Novel blood glucose monitoring method based on optical coherence tomography technology

Abstract

The invention discloses a novel blood glucose monitoring method based on an optical coherence tomography technology, which comprises the following steps of: reconstructing and positioning a blood vessel by using a Doppler optical coherence tomography technology, extracting a near infrared spectrum of the blood vessel through windowing Fourier transform, and establishing a model of spectral data and blood glucose concentration by using a partial least square method; therefore, nondestructive blood glucose monitoring is realized. By combining the spectrum technology and Doppler optical coherence tomography, the interference of the signal slope of the optical coherence tomography in blood glucose monitoring due to motion artifacts can be effectively overcome, and the hysteresis effect during measurement of the intercellular fluid glucose concentration is solved, so that the accuracy and the stability of real-time blood glucose monitoring are improved. The method is the intersection of advanced instruments and clinical medical diagnosis, has good social application value and application prospect, can effectively provide guidance for diagnosis of diabetes and design of a disease control scheme, and can prevent patients from suffering from pain caused by frequent blood sampling.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing, relates to a method for processing OCT images, and particularly relates to a novel blood glucose monitoring method based on optical coherence tomography technology. Background Art

[0002] Optical Coherence Tomography (OCT) is a high-resolution imaging technology based on low-coherence interference, which can accurately image the two-dimensional or three-dimensional structure of biological tissues in a non-invasive manner. Its high sensitivity and real-time performance have made it widely used in medical fields such as ophthalmology, cardiology, and dermatology. In recent years, researchers have begun to explore the possibility of using OCT technology for non-invasive blood glucose monitoring and obtaining blood glucose-related information using its imaging characteristics.

[0003] The application of OCT in blood glucose monitoring mainly relies on its sensitivity to changes in tissue optical properties and its ability to perform real-time analysis of hemodynamics. On the one hand, changes in blood glucose concentration can affect the refractive index and scattering characteristics of blood and tissues, and OCT can capture these minute changes to indirectly infer blood glucose concentration. On the other hand, Doppler OCT technology can analyze blood flow information in blood vessels in real time, providing a basis for dynamic monitoring. In addition, the non-invasive feature of OCT avoids the blood sampling process in traditional blood glucose monitoring and is particularly suitable for patients who need to measure blood glucose frequently.

[0004] However, the application of OCT in blood glucose monitoring also faces some technical challenges. First, patient movement may cause signal distortion, i.e., motion artifacts, which affect the measurement accuracy. Second, when measuring the glucose concentration in interstitial fluid based on optical methods, there may be a time lag relative to the change in blood glucose, i.e., the hysteresis effect, which limits the real-time performance of the monitoring results. In addition, since the optical signal changes caused by blood glucose concentration changes are relatively weak, higher requirements are imposed on the device sensitivity and signal processing technology.

[0005] To overcome these challenges, researchers have proposed an innovative method of combining OCT with spectroscopy technology. Absorption and scattering information related to blood glucose concentration is extracted through near-infrared spectroscopy technology to supplement the limitations of traditional OCT imaging in biochemical parameter detection. In addition, using Doppler OCT technology to accurately locate the blood flow area in blood vessels can effectively reduce the error caused by motion artifacts. By combining Fourier transform and machine learning modeling technology, the spectral data is correlated with blood glucose concentration, greatly improving the measurement accuracy and stability.

[0006] Currently, the OCT-based non-invasive blood glucose monitoring technology is still in the research stage. However, its characteristics of non-invasiveness, high resolution, and real-time monitoring have shown broad prospects. In the future, with the improvement of device hardware, algorithm optimization, and the development of portable devices, OCT technology is expected to become an important innovative means in the field of blood glucose monitoring, providing more convenient and accurate blood glucose management tools for diabetic patients, and promoting the development of precision medicine at the same time. Summary of the Invention

[0007] Object of the Invention: In order to significantly improve the quality of life of diabetic patients and solve the deficiencies of traditional blood glucose monitoring methods, the present invention discloses a novel blood glucose monitoring method based on optical coherence tomography technology. Through Doppler imaging technology, OCT also reduces the influence of motion artifacts, is suitable for blood glucose monitoring during exercise, and solves the hysteresis effect problem in traditional non-invasive monitoring, improving the accuracy and real-time performance of measurement.

[0008] Technical Solution: The present invention discloses a novel blood glucose monitoring method based on optical coherence tomography imaging technology, including the following steps:

[0009] Step 1) Collect OCT images of the glucose concentration in the interstitial fluid of the human body as sample images, and obtain the blood glucose concentration in the traditional way, corresponding the OCT signal to the blood glucose concentration to form a modeling set;

[0010] Step 2) Use the GPU to load the sample images in Step 1), quickly reconstruct three-dimensional images, and restore and accurately locate blood vessels;

[0011] Step 3) Remove the spectral offset term from the three-dimensional images established in Step 2);

[0012] Step 4) After removing the spectral offset term in Step 3), extract the near-infrared spectrum of blood without spectral offset through windowed Fourier transform, and establish a prediction model of near-infrared spectrum data and blood glucose concentration through chemometric algorithms. Use the modeling set to train the prediction model of near-infrared spectrum data and blood glucose concentration, and use the trained prediction model of near-infrared spectrum data and blood glucose concentration to predict the blood glucose concentration of unknown sample spectra to obtain the accurate blood glucose concentration of the human body.

[0013] Further, the reconstruction of the three-dimensional image in Step 2) is specifically as follows:

[0014] Step 2.1) Each interference spectrum collected by the CCD is a longitudinal scan, and the interference signal is expressed by the following formula:

[0015]

[0016] where k is the wave number, t is the time, S(k) is the spectral density function of the light source, R lis the reflectivity of the sample of the l-th layer, R r is the reflectivity of the reference arm plane mirror, z l is the optical path difference between the sample of the l-th layer and the reference plane mirror, v l is the velocity of the sample of the l-th layer in the direction parallel to the incident light velocity;

[0017] The phase signal is expressed by the following formula:

[0018] Step 2.2) The blood movement velocity v is decomposed into a velocity v p perpendicular to the incident light velocity direction and a velocity v l parallel to the incident light velocity direction. Due to the Doppler effect, the phase difference of the blood flow velocity parallel to the incident light velocity direction within the time interval Δt is:

[0019] Since the time interval Δt is an integer multiple of the exposure time and is a known quantity, the component of the blood flow velocity in the direction parallel to the incident light velocity direction can be obtained from the above formula as:

[0020]

[0021] Considering that in the ideal case, the tissues in the skin other than blood are stationary. When performing multiple A-scans at the same position, there is a phase difference in the blood signal between different A-scan signals, while there is no phase difference in other tissues in the skin, thus distinguishing blood vessels from other tissues, and realizing three-dimensional imaging of blood vessels through the scanning of the X-Y galvanometer;

[0022] 2.3) Use the GPU for parallel computing. When the GPU processes OCT data, the system first copies the data from the buffer of the host to the buffer of the graphics card, and then calls the kernel function in CUDA to parallel process the data in the buffer using the stream processors. The GPU processes N groups of interference spectral signals simultaneously, including resampling, DC removal, and removal of complex conjugate images, dispersion compensation, and phase calculation and blood flow relative velocity calculation algorithms. The number of interference spectral signals processed simultaneously depends on the number of stream processors of the GPU.

[0023] Furthermore, in the step 2.3), when the GPU processes N groups of interference spectral signals simultaneously, to remove the complex conjugate image, a method of deviating the light spot from the axis of the scanning galvanometer is used to introduce a modulation frequency in different B-scan signals, and the complex interference spectrum is reconstructed by performing a Hilbert transform on the B-scan signals, thereby realizing the removal of the complex conjugate image.

[0024] Further, the specific method for removing the complex conjugate image: Select a collimating lens according to the size of the galvanometer mirror surface, ensure that the spot size is much smaller than the offset distance of the spot, collimate the outgoing light of the light onto the scanning galvanometer, and adjust the scanning galvanometer so that the incident spot deviates from the axis of rotation of the scanning galvanometer; perform a Hilbert transform on the interference spectrum along the B-scan direction to reconstruct the complex interference spectrum, and then perform a Fourier transform on the reconstructed complex interference spectrum along the CCD pixel number direction to obtain the reconstructed image of the tissue.

[0025] Further, specifically for removing the spectral offset term from the established three-dimensional image in step 3) is as follows:

[0026] Step 3.1) The prediction error introduced due to spectral offset is:

[0027] Δc = c′ - c = b(λ 1 )[I(λ 1 + δλ) - I(λ 1 )] +... + b(λ n )[I(λ n + δλ) - I(λ n )]

[0028] where λ is the wavelength, λ 1 ...λ n is a specific value in a wavelength sequence, n identifies the wavelength sequence, b(λ n ) is a coefficient related to the wavelength, [I(λ 1 )...I(λ n )] is the near-infrared spectrum, c is the blood glucose concentration, c′ is the predicted value of the blood glucose concentration. Considering that the spectral offset amount δλ is much smaller than λ, performing a Taylor expansion on the near-infrared spectrum gives:

[0029]

[0030] where R n (λ) is the high-order term and can be ignored. Substituting the above formula into the following formula to obtain the relationship between the prediction error and the spectral offset amount:

[0031]

[0032] Step 3.2) Using the properties of the Fourier transform, convert the translation term in the time domain into a phase term in the frequency domain, and perform PLS modeling on the near-infrared spectrum of blood in the frequency domain to effectively remove the influence of the spectral offset amount; assuming that the Fourier transform of the original spectrum I(λ) is:

[0033] G(ω) = FFT(I(k))

[0034] where k is the wave number, and the Fourier transform when there is a spectral offset and the offset amount is δλ is:

[0035] G′(ω) = FFT(I(λ + δλ)) = G(ω)exp(-jωδλ)

[0036] Where ω is the angular frequency, j is the imaginary unit, and λ is the wavelength. Combining the above two equations, we can obtain

[0037] |G(ω)| = |G′(ω)|

[0038] Perform PLS modeling on the modulus of the Fourier transform in the above equation and the concentration, completely avoiding the influence of the spectral offset on the model and improving the stability of the model.

[0039] Furthermore, the specific content of step 4) is as follows:

[0040] Step 4.1) Select a blood vessel within the lateral position range of 200×200 μm in the three-dimensional image of the blood vessel, and obtain the central position z 0 and diameter d at different lateral positions of the blood vessel, and perform windowed Fourier transform on the interference signal at this position. The window center is k n , and the window width is Δk, as shown in the following equation:

[0041] I(k,z) = FFT -1 (I(k)g(k n ,Δk))

[0042] Where I(k,z) is the signal at wave number k and a certain spatial coordinate z, I(k) is the original signal in the frequency space. There is an inverse relationship between the wave number k and the wavelength λ, that is, k = 2π / λ. Adjust the window width Δk so that the depth resolution Δz ≤ d. Δz represents the resolution ability in the depth direction and describes the minimum depth difference that the system can distinguish. Select z = z 0 , so as to obtain the near-infrared spectrum I(k,z 0 ) of the blood;

[0043] Step 4.2) Preprocess the obtained near-infrared spectrum, and use partial least squares (PLS) to establish a model for predicting blood glucose concentration. Use Doppler OCT to image the same position to obtain the OCT signal, and extract the near-infrared spectrum of the blood. Correlate the OCT signal with the blood glucose concentration in the modeling set to obtain the near-infrared spectrum matrix X and the concentration matrix C of the modeling set. Obtain the Beta coefficient of the PLS model according to the spectral matrix X and the concentration matrix C of the modeling set:

[0044] C ≈ XBeta

[0045] Where Beta = [b 0 b(λ 1 )... b(λ n )], b 0is a constant term, and b(λ n ) is a coefficient related to the wavelength;

[0046] For an unknown concentration sample, if its near-infrared spectrum is known as [I(λ 1 )...I(λ n )], then the blood glucose concentration is:

[0047] c = [1 I(λ 1 )... I(λ n )][b 0 b(λ 1 )... b(λ n )] T

[0048] In the formula, [b 0 b(λ 1 )... b(λ n )] T is the transpose of the Beta matrix;

[0049] When there is an offset in the measured spectrum, assuming the offset is δλ, then the predicted value of the PLS model is:

[0050] c′ = [1 I(λ 1 + δλ)... I(λ n + δλ)][b 0 b(λ 1 )... b(λ n )] T

[0051] The coefficient of determination of the model is:

[0052]

[0053] Among them, c p is the fitting value of the model, c a is the average value of the sample concentration, c m is the measured value of the sample concentration. Calculate the coefficient of determination of the model under different numbers of principal components, and compare the values of the coefficient of determination to determine the optimal number of principal components.

[0054] Beneficial effects:

[0055] 1. The present invention combines spectroscopic technology with Doppler optical coherence tomography: First, OCT is used to collect images of glucose concentration in interstitial fluid of the human body, and a GPU is utilized to rapidly reconstruct three-dimensional images for vessel restoration and precise positioning. Then, in combination with the near-infrared spectrum after removing spectral shift, its absorption characteristics are analyzed through a chemometric model to address the motion artifacts and the influence of free blood glucose existing in current OCT technology for blood glucose monitoring. In this way, the chemical absorption characteristics and physical scattering characteristics complement each other, achieving high precision and high robustness in blood glucose concentration monitoring. By obtaining blood glucose data through optical means, it is completely unnecessary to puncture the skin, avoiding the pain and infection risks brought by traditional blood sampling methods and reducing the discomfort of patients. For diabetic patients who need to frequently measure blood glucose, this non-invasive method provides a more comfortable experience and improves their quality of life.

[0056] 2. The present invention combines Doppler imaging technology to provide relatively stable blood glucose data even when the patient is active, which means that even during the patient's daily activities or exercise, the system can still maintain a high monitoring accuracy. This feature solves the problem that traditional non-invasive monitoring is vulnerable to interference during exercise, enabling patients to obtain reliable blood glucose data in various situations.

[0057] 3. Traditional non-invasive monitoring methods such as near-infrared spectroscopy often result in measurement delays due to the lag in glucose concentration changes in interstitial fluid. The blood glucose monitoring based on OCT in the present invention combines sophisticated spectral modeling technology, effectively reducing this lag effect, making the measurement faster and more accurate, enabling patients to promptly understand their true blood glucose levels and avoiding misjudgment or risks caused by delays.

[0058] 4. The present invention can be integrated into portable or wearable devices such as bracelets and wristwatches, enabling patients to perform non-invasive blood glucose monitoring anytime and anywhere in their daily lives. This convenience supports long-term and continuous blood glucose management, especially suitable for patients who need to frequently detect blood glucose, reducing their dependence on traditional blood glucose meters. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flowchart of the method of the present invention;

[0060] Figure 2 is a schematic diagram of Doppler OCT;

[0061] Figure 3 is a two-dimensional structure diagram of esophageal tissue. DETAILED DESCRIPTION OF THE INVENTION

[0062] This application relates to multiple scientific and technological fields such as medicine, OCT, and image processing, with a wide coverage. Therefore, unless otherwise specifically defined, the meanings of all technical and scientific terms used herein are the same as those commonly understood by those skilled in the art of this application field; the terms described herein are only for the purpose of describing specific embodiments and do not imply any limitation to this application; the terms "including" and "having" used in this application, as well as any variants thereof, are intended to cover non-exclusive meanings.

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0064] As Figure 1 shown, a specific embodiment of the present invention discloses a novel blood glucose monitoring method based on optical coherence tomography technology, including the following steps:

[0065] 1) Diabetes is divided into type I and type II diabetes. Type I diabetes mostly occurs in young people, and type II diabetes mostly occurs in people over 40 years old. Generally speaking, the incidence of type II diabetes is much higher than that of type I diabetes. Considering that the high-incidence age of diabetes is between 40 and 59 years old, the age distribution ratios of all the participating volunteers in the modeling are 54% and 46% respectively (the male-female ratio of diabetic patients is 1:0.84). After obtaining the blood glucose concentration of the volunteers, Doppler OCT is used to image the same position to obtain OCT signals, and the near-infrared spectrum of the blood is extracted. The OCT signals of the same volunteer are corresponded to the blood glucose concentration to obtain the near-infrared spectrum matrix X and the concentration matrix Y.

[0066] 2) Load the sample images in step 1) through the GPU to efficiently realize the reconstruction of three-dimensional images.

[0067] 2.1) Each interference spectrum collected by the CCD corresponds to a longitudinal scan, and its interference signal can be expressed by the following formula

[0068]

[0069] where k represents the wave number, t represents the time, S(k) represents the spectral density function of the light source, R l represents the reflectivity of the l-th layer of the sample, R r represents the reflectivity of the reference arm plane mirror, z l represents the optical path difference between the l-th layer of the sample and the reference plane mirror, v l represents the velocity of the l-th layer of the sample parallel to the incident light velocity direction.

[0070] Its phase signal is expressed by the following formula

[0071]

[0072] 2.2) Since the blood is in a relative motion state with respect to the surrounding skin tissue, its motion speed v can be decomposed into a speed v perpendicular to the direction of the incident light speed p and a speed v parallel to the direction of the incident light speed l , affected by the Doppler effect, the blood flow velocity in the direction parallel to the incident light direction will cause a phase difference within the time interval Δt, and the specific expression is:

[0073] Since the time interval Δt is an integer multiple of the exposure time and is a known quantity, according to the above formula, the component of the blood flow velocity in the direction parallel to the incident light direction can be obtained as:

[0074]

[0075] In an ideal situation, other tissues in the skin except blood are stationary. When performing multiple A-scans on the same position, a phase difference will be generated between blood signals in different A-scans, while the signals of other tissues remain without phase difference. Through this characteristic, blood vessels can be distinguished from other tissues, and three-dimensional imaging of blood vessels can be achieved by combining the scanning of the X-Y galvanometer.

[0076] 2.3) In three-dimensional imaging, the amount of data is extremely large, possibly reaching millions or even tens of millions of interference spectra. Using the traditional CPU serial processing method to reconstruct the OCT three-dimensional image takes a long time. The present invention significantly improves the processing efficiency by introducing GPU parallel computing. When the GPU processes OCT data, the system first transfers the data from the host buffer to the video card buffer, and then uses the kernel function in CUDA to perform parallel processing on the buffered data through the stream processors of the GPU. Different from the serial processing of the CPU, the GPU can simultaneously process multiple interference spectrum signals, covering algorithms such as resampling, DC removal and conjugate image removal, dispersion compensation, phase calculation, and blood flow relative velocity calculation. This method significantly shortens the data processing time, and the number of interference spectrum signals simultaneously processed by the GPU is mainly determined by the number of stream processors.

[0077] 3) Remove the spectral offset term from the three-dimensional image established in step 2).

[0078] 3.1) The prediction value error introduced due to spectral offset is

[0079] Δc = c′ - c = b(λ 1 )[I(λ 1 + δλ) - I(λ 1 )] +... + b(λ n )[I(λ n + δλ) - I(λ n )]

[0080] where λ is the wavelength, λ1 ...λ n is a specific value in the wavelength sequence, n identifies the wavelength sequence, b(λ n ) is a coefficient related to the wavelength, [I(λ 1 )...I(λ n )] is the near-infrared spectrum, c is the blood glucose concentration, c′ is the predicted value of the blood glucose concentration. Considering that δλ is much smaller than λ, performing a Taylor expansion on the near-infrared spectrum gives:

[0081]

[0082] where R n (λ) is the high-order term and can be ignored. Substituting the above equation into the following equation, the relationship between the prediction error and the spectral offset can be obtained

[0083]

[0084] It can be seen that the spectral offset has an important impact on the model stability. Therefore, solving the influence of spectral offset in blood near-infrared spectra on the PLS model is crucial for improving the accuracy of blood glucose monitoring.

[0085] 3.2) Through the properties of the Fourier transform, the translation term in the time domain can be transformed into a phase term in the frequency domain. Therefore, performing PLS modeling on the near-infrared spectrum of blood in the frequency domain helps to effectively eliminate the influence of spectral offset. Assume that the Fourier transform of the original spectrum I(λ) is:

[0086] G(ω) = FFT(I(k))

[0087] where k is the wave number. Then, when there is a spectral offset and the offset is δλ, the Fourier transform is

[0088] G′(ω) = FFT(I(λ + δλ)) = G(ω)exp(-jωδλ)

[0089] Combining the above two equations, we get

[0090] |G(ω)| = |G′(ω)|

[0091] Therefore, it can be seen that the spectral offset does not affect the amplitude after Fourier transform. By performing PLS modeling on the amplitude after Fourier transform and the concentration data, the interference of the spectral offset on the model can be effectively avoided, thereby improving the stability of the model.

[0092] 3.3) Using the established model above, predict the blood glucose concentration in the blood spectra measured every 15 minutes within 30 minutes after oral glucose administration respectively; calculate the scattering coefficient within the skin depth range of 200 - 400 μm, and measure the glucose concentration in the interstitial fluid using the OCTSS method, so as to reveal the dynamic change law of blood glucose concentration and interstitial fluid glucose concentration. Furthermore, study the accuracy of measuring blood glucose concentration using the two methods within different concentration ranges.

[0093] 4) After removing the spectral offset term in step 3), using the previously obtained three-dimensional blood vessel image, select a blood vessel within the lateral area of 200×200 μm, and calculate the central position z of the blood vessel at different lateral positions 0 and diameter d (in the depth direction). Then, perform a windowed Fourier transform on the interference signal at this position, with the window center at k n , and the window width at Δk. The specific expression is as follows: I(k,z) = FFT -1 (I(k)g(k n ,Δk))

[0094] where I(k,z) is the signal at wave number k and a certain spatial coordinate z, I(k) is the original signal in the frequency space. There is an inverse relationship between the wave number k and the wavelength λ, that is, k = 2π / λ. Adjust the window width Δk such that the depth resolution Δz ≤ d. Δz represents the resolution ability in the depth direction and describes the minimum depth difference that the system can distinguish. Select z = z 0 , so as to obtain the near-infrared spectrum I(k,z 0 ) of the blood.

[0095] 4.1) Match the OCT signal of the same volunteer with the blood glucose concentration data, aiming to obtain the near-infrared spectrum matrix X and the concentration matrix C. Through the Doppler OCT technology, collect the OCT signal at the same position every 15 minutes. These generated data will be used to study the dynamic change model of blood glucose and interstitial fluid glucose concentration.

[0096] The present invention combines the Doppler OCT technology, locates the blood vessels through three-dimensional reconstruction, and then extracts the near-infrared spectrum of the blood through windowed Fourier transform, realizing the direct monitoring of blood glucose concentration and avoiding the hysteresis effect that may occur in indirect measurement. The windowed Fourier transform can perform time-frequency analysis on the interference spectrum signal, and the window width affects the balance between the spectrum and the depth resolution. Since there is a mutual restriction relationship between the spectral resolution and the depth resolution, according to the standard deviation and variance of the predicted values in the cross-validation of the modeling set, by selecting an appropriate window width, the blood infrared spectrum extraction algorithm can be optimized to reduce the influence of the window width on the prediction accuracy.

[0097] 4.2) After preprocessing the obtained spectral data, the PLS method is used to establish a blood glucose concentration prediction model. By using the spectral matrix X and the concentration matrix C in the modeling set, the Beta coefficients of the PLS model can be calculated

[0098] C ≈ XBeta

[0099] where Beta = [b 0 b(λ 1 )... b(λ n )], b 0 is the constant term, and b(λ n ) is the coefficient related to the wavelength.

[0100] For a sample with unknown concentration, if its near-infrared spectrum is known as [I(λ 1 )... I(λ n )], then the blood glucose concentration is

[0101] c = [1 I(λ 1 )... I(λ n )][b 0 b(λ 1 )... b(λ n )] T

[0102] In the formula, [b 0 b(λ 1 )... b(λ n )] T is the transpose of the Beta matrix.

[0103] When there is an offset in the measured spectrum, assuming the offset is δλ, then the predicted value of the PLS model is

[0104] c′ = [1 I(λ 1 + δλ)... I(λ n + δλ)][b 0 b(λ 1 )... b(λ n )] T

[0105] The coefficient of determination of the model is:

[0106]

[0107] where c p is the fitted value of the model, c a is the average value of the sample concentration, and c m is the measured value of the sample concentration. Calculate the coefficient of determination of the model under different numbers of principal components, and compare the values of the coefficient of determination to determine the optimal number of principal components.

[0108] 5) Through the above steps, the three-dimensional model established by using OCT technology realizes the accurate measurement of human blood glucose concentration.

[0109] The central wavelength of the light source adopted by the OCT system of the present invention is 1200 nm, and the bandwidth is 280 nm; the number of pixels of the CCD is 2048 pixels, the spectral resolution of the system is ~0.068 nm (in air), so the longitudinal resolution of the system can be calculated to be ~2.3 μm, and the imaging depth of the system is ~9 mm (in air). And the blood vessels in the skin are in the dermis layer. The blood vessels in the fingertip skin are less than 1 mm away from the skin surface, and the average diameter of the thinnest capillaries is 7 - 9 μm. According to previous studies, for whole blood samples, the correlation coefficient between the spectral data in the wavenumber range of 7540 cm -1 -9425 cm -1 (1061 nm - 1330 nm) and blood glucose concentration can reach 0.9999, and the error is less than 0.13 mmol / L.

[0110] The present invention relies on the spectrometer of Andor Company, and has built a proof-of-principle endoscopic OCT system by using a laser with a central wavelength of 1310 nm and a bandwidth of 85 nm, and has imaged the esophagus to obtain a two-dimensional structure diagram of the esophageal tissue. See Figure 3 . On this basis, it is planned to replace the laser (S5FC1021S) in the original system with an ultra-wideband continuous laser (szu-sc001), and add corresponding devices such as filters and fiber optic couplers, so as to possibly realize an OCT system applied to blood glucose monitoring.

[0111] Obviously, the embodiments described above only constitute part of the implementation schemes of this application, rather than all implementation schemes. The preferred embodiments of this application are shown in the drawings, but are not limited thereto. The patent scope of this application is not limited to the content shown in the drawings. This application can be implemented in many different forms. The purpose of providing these embodiments is to more comprehensively clarify the content of this application. Although the foregoing embodiments have been described in detail, for those skilled in the art, the specific implementation manners can still be modified, or some technical features can be equivalently replaced. Any equivalent structure using the content of the specification and drawings of this application and directly or indirectly applied in other related technical fields should be included in the patent protection scope of this application.

Claims

1. A novel blood glucose monitoring method based on optical coherence tomography technology, characterized in that: The following steps are involved: Step 1) collecting an OCT image of glucose concentration in human intercellular fluid as a sample image, and obtaining blood glucose concentration using a traditional method, and matching the OCT signal with the blood glucose concentration to form a modeling set; Step 2) Using GPU to load the sample image of step 1), quickly reconstruct the three-dimensional image, restore and accurately locate the blood vessels; Step 3) removing the spectral offset term from the three-dimensional image created in step 2); Step 4) After removing the spectral offset term in step 3), the near-infrared spectrum of blood without spectral offset is extracted by windowed Fourier transform, and a prediction model of near-infrared spectral data and blood glucose concentration is established by a chemometric algorithm. The prediction model of near-infrared spectral data and blood glucose concentration is trained using the modeling set, and the trained near-infrared spectral data and blood glucose concentration prediction model is used to predict the blood glucose concentration of the unknown sample spectrum to obtain the accurate blood glucose concentration of the human body.

2. According to claim 1, a novel blood glucose monitoring method based on optical coherence tomography technology is characterized in that: The three-dimensional image reconstruction in step 2) is specifically as follows: Step 2.1) Each interference spectrum collected by the CCD is a longitudinal scan, and the interference signal is expressed as follows: Where k is the wave number, t is the time, S(k) is the spectral density function of the light source, R l is the reflectivity of the sample in the first layer, R r is the reflectivity of the reference arm plane mirror, z l is the optical path difference between the sample layer l and the reference plane mirror, v l is the velocity of the sample in the lth layer parallel to the direction of the incident light velocity; The phase signal is expressed as follows: Step 2.2) The blood velocity v is decomposed into the velocity v perpendicular to the incident light velocity p and the velocity v parallel to the direction of the incident light velocity l , due to the Doppler effect, the phase difference of the blood flow velocity parallel to the direction of the incident light velocity within the time interval Δt is: Since the time interval Δt is an integer multiple of the exposure time and is a known quantity, the component of the blood flow velocity in the direction parallel to the incident light velocity can be obtained from the above formula: Considering that the tissues in the skin except the blood are stationary in ideal conditions, when multiple A-scans are performed at the same location, there is a phase difference between the blood signal and other tissues in the skin between different A-scan signals, thereby distinguishing blood vessels from other tissues and achieving three-dimensional imaging of blood vessels through scanning with an XY galvanometer. Step 2.3) GPU is used for parallel computing. When GPU processes OCT data, the system first copies the data from the host's cache to the graphics card's cache, and then uses the stream processor to process the data in the cache in parallel by calling the kernel function in CUDA. The GPU processes N groups of interference spectral signals simultaneously, including resampling, DC removal and removal of complex conjugate images, dispersion compensation, phase calculation and blood flow relative velocity calculation algorithms, and the number of interference spectral signals processed simultaneously depends on the number of stream processors of the GPU.

3. The novel blood glucose monitoring method based on optical coherence tomography technology according to claim 2, characterized in that: In the step 2.3), the GPU processes N groups of interference spectrum signals simultaneously, and the complex conjugate image is removed by introducing modulation frequency into different B-scan signals by deviating the light spot from the scanning galvanometer axis. The complex interference spectrum is reconstructed by performing Hilbert transform on the B-scan signal, thereby removing the complex conjugate image.

4. The novel blood glucose monitoring method based on optical coherence tomography technology according to claim 3 is characterized in that: Specific method for removing the complex conjugate image: select a collimating lens according to the size of the galvanometer mirror to ensure that the spot size is much smaller than the offset distance of the spot, collimate the outgoing light onto the scanning galvanometer, and adjust the scanning galvanometer so that the incident light spot deviates from the scanning galvanometer axis; perform Hilbert transform on the interference spectrum along the B scan direction to reconstruct the complex interference spectrum, and then perform Fourier transform on the reconstructed complex interference spectrum along the direction of the CCD pixel number to obtain a reconstructed image of the tissue.

5. The novel blood glucose monitoring method based on optical coherence tomography technology according to claim 1, characterized in that: The specific method of removing the spectral offset item from the established three-dimensional image in step 3) is as follows: Step 3.1) The prediction value error introduced by the spectral shift is: Δc=c′-c=b(λ1)[I(λ1+δλ)-I(λ1)]+...+b(λ n )[I(λ n +sλ)-I(λ n )] Where λ is the wavelength, λ1...λ n is a specific value in the wavelength sequence, n identifies the wavelength sequence, b(λ n ) is a coefficient related to wavelength, [I(λ1)...I(λ n )] is the near-infrared spectrum, c is the blood glucose concentration, c′ is the predicted value of the blood glucose concentration. Considering that the spectral offset δλ is much smaller than λ, Taylor expansion of the near-infrared spectrum can be obtained: Among them, R n (λ) is a high-order term and can be ignored. Substituting the above equation into the following equation, we can obtain the relationship between the prediction error and the spectral offset: Step 3.2) Using the properties of Fourier transform, the translation term in the time domain is converted into the phase term in the frequency domain, and PLS modeling is performed on the near-infrared spectrum of blood in the frequency domain to effectively remove the influence of spectral offset; assuming that the original spectrum is I(λ), the Fourier transform is: G(ω)=FFT(I(k)) Where k is the wave number. When there is a spectral shift and the shift is δλ, the Fourier transform is: G′(ω)=FFT(I(λ+δλ))=G(ω)exp(-jωδλ) Where ω is the angular frequency, j is the imaginary unit, and λ is the wavelength. Combining the above two equations, we can get |G(ω)|=|G′(ω)| The modulus and concentration of the Fourier transform in the above formula are used for PLS modeling to completely avoid the influence of the spectral offset on the model and improve the stability of the model.

6. The novel blood glucose monitoring method based on optical coherence tomography technology according to claim 1, characterized in that: The step 4) is specifically as follows: Step 4.1) Select a blood vessel within the lateral position range of 200×200 μm in the three-dimensional image of the blood vessel, calculate the center position z0 and diameter d of the blood vessel at different lateral positions, and perform a windowed Fourier transform on the interference signal at this position, with the center of the window being k. n , the window width is Δk, as shown in the following formula: I(k,z)=FFT -1 (I(k)g(k n ,Δk)) Where, I(k,z) is the signal at wave number k and a certain spatial coordinate z, I(k) is the original signal in the frequency space, there is an inverse relationship between wave number k and wavelength λ, that is, k = 2π / λ, adjust the window width Δk so that the depth resolution Δz ≤ d, Δz represents the resolution in the depth direction, and describes the minimum depth difference that the system can distinguish, select z = z0, and obtain the near-infrared spectrum I(k,z0) of blood; Step 4.2) Preprocess the obtained near-infrared spectrum, and use the partial least squares method PLS to establish a model to predict the blood glucose concentration. Use Doppler OCT to image the same position to obtain the OCT signal, and extract the near-infrared spectrum of the blood. The OCT signal is matched with the blood glucose concentration of the modeling set to obtain the near-infrared spectrum matrix X and concentration matrix C of the modeling set. According to the spectrum matrix X and concentration matrix C of the modeling set, the Beta coefficient of the PLS model is obtained: C≈XBeta Among them, Beta=[b0 b(λ1)...b(λ n )], b0 is a constant term, b(λ n ) is a coefficient related to wavelength; For a sample of unknown concentration, if its near-infrared spectrum is known to be [I(λ1)...I(λ n )], then the blood glucose concentration is: c=[1I(λ1)...I(λ n )][b0 b(λ1)...b(λ n )] T Where [b0 b(λ1)...b(λ n )] T is the transpose of the Beta matrix; When the measured spectrum is offset, assuming the offset is δλ, the predicted value of the PLS model is: c′=[1I(λ1+δλ)...I(λ n +sλ)][b0 b(λ1)...b(λ n )] T The coefficient of determination of the model is: Among them, c p is the fitted value of the model, c a is the average value of sample concentration, c m For the measured value of sample concentration, calculate the determination coefficient of the model under different numbers of principal components, and compare the values ​​of the determination coefficient to determine the optimal number of principal components.

Images