A method for optical non-invasive blood glucose concentration detection based on time-spectrum-space multi-parameter fusion

Abstract

The application discloses a method for optical non-invasive blood glucose concentration detection based on time-spectrum-space multi-parameter fusion, and belongs to the field of physiological signal detection. The method comprises the following steps: synchronously collecting video of skin tissue of a subject at a position with pulse beats and corresponding diffuse reflection spectrum, and obtaining a concentration value of blood glucose in a manner of invasive finger blood sampling as a true value; extracting an IPPG signal of a G channel from the video of the skin tissue as a time domain signal through image data processing technology and spectrum analysis technology, extracting the diffuse reflection spectrum of the skin tissue as a spectrum domain signal, and extracting a spatial dimension feature of a single frame image in the video of the skin tissue as a space domain signal; taking time-spectrum-space multi-parameter fusion and the blood glucose value in the invasive manner as input signals of a deep model; and building a non-specific optical non-invasive blood glucose concentration detection model through a deep learning algorithm, so as to achieve the purpose of predicting blood glucose concentration. The application realizes establishment of a non-specific non-invasive blood glucose general prediction model, improves the prediction accuracy of blood glucose concentration through multi-parameter feature fusion, and conveniently, quickly and accurately obtains the blood glucose concentration value of the subject.

Description

Technical Field

[0001] This invention relates to the field of physiological signal detection, specifically to a method for non-invasive optical blood glucose concentration detection based on time-spectrum-spatial multi-parameter fusion. Background Technology

[0002] Regular and frequent blood glucose concentration monitoring, used as a basis for drug treatment, is an effective means of preventing and treating diabetes and its complications. Optical non-invasive detection technology, with its characteristics of speed, non-invasiveness, and multidimensional information, has become a research hotspot in non-invasive blood glucose measurement as a way to obtain optical parameter information from living tissues. However, blood glucose levels in blood and body fluids are very low, and the spectral responses of various components in human tissues overlap with the characteristic peaks of glucose. Therefore, the changes in optical information caused by changes in blood glucose concentration are relatively weak. Furthermore, variations in physiological background information during measurement, such as fluctuations in metabolism, emotion, blood pressure, body temperature, sweating, and individual differences in health status, cannot yet be clearly explained or precisely quantified to explain their interference with blood glucose concentration measurement. These factors result in low signal-to-noise ratios in current optical non-invasive methods for blood glucose concentration detection, and improving the accuracy of blood glucose concentration detection remains a key research topic that optical non-invasive blood glucose measurement systems have yet to overcome. Summary of the Invention

[0003] The purpose of this invention is to address the problems of low detection accuracy and low signal-to-noise ratio in current optical non-invasive methods for blood glucose detection. This invention proposes a non-invasive optical blood glucose concentration detection method based on time-spectrum-spatial multi-parameter fusion. This method uses physiological parameter information from three dimensions—time domain, spectral domain, and spatial domain—as input signals for the blood glucose model, and combines this with deep learning algorithms to build a general non-invasive blood glucose model. This invention expands the use of physiological parameter information from different dimensions as input signals for the blood glucose model, with their information complementing each other. Combined with deep learning algorithms, a non-adaptive non-invasive blood glucose concentration detection model is established, effectively improving the accuracy of blood glucose detection. This invention proposes a multi-parameter fusion method to avoid low-precision blood glucose concentration detection, effectively improving the limitations of current optical non-invasive detection methods. The objective of this invention is achieved through the following technical solutions.

[0004] A non-invasive optical method for detecting blood glucose concentration based on time-spectrum-spatial multi-parameter fusion includes the following steps:

[0005] Step 1: Simultaneously acquire video of the skin tissue of the subject with a pulse and the corresponding diffuse reflectance spectrum, and obtain the blood glucose concentration value as the true value by invasive finger blood sampling.

[0006] Step 2: Perform image processing on the video of the skin tissue acquired in Step 1 to obtain the IPPG signal;

[0007] 1) For the first frame of the acquired subject skin tissue video, a rectangular matrix is ​​used to select any pixel region [a1, b1] containing the skin tissue image as the region of interest for IPPG signal extraction; where a1 is the height and b1 is the width, and a1 and b1 are smaller than the image size acquired by the CCD camera.

[0008] 2) Calculate the average pixel value of the region of interest to obtain the original IPPG signal;

[0009] 3) The original IPPG signal is subjected to RGB color channel separation and a preprocessing algorithm to remove interference from non-physiological parameters, and then the IPPG signal of the G channel is extracted.

[0010] 4) Use the preprocessed IPPG signal of the G channel as the time-domain signal of blood glucose;

[0011] Step 3: Process the diffuse reflectance spectrum of the skin tissue collected in Step 1 to obtain the diffuse reflectance spectrum of human skin tissue;

[0012] 1) Collect the diffuse reflectance spectrum of the subject's skin tissue, and calculate the average diffuse reflectance spectrum of the human tissue by averaging multiple diffuse reflectance spectrum data.

[0013] 2) Use the average diffuse reflectance spectrum as the spectral domain signal of blood glucose;

[0014] Step 4: Perform image processing on the skin tissue video acquired in Step 1 to obtain a single-frame image. Select any frame from the single-frame image to obtain the image's spatial dimension features.

[0015] 1) For any frame of a single-frame image, select the region containing the skin tissue image, and use a rectangular matrix to select any pixel position [a2, b2] as the region of interest for image spatial dimension feature extraction; where a2 is the height and b2 is the width, and a2 and b2 are smaller than the image size captured by the CCD camera (a2 and b2 can be the same as a1 and b1).

[0016] 2) Extract spatial dimension features of the region of interest image using a noise reduction autoencoder algorithm;

[0017] 3) Process the spatial dimension features, remove redundant features, and obtain the spatial dimension features related to blood glucose concentration as the spatial signal of blood glucose.

[0018] Step 5: Repeat steps 1, 2, 3 and 4 to collect time-domain, spectral domain, and spatial domain signals, as well as blood glucose values, from different subjects. The IPPG signal of the G channel of multiple subjects is used as the time-domain signal of blood glucose, the diffuse reflectance spectrum is used as the spectral domain signal of blood glucose, the spatial dimension features are used as the spatial domain signal of blood glucose, and the invasive blood glucose value is used as the true value.

[0019] Step 6: Use the time-spectrum-space signal and invasive blood glucose reference values ​​obtained in Step 5 as the dataset. Divide the dataset into a training set and a test set. Use a deep learning algorithm to train a general blood glucose concentration prediction model on the training set. Then, input the test set into the general blood glucose concentration prediction model to obtain the predicted blood glucose concentration value for each subject in the test set.

[0020] Beneficial effects

[0021] 1. The present invention provides a non-invasive optical blood glucose concentration detection method based on time-spectrum-spatial multi-parameter fusion, which is a non-invasive optical blood glucose concentration measurement method based on multi-parameter fusion, and can effectively improve the problem of low signal-to-noise ratio in blood glucose concentration detection accuracy.

[0022] 2. This invention is applicable to the acquisition of time-spectral-spatial multi-parameter signals from different parts of the human body and the fusion of multi-parameter features.

[0023] 3. This invention utilizes a non-contact optical detection method to establish a universal blood glucose concentration detection model for different subjects in a non-invasive, accurate, and convenient manner. The results can be used as an auxiliary basis for doctors to make professional judgments. Attached Figure Description

[0024] Figure 1 A schematic diagram of the apparatus for a non-invasive optical blood glucose concentration detection method based on time-spectrum-space multi-parameter fusion provided in the embodiments;

[0025] Figure 2 The overall flowchart of the optical non-invasive blood glucose concentration detection method based on time-spectrum-space multi-parameter fusion provided for the embodiments is shown below. Detailed Implementation

[0026] To make the objectives, advantages, and features of this invention clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a method for non-invasive optical blood glucose concentration detection based on time-spectrum-spatial multi-parameter fusion. It should be noted that the accompanying drawings are all in a very simplified form and use non-precise proportions, used only to facilitate and clarify the illustration of the embodiments of this invention. The structures shown in the drawings are only a part of the actual structures. The human skin tissue described in this invention is not limited to the fingers; it is also applicable to other human tissues where pulse waves can be extracted and diffuse reflectance spectra of the skin tissue can be obtained. The deep learning algorithm used in this invention is not limited to a specific regression model algorithm.

[0027] A method for non-invasive optical blood glucose concentration detection based on time-spectrum-spatial multi-parameter fusion is illustrated in the following schematic diagram. Figure 1 As shown, the overall flowchart is as follows: Figure 2 As shown.

[0028] Step 1: Use a visible light source to illuminate the skin tissue of different parts of the human body, and at the same time turn on the spectrometer and CCD camera to collect video containing pulse information and the corresponding diffuse reflectance spectrum of the illuminated area.

[0029] 1) Start the light source, CCD camera, and spectrometer:

[0030] The subject sat still in a chair, and a visible light source was used to uniformly illuminate different parts of the skin tissue. A CCD camera and a spectrometer were simultaneously activated. A polarizer was placed in front of the visible light source, and an analyzer was placed in front of the CCD camera. The polarizer and analyzer eliminated the influence of specular reflection. The different parts of the skin tissue refer to areas such as the face, arms, and fingers from which pulse waves can be extracted. This embodiment uses a video of a human finger as the detection area for illustration.

[0031] 2) Simultaneously acquire video and corresponding diffuse reflectance spectra of the subject's finger area, and obtain blood glucose concentration values ​​through invasive methods as the true values; focus the optical fiber of the spectrometer onto the skin tissue area of ​​the finger under illumination for acquisition, with an acquisition frame rate of 30fps and an acquisition time of 30 seconds (or more); acquire video of the skin tissue area of ​​the finger under illumination, including the optical fiber acquisition, with the acquisition time synchronized with the spectrometer acquisition time, the CCD camera frame rate of 30fps, and the acquired image size of 1920×1000; the subject remains relatively still during the shooting.

[0032] Step 2: Perform image processing on the video acquired in Step 1 to obtain the IPPG signal;

[0033] 1) For the first frame of the acquired subject skin tissue video, a rectangular matrix is ​​used to select any pixel region [a1, b1] containing the skin tissue image as the region of interest for IPPG signal extraction; where a1 is the height and b1 is the width, and a1 and b1 are smaller than the image size acquired by the CCD camera.

[0034] 2) Calculate the average pixel value of the region of interest to obtain the original IPPG signal;

[0035] The pixel value within the region of interest [a1, b1] can be calculated using formula (1):

[0036] C(x,y)=I×(ρ s (t)+ρ d (t))+V n (1)

[0037] Where C(x,y) represents the light intensity value corresponding to the pixel with coordinates (x,y); I represents the light intensity of the light source; ρ s (t) and ρ d (t) represent the specular reflection coefficient and the diffuse reflection coefficient, respectively; V n This represents the quantization noise of the CCD camera.

[0038] Formula (2) is used to average all pixels in each frame of the image to remove V. n ,Right now:

[0039]

[0040] in, This represents the average light intensity of pixels in a frame of an image. Polarizers and analyzers remove specular reflections ρ. s (t), then all of them under time series t The IPPG signal composed of the set is shown in formula (3):

[0041]

[0042] 3) The original IPPG signal is subjected to RGB color channel separation and a preprocessing algorithm to remove interference from non-physiological parameters, and then the IPPG signal of the G channel is extracted.

[0043] 4) Use the preprocessed IPPG signal of the G channel as the time-domain signal of blood glucose;

[0044] Step 3: Process the diffuse reflectance spectrum of the skin tissue collected in Step 1 to obtain the diffuse reflectance spectrum of human skin tissue;

[0045] 1) Collect the diffuse reflectance spectrum of the subject's skin tissue, and calculate the average diffuse reflectance spectrum of the human tissue by averaging multiple diffuse reflectance spectrum data.

[0046] 2) Use the average diffuse reflectance spectrum as the spectral domain signal of blood glucose;

[0047] Step 4: Perform image processing on the skin tissue video acquired in Step 1 to obtain a single-frame image. Select any frame from the single-frame image to obtain the image's spatial dimension features.

[0048] 1) For any frame of a single-frame image, select the region containing the skin tissue image, and use a rectangular matrix to select any pixel position [a2, b2] as the region of interest for image spatial dimension feature extraction; where a2 is the height and b2 is the width, and a2 and b2 are smaller than the image size captured by the CCD camera (a2 and b2 can be the same as a1 and b1).

[0049] 2) Extract spatial dimension features of the region of interest image using a noise reduction autoencoder algorithm;

[0050] 3) Process the spatial dimension features, remove redundant features, and obtain the spatial dimension features related to blood glucose concentration as the spatial signal of blood glucose.

[0051] Step 5: Repeat steps 1, 2, 3 and 4 to collect time-domain, spectral domain, and spatial domain signals, as well as blood glucose values, from different subjects. The IPPG signal of the G channel of multiple subjects is used as the time-domain signal of blood glucose, the diffuse reflectance spectrum is used as the spectral domain signal of blood glucose, the spatial dimension features are used as the spatial domain signal of blood glucose, and the invasive blood glucose value is used as the true value.

[0052] 1) The number of subjects collected is more than 100 to obtain a wider dataset.

[0053] Step 6: Use the time-spectrum-space signal and invasive blood glucose reference values ​​obtained in Step 5 as the dataset. Divide the dataset into a training set and a test set. Use a deep learning algorithm to train a general blood glucose concentration prediction model on the training set. Then, input the test set into the general blood glucose concentration prediction model to obtain the predicted blood glucose concentration value for each subject in the test set.

[0054] 1) Use the spatiotemporal signal obtained in step 5 and the invasive blood glucose reference value as the dataset. Specifically, the spatiotemporal signal of each subject is the input to the dataset, and the corresponding invasive blood glucose value is the output. Randomly divide the dataset into training and test sets with different proportions.

[0055] 2) Using random forest regression model, support vector machine regression model, etc., based on the training set, input time-spectral-space blood glucose related signals and invasive blood glucose concentration values ​​to establish a non-specific blood glucose concentration prediction model.

[0056] 3) Input the time-spectral-space blood glucose correlation signal of the test set into the non-specific blood glucose concentration prediction model to predict the blood glucose concentration value;

[0057] The above embodiments are only used to illustrate and not limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention without departing from the spirit and scope of the present invention. Any modifications or partial substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for non-invasive optical blood glucose concentration detection based on time-spectrum-spatial multi-parameter fusion, characterized in that: Includes the following steps: Step 1: Simultaneously acquire video of the skin tissue of the subject with a pulse and the corresponding diffuse reflectance spectrum, and obtain the blood glucose concentration value as the true value by invasive finger blood sampling. Step 2: Perform image processing on the video of the skin tissue acquired in Step 1 to obtain the IPPG signal; 1) For the first frame of the acquired subject skin tissue video, a rectangular matrix is ​​used to select any pixel region [a1, b1] containing the skin tissue image as the region of interest for IPPG signal extraction; where a1 is the height and b1 is the width, and a1 and b1 are smaller than the image size acquired by the CCD camera; 2) Calculate the average pixel value of the region of interest to obtain the original IPPG signal; 3) The original IPPG signal is subjected to RGB color channel separation and a preprocessing algorithm to remove non-physiological parameter interference, and then the IPPG signal of the G channel is extracted; 4) The preprocessed IPPG signal of the G channel is used as the time-domain signal of blood glucose. Step 3: Process the diffuse reflectance spectrum of the skin tissue collected in Step 1 to obtain the diffuse reflectance spectrum of human skin tissue; The diffuse reflectance spectrum of the subject's skin tissue was collected, and the average diffuse reflectance spectrum of the human tissue was obtained by averaging multiple diffuse reflectance spectrum data. The average diffuse reflectance spectrum is used as the spectral domain signal of blood glucose. Step 4: Perform image processing on the skin tissue video acquired in Step 1 to obtain a single-frame image. Select any frame of the single-frame image to obtain the image spatial dimension features. 1) For any frame of a single-frame image, select the region containing the skin tissue image, and use a rectangular matrix to select any pixel position [a2, b2] as the region of interest for image spatial dimension feature extraction; where a2 is the height, b2 is the width, a2 and b2 are smaller than the image size captured by the CCD camera, and a2 and b2 are the same as a1 and b1. 2) Extract spatial dimensional features of the region of interest image using a noise reduction autoencoder algorithm; 3) Process the spatial dimension features, remove redundant features, and obtain the spatial dimension features related to blood glucose concentration as the spatial signal of blood glucose. Step 5: Repeat steps 1, 2, 3 and 4 to collect time-domain, spectral domain, and spatial domain signals, as well as blood glucose values, from different subjects. The IPPG signal of the G channel of multiple subjects is used as the time-domain signal of blood glucose, the diffuse reflectance spectrum is used as the spectral domain signal of blood glucose, the spatial dimension features are used as the spatial domain signal of blood glucose, and the invasive blood glucose value is used as the true value. Step 6: Using the time-spectrum-space signal and invasive blood glucose reference value obtained in Step 5 as a dataset, divide the dataset into a training set and a test set. Use a deep learning algorithm to train a general blood glucose concentration prediction model on the training set, and then input the test set into the general blood glucose concentration prediction model to obtain the predicted blood glucose concentration value for each subject in the test set.

Images