] Step 4: The data processing module integrates the features of PPG signal and photoacoustic signal, inputs them into the blood glucose prediction regression model to obtain the blood glucose prediction result.

[0019] Furthermore, the blood glucose prediction regression model is established through the following steps:

[0020] 1) Collect blood glucose values ​​from M subjects using traditional medical equipment, and use this as a blood glucose label set S=(s1, s2, …s m ), where sm Let m be the blood glucose value of the m-th subject, where m = 1, 2 … M;

[0021] 2) Based on steps 1, 2 and 3 above, collect PPG and photoacoustic signals from M subjects;

[0022] 3) Perform morphological analysis on the PPG signal, extract PPG morphological features, and calculate the amplitude and peak-to-peak value of the photoacoustic signal to form a sample set P=(P1, P2, … P m ), where P m For the data features extracted from the m-th subject, P m =(t 1m , t 2m ,… t nm , p m1 , p m2 ) T , where t nm For the morphological characteristics of the nth PPG of the mth subject, p m1 p represents the amplitude of the photoacoustic signal of the m-th subject. m2 The peak-to-peak value of the photoacoustic signal of the m-th subject;

[0023] 4) Using the sample set P and the label set S as training datasets, a blood glucose prediction regression model is trained using machine learning methods.

[0024] This invention provides a non-invasive blood glucose detection device based on multi-signal feature fusion, aiming to solve the problems of inaccurate blood vessel location during photoacoustic blood glucose detection and the weakened correlation between PPG signals and blood glucose levels due to interference, thereby improving the accuracy of non-invasive blood glucose detection. Compared with existing technologies, the advantages of this invention are:

[0025] (1) This invention achieves precise positioning of laser in the human blood vessel area by evaluating the quality of the collected PPG signal, avoiding the limitations of traditional photoacoustic technology in blood glucose detection that relies on large instruments such as scanning platforms to locate blood vessels, simplifying the system structure and shortening the data acquisition time.

[0026] (2) The acquisition process of this invention is simple. After the PPG signal acquisition is completed, no device needs to be replaced or moved to complete the acquisition of photoacoustic signal and obtain real blood glucose photoacoustic information in blood vessels.

[0027] (3) This invention further explores the factors related to blood glucose concentration in human blood, extracts and fuses the multidimensional features of photoacoustic signal and PPG signal as input to the blood glucose prediction regression model. Compared with the method of predicting blood glucose using only a single signal of photoacoustic signal or PPG signal, the accuracy of blood glucose detection is higher. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the non-invasive blood glucose detection device based on multi-signal feature fusion in this invention.

[0029] Figure 2 This is a schematic diagram of the acquisition probe in the non-invasive blood glucose detection device based on multi-signal feature fusion in this invention.

[0030] Figure 3 This is a schematic diagram of a high-quality PPG signal acquired by a non-invasive blood glucose detection device based on multi-signal feature fusion in this invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0032] From the perspective of measurement principles:

[0033] One of the main theoretical bases of this invention is the photoacoustic effect, and the equation for the sound wave generated by the photoacoustic effect can be expressed by the following formula:

[0034] (1)

[0035] Where P represents the sound pressure of the photoacoustic signal, v represents the speed of sound in the medium, t represents time, α represents the light absorption coefficient of the medium, β is the volumetric thermal expansion coefficient of the medium, and C p Let P represent the specific heat capacity of the medium, and I represent the incident laser light intensity. When the light absorption of the medium is weak, the sound pressure P can be expressed as:

[0036] (2)

[0037] Where c represents the system constant and E0 represents the incident laser energy, it can be seen that the sound pressure P is proportional to the physical properties of the medium:

[0038] (3)

[0039] Therefore, when other parameters are constant, if the laser wavelength corresponds to the characteristic absorption wavelength of the analyte, changes in the concentration of the analyte mainly affect the light absorption coefficient α, and thus the sound pressure P. Based on this, it is feasible to select the laser source wavelength as the characteristic absorption wavelength of glucose, and then deduce the blood glucose concentration of the human body by measuring the photoacoustic signal of a certain part of the body.

[0040] The physical basis for PPG signal measurement is the Lamb-Beer law, as shown in the following formula:

[0041] (4)

[0042] Where A is absorbance, and I... out Represents the emitted light intensity, I in Here, A represents the incident light intensity, T is the transmittance, K is the absorption coefficient, b is the thickness of the absorbing medium, and c is the medium concentration. The blood in blood vessels changes periodically with the heart's diastolic and systolic rhythms, and the thickness b of the absorbing medium also changes accordingly. Since the absorption coefficient K does not change at a given wavelength, the change in absorbance A depends on the change in b, i.e., it is related to the changes in blood flow within the blood vessels. When the incident light irradiates a blood vessel region, the change in absorbance A is most pronounced, and the corresponding PPG signal quality is optimal. Based on this, by analyzing the changes in PPG signal quality, it is possible to determine whether the incident light has irradiated a blood vessel region, thereby achieving precise location of blood vessels in the human body.

[0043] Based on the above principles, this embodiment provides a non-invasive blood glucose detection device based on multi-signal feature fusion, the detection device as follows: Figure 1 As shown, it includes: a host and a data acquisition probe; wherein, the host includes a laser, a signal generator, a lock-in amplifier, a control module, a data processing module, and an embedded system circuit board.

[0044] Furthermore, the structure of the acquisition probe is as follows: Figure 2 As shown, the device used for simultaneous acquisition of photoacoustic and PPG signals includes: a pulse wave sensor, a laser collimator, a focusing lens, and an ultrasonic transducer. The pulse wave sensor and the laser beam are located on the same side of the tissue being tested to acquire reflective PPG signals, while the ultrasonic transducer is located on the other side of the tissue to receive photoacoustic signals.

[0045] The specific working process is as follows, and the tissue to be tested is generally the earlobe or fingertip;

[0046] 1. Vessel localization and PPG signal acquisition

[0047] Under the control of the control module, the signal generator first generates a low-frequency square wave signal with a frequency range of 50Hz to 400Hz and a duty cycle of 50%. This square wave signal is used as a sampling trigger signal input to the embedded system circuit board to control the acquisition of PPG signals. During this stage, the laser maintains a constant output power to irradiate the tissue under test. The data processing module displays the acquired PPG waveform and its quality evaluation indicators in real time, and adjusts the position of the acquisition probe on the surface of the tissue under test according to changes in signal quality until a high-quality PPG signal with a stable waveform and clear morphology is obtained. Figure 3 As shown in the image. This indicates that the laser incident position is located in or near the blood vessel region. Subsequently, the data processing module begins recording PPG signal data.

[0048] 2. Acquisition of photoacoustic signals

[0049] After vascular localization, the control module disconnects the signal generator from the embedded system circuit board and controls the signal generator to produce a high-frequency square wave signal with a frequency range of 350kHz to 400kHz and a duty cycle of 50%. This high-frequency square wave signal is split into two paths: one path serves as a reference signal input to the lock-in amplifier, and the other path connects to the laser to achieve laser intensity modulation. The modulated laser then passes through a laser collimator and a focusing lens before irradiating the surface of the skin tissue to be tested. To improve the transmission efficiency of the photoacoustic signal, a medical ultrasound coupling agent is coated on the surface of the test area to enhance the acoustic coupling with the ultrasound transducer. The ultrasound transducer located on the other side of the test tissue receives the photoacoustic signal generated after the tissue absorbs the laser energy and transmits it to the lock-in amplifier. The lock-in amplifier outputs the processed signal to the data processing module.

[0050] 3. Data Processing

[0051] The data processing module extracts features from the PPG and photoacoustic signals and inputs them into the blood glucose prediction regression model to obtain blood glucose prediction results.

[0052] Furthermore, the blood glucose prediction regression model is established through the following steps:

[0053] 1) Collect blood glucose values ​​from M subjects using traditional medical equipment, and use this as a blood glucose label set S=(s1, s2, …s m ), where s m Let m be the blood glucose value of the m-th subject, where m = 1, 2 … M;

[0054] 2) Based on the above working process, collect the PPG signal and photoacoustic signal of the subject;

[0055] 3) Perform morphological analysis on the PPG signal, extract PPG morphological features, and calculate the amplitude and peak-to-peak value of the photoacoustic signal to form a sample set P=(P1, P2, … P m ), where P m For the data features extracted from the m-th subject, P m =(t 1m , t 2m ,… t nm , p m1 , p m2 ) T , where t nm For the nth PPG morphological feature of the mth subject, p m1 p represents the amplitude of the photoacoustic signal of the m-th subject. m2 The peak-to-peak value of the photoacoustic signal of the m-th subject;

[0056] 4) Using the sample set P and the label set S as training datasets, a blood glucose prediction regression model is trained using machine learning methods.

[0057] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. A non-invasive blood glucose detection device based on multi-signal feature fusion, comprising a main unit and a acquisition probe; characterized in that, The main unit includes: a laser, a signal generator, a lock-in amplifier, a control module, a data processing module, and an embedded system circuit board; the acquisition probe is used to simultaneously acquire photoacoustic signals and PPG signals.

2. The non-invasive blood glucose detection device based on multi-signal feature fusion according to claim 1, characterized in that, The laser is a power-stable semiconductor continuous wave laser, and it can be square wave intensity modulated to output lasers with different modulated light intensities; the wavelength of the laser is the characteristic absorption wavelength of glucose.

3. The non-invasive blood glucose detection device based on multi-signal feature fusion according to claim 1, characterized in that, The control module is used to control the signal generator to generate square wave signals of different frequencies with a duty cycle of 50% in succession. The low-frequency signal has a frequency range of 50Hz to 400Hz, and the high-frequency signal has a frequency range of 350kHz to 400kHz.

4. The non-invasive blood glucose detection device based on multi-signal feature fusion according to claim 1, characterized in that, The acquisition probe includes: a pulse wave sensor, a laser collimator, a focusing lens, and an ultrasonic transducer; wherein the pulse wave sensor and the ultrasonic transducer are distributed on both sides of the tissue to be tested, and the pulse wave sensor and the laser beam are located on the same side of the tissue to be tested.

5. A non-invasive blood glucose detection method based on multi-signal feature fusion, implemented based on the non-invasive blood glucose detection device based on multi-signal feature fusion as described in claim 1, characterized in that, Includes the following steps: Step 1: Fix the acquisition probe to the tissue to be tested, connect the control module to the signal generator and the embedded system circuit board, and the embedded system circuit board starts acquiring PPG signals; Step 2: The data processing module evaluates the PPG signal quality, adjusts the laser incident position of the acquisition probe according to the signal quality evaluation index, determines the optimal acquisition area, and the data processing module begins to record the PPG signal. Step 3: The control module switches the signal channel, connects the signal generator and the lock-in amplifier, collects the photoacoustic signal, and sends it to the data processing module; Step 4: The data processing module integrates the features of photoacoustic signals and PPG signals, inputs them into the blood glucose prediction regression model, and obtains the blood glucose prediction results.

6. The non-invasive blood glucose detection method based on multi-signal feature fusion according to claim 5, characterized in that, The blood glucose prediction regression model is established through the following steps: 1) Blood glucose levels of M subjects were collected using a medical blood glucose meter, and used as a blood glucose label set S = (s1, s2, ... s... m ), where s m Let m be the blood glucose value of the m-th subject, where m = 1, 2 … M; 2) According to steps 1, 2 and 3 in claim 5, collect PPG signals and photoacoustic signals from M subjects; 3) Perform morphological analysis on the PPG signal, extract PPG morphological features, and calculate the amplitude and peak-to-peak value of the photoacoustic signal to form a sample set P=(P1, P2, … P m ), where P m For the data features extracted from the m-th subject, P m =(t 1m , t 2m , …t nm , p m1 , p m2 ) T , where t nm For the nth PPG morphological feature of the mth subject, p m1 p represents the amplitude of the photoacoustic signal of the m-th subject. m2 The peak-to-peak value of the photoacoustic signal of the m-th subject; 4) Using the sample set P and the label set S as training datasets, a blood glucose prediction regression model is trained using machine learning methods.

Images