Method, system, medium and device for image processing in analyte detection

Abstract

To provide an image processing method or the like belonging to the field of optical analysis and used for detecting an object to be analyzed.SOLUTION: The image processing method includes an image acquisition step of acquiring an infrared grayscale image obtained by photographing a first region, a segmentation step of distinguishing a dark point and a bright point according to a gradient of a pixel and segmenting the infrared grayscale image into a dark point region and a bright point region, a point selection step of calculating and selecting a first target candidate point, and a screening step of performing outlier detection and screening on the first target candidate point to obtain a position of a first target. The area of the skin where the blood vessel is located can be distinguished from the area other than the area where the blood vessel is located, and the optical data acquisition points are selected in the corresponding area, so that the acquired optical data is more accurate. When the region where the blood vessel is located is distinguished from the region other than the region where the blood vessel is located, outliers in the candidate points may be effectively screened out, thereby further reducing the influence of the position of the abnormal point on the accuracy of the collection result.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present invention relates to the field of optical analysis, and in particular to image processing methods, systems, media and devices in analyte detection. [Background technology]

[0002] Commonly used analyte detection techniques include electrochemical and optical methods. Taking glucose detection in humans as an example, Patent Document US20100065441A1 discloses an analyte monitoring system, device, and method in which a sensor is implanted subcutaneously and electrochemically reacts with subcutaneous glucose to obtain glucose levels. The advantage of such a solution is that it significantly simplifies user experience, eliminating the need for multiple punctures and blood sampling, and allowing real-time glucose data to be collected as needed within a day. However, the disadvantage is that the subcutaneous implantation of the sensor is an invasive method. Patent Document US20160287147A1 discloses a non-invasive in-vivo measurement device using Raman spectroscopy to measure blood glucose levels in vivo. The advantage of such a solution is that it is more accurate than the electrochemical method of US20100065441A1, but the disadvantage is that it currently relies on laboratory-level Raman spectroscopy systems to implement, which are large, expensive, and difficult to make portable.

[0003] Furthermore, Patent Document CN118078277A discloses a non-invasive blood glucose detection method based on hyperspectral data analysis, which enables non-invasive detection. In this document, absorption spectroscopy is used, and the collected and analyzed spectral signals include not only the spectral signal of blood glucose but also the spectral signals of components such as skin tissue, resulting in a mixture of spectral signals of different wavelengths, making it difficult to precisely separate and extract the spectral signals related to blood glucose. At the same time, factors such as the excitation light source, differences in human skin color, and epidermal layer thickness also affect the intensity of the spectral signals, resulting in differences in the intensity of the spectral signals. The final collected spectral signals are easily affected and cannot be strongly correlated with blood glucose concentration, thereby affecting the accurate measurement of blood glucose concentration. Therefore, unless spectral data solely related to the location of blood vessels can be collected, it is always difficult to significantly improve the accuracy of optical detection results. Summary of the Invention [Problem to be solved by the invention]

[0004] SUMMARY OF THE INVENTION In view of the shortcomings of the prior art, it is an object of the present invention to provide an image processing method, system, medium and device for analyte detection. [Means for solving the problem]

[0005] The image processing method for analyte detection provided by the present invention comprises: an image capturing step of capturing an infrared grayscale image of the first region; a segmentation step of distinguishing between dark spots and bright spots according to pixel gradients and segmenting the infrared grayscale image into dark spot regions and bright spot regions; A point selection step of calculating a grayscale gradient of a pixel point combination consisting of each pixel point in the dark point region and a surrounding pixel point, and selecting a predetermined number of pixel point combinations having a large grayscale gradient as first target candidate points; The method includes a screening step of detecting outliers for the first target candidate points, screening and obtaining the first target candidate points after screening using a predetermined multiple of standard deviation, and determining the positions of the first target candidate points after screening as the positions where the first target exists.

[0006] Furthermore, the screening conditions in the screening step are: |X-μ|≦T*σ, Here, X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a preset multiple, and σ represents the standard deviation.

[0007] Furthermore, the point selection step comprises: Further comprising: calculating a grayscale gradient of a pixel point combination consisting of each pixel point and a surrounding pixel point in the bright point region; and selecting a predetermined number of pixel point combinations having a small grayscale gradient as second target candidate points; The screening step comprises: The method further includes performing outlier detection on the second target candidate points, screening and obtaining the second target candidate points after screening using a predetermined multiple of the standard deviation, and determining the positions of the second target candidate points after screening as positions where the second target is present.

[0008] The method for detecting an analyte provided by the present invention employs the image processing method for detecting an analyte described above.

[0009] The image processing system for analyte detection provided by the present invention comprises: an image acquisition module for acquiring an infrared grayscale image obtained by imaging the first region; a segmentation module for distinguishing between dark spots and bright spots according to pixel gradients and segmenting the infrared grayscale image into dark spot regions and bright spot regions; a point selection module that calculates grayscale gradients of pixel point combinations consisting of each pixel point and surrounding pixel points in the dark point region, and selects pixel point combinations having a predetermined number of large grayscale gradients as first target candidate points; and a screening module that performs outlier detection on the first target candidate points, screens and obtains the first target candidate points after screening using a preset multiple of standard deviation, and determines the position of the first target candidate points after screening as the position where the first target is located.

[0010] Furthermore, the screening conditions of the screening module are: |X-μ|≦T*σ, Here, X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a preset multiple, and σ represents the standard deviation.

[0011] Further, the point selection module: Further comprising: calculating a grayscale gradient of a pixel point combination consisting of each pixel point and a surrounding pixel point in the bright point region; and selecting a predetermined number of pixel point combinations having a small grayscale gradient as second target candidate points; The screening module comprises: The method further includes performing outlier detection on the second target candidate points, screening and obtaining the second target candidate points after screening using a predetermined multiple of the standard deviation, and determining the positions of the second target candidate points after screening as positions where the second target is present.

[0012] The analyte detection system provided by the present invention employs the image processing system for analyte detection described above.

[0013] The computer-readable storage medium having a computer program stored thereon provided by the present invention realizes the image processing step in the analyte detection when the computer program is executed by a processor.

[0014] The electronic device provided by the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, which, when executed by the processor, realizes the image processing step in the analyte detection. [Effects of the Invention]

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The technical solution of the present application can distinguish between blood vessel areas and non-blood vessel areas on the skin, thereby selecting optical data collection points within the corresponding areas and making the collected optical data more accurate. In addition, when distinguishing between blood vessel areas and non-blood vessel areas, outliers in the candidate points can be effectively screened out, thereby further reducing the impact of abnormal point locations on the accuracy of the collection results. 2. The detection method of the present application can obtain spectral data of various regions by utilizing the characteristics of the non-uniform distribution of the analyte in the imaging area, and since the distribution of components other than the analyte in the imaging area is relatively uniform, the difference in spectral data in different regions directly reflects information correlated with the spectral data of the analyte, such as the concentration of the analyte, which basically eliminates the influence of components other than the analyte. 3. The detection method of the present application uses fluorescence spectrum for detection, avoiding the traditional method of using Raman method to measure the analyte, realizing low-cost and miniaturized detection system, and achieving the goal of real-time detection.

[0016] Other features, objects and advantages of the present invention will become more apparent through the detailed description given below, given by way of non-limiting example with reference to the drawings in which: [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a flowchart of the first embodiment. [Figure 2] FIG. 1 is a schematic diagram of a first image collected in Example 2. [Figure 3] FIG. 10 is a schematic diagram of a second image collected in Example 2. [Figure 4] FIG. 10 is a diagram illustrating the principle of a detection model according to a second embodiment. [Figure 5] FIG. 1 is a schematic diagram of detection point-reference point spectrum data obtained in Example 2. [Figure 6] These are experimental results of the accuracy of the analytical results of the analytical model. [Figure 7] 1 is a structural schematic diagram of an analyte detection device provided by Example 5. FIG. [Figure 8] FIG. 1 is a structural schematic diagram of an electronic device provided by Example 6. [Figure 9] FIG. 1 is a schematic diagram of image processing in analyte detection. [Figure 10] FIG. 10 is a structural schematic diagram of a wristwatch for detecting an analyte provided by Example 5. [Figure 11] FIG. 2 is a schematic diagram of the back of the wristwatch for detecting an analyte. [Figure 12] FIG. 1 is an exploded view of a wristwatch for detecting an analyte. [Figure 13] FIG. 1 is a schematic diagram illustrating a state in which a wristwatch for detecting an analyte is used. DETAILED DESCRIPTION OF THE INVENTION

[0018] The present invention will be described in detail below with reference to specific examples. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any way. It should be noted that those skilled in the art may make some modifications and improvements without departing from the concept of the present invention, all of which fall within the protection scope of the present invention.

[0019] Example 1 FIG. 1 is a flow chart of this embodiment, and the method for detecting an analyte of this embodiment includes the following steps:

[0020] Imaging step: Irradiating a first region with light within a predetermined wavelength range through a light source, and imaging the first region through an imaging spectrum detector to obtain an image of the imaging region. By irradiating the first region with light within the predetermined wavelength range, distribution data and spectral data of the reflected or excited signals generated by the analyte when the light is irradiated in the imaging region can be reflected in the image. The first region may be a specific region on the surface of human skin. To prevent external light such as ambient light from affecting the detection, the collection window of the imaging spectrum detector must be tightly attached to the surface of the human skin in the first region. The imaging region refers to the region within the lens range of the imaging spectrum detector. Generally, the imaging region may be a part of the first region or the same region as the first region.

[0021] Since light rays with different wavelength ranges are required to obtain analyte distribution data and spectral data, two methods can be used: light rays with a wider wavelength range provided by one light source, or light rays with narrower wavelength ranges provided by two light sources. When one light source is used, the wavelength range of the light rays provided by the light source must simultaneously cover the wavelength range for obtaining analyte distribution data and the wavelength range for obtaining analyte spectral data. When two light sources are used, the two light sources provide different light rays, with the wavelength of one light ray covering the wavelength range for obtaining analyte distribution data and the wavelength of the other light ray covering the wavelength range for obtaining analyte spectral data. At the same time, when one light source is used, only one image is captured, and when two light sources are used, two images are captured. For ease of processing, the capturing areas of the two images must be the same, i.e., the collection window of the imaging spectrum detection device does not move on the surface of human skin.

[0022] In the present application, the analyte may be vascular glucose, ketones, alcohol, lactate, oxygen, hemoglobin A1C, acetylcholine, amylase, bilirubin, cholesterol, chorionic gonadotropin, creatine kinase (e.g., CK-MB), creatine, creatinine, DNA, fructosamine, glutamine, growth hormone, hormones, peroxides, prostate-specific antigen, prothrombin, RNA, thyroid-stimulating hormone, or troponin, or may be a drug such as an antibiotic (e.g., gentamicin, vancomycin, etc.), digitoxin, digoxin, a drug of abuse, theophylline, or warfarin. In embodiments where more than one analyte is detected, the analytes may be monitored at the same or different times. In other embodiments, the analyte may be any other substance within a body surface that can be noninvasively detected using the present invention.

[0023] Spectral acquisition step: Acquire spectral data from the image, which reflects the non-uniform distribution of reflected or excited signals generated by the object to be analyzed when the light beam is irradiated in the imaging region via the imaging spectral detection device. Specifically, the imaging region can be divided based on different distribution status data, and positions for acquiring spectral data from different sections can be selected.

[0024] Analysis stage: Based on the acquired spectral data, information about the analyte in the imaging area is acquired, including information about the analyte correlated with the spectral data. Because the distribution of the analyte in different sections is different, the reflected or excited signals generated by the analyte when irradiated with light also differ. Taking human skin as an example, it is divided into three parts: the epidermis, the dermis, and the subcutaneous tissue, and blood vessels such as veins are located in the subcutaneous tissue. UV light can be used to irradiate skin areas with blood vessels and skin areas without blood vessels to acquire corresponding spectral data, or to irradiate skin areas with thick blood vessels and skin areas with thin blood vessels to acquire corresponding spectral data. The difference between the two spectral data can reflect information about the analyte correlated with the spectral data in the blood vessels. Intermediate information, such as data about the degree of impact of the analyte on the spectral data, can be acquired for further analysis, or information such as the concentration of the analyte can be directly acquired through an analytical model.

[0025] As shown in Figure 9, this embodiment provides an image processing method for analyte detection for segmentation in the spectrum acquisition stage, and the image processing method specifically includes the following steps:

[0026] Image acquisition step: An infrared grayscale image obtained by imaging the first region is acquired.

[0027] Segmentation stage: distinguish between dark and bright points according to the pixel gradient, and segment the infrared grayscale image into dark and bright areas.

[0028] Point selection stage: Calculate the grayscale gradient of pixel point combinations consisting of each pixel point in the dark point area and surrounding pixel points, and select a predetermined number of pixel point combinations with the largest grayscale gradients as first target candidate points.

[0029] The point selection step comprises: The method further includes calculating the grayscale gradient of pixel point combinations consisting of each pixel point and surrounding pixel points in the bright point region, and selecting a predetermined number of pixel point combinations with small grayscale gradients as second target candidate points.

[0030] Screening stage: Outlier detection is performed on the first target candidate points, and the first target candidate points after screening are obtained by screening using a preset multiple of the standard deviation. The position of the first target candidate points after screening is determined to be the position where the first target is located.

[0031] The screening conditions of the screening step are: |X-μ|≦T*σ, Here, X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a preset multiple, and σ represents the standard deviation.

[0032] The screening step comprises: The method further includes performing outlier detection on the second target candidate points, screening and obtaining the second target candidate points after screening using a predetermined multiple of the standard deviation, and determining the positions of the second target candidate points after screening as positions where the second target is present.

[0033] The blood vessels can be found accurately and well by the following steps: Step 1: In the infrared grayscale image, extract the main sub-regions containing the skin and blood vessels, and smooth them to reduce the influence of noise.

[0034] Step 2: Distinguish dark points from bright points based on pixel gradients. Dark points represent blood vessel regions, and bright points represent skin regions. First, calculate the grayscale gradient of a 2x2 period around each pixel point in the subregion. Then, find the peaks of the subregion, and select 20 consecutive maximum values ​​for the maximum (dark points) and minimum (bright points) as candidate locations for dark points.

[0035] Step 3: Perform outlier detection on the candidate scotoma, set a threshold based on the standard deviation, and perform filtering and screening to remove abnormal values ​​such as skin dullness and fluorescence. The threshold determination condition is |X-μ|≦T*σ, where μ represents the mean value of the scotoma or bright point, σ represents the standard deviation of the scotoma or bright point, and T sets the threshold value, usually a multiple of the standard deviation σ. Here, a threshold of 1 is set to determine whether the selected data point is an outlier (abnormal value), and more accurate coordinate locations of the scotoma, i.e., the coordinate locations of the blood vessels, are obtained.

[0036] Example 2 This embodiment is based on the first embodiment and takes glucose detection in human blood vessels as an example to provide a non-invasive glucose detection method, which includes the following steps:

[0037] Imaging step: Irradiate the skin where veins are located on the wrist, back of the hand, etc. with infrared light in a first wavelength range of 800 to 1000 nanometers, preferably in the near-infrared band, to capture a first image of the imaging area, and then irradiate the same location with ultraviolet light in a second wavelength range of 300 to 390 nanometers to capture a second image of the imaging area.

[0038] 2, where the horizontal axis is the horizontal coordinate of the first image, the vertical axis is the vertical coordinate of the first image, the white box represents the pixel point block of the detection point selected on the venous blood vessel, and the black box represents the pixel point block of the reference point selected on the surrounding skin. In the first image, part of the infrared light passes through the human skin, and part is absorbed by the human skin. At the same time, the area where the venous blood vessel is located is absorbed in large amounts by the venous blood vessel, so the grayscale value of the pixel in the area where the venous blood vessel is located is smaller, and the grayscale value of the pixel outside the area where the venous blood vessel is located is larger, which makes it easy to divide the imaging area into the area where the venous blood vessel is located and the area outside the area where the venous blood vessel is located.

[0039] As shown in Figure 3, the horizontal axis represents the horizontal coordinate of the second image, the vertical axis represents the vertical coordinate of the second image, the white box represents the pixel point block of the detection point selected on the venous blood vessel, and the black box represents the pixel point block of the reference point selected on the surrounding skin. Because it is difficult to distinguish between the area where the venous blood vessel is located and the area other than the area where the venous blood vessel is located in the second image, it is necessary to distinguish the first image. The purpose of using excitation light in the second wavelength range of 300-390 nanometers is to obtain a high-quality effective fluorescence spectrum signal. This is because the main response band of the imaging spectrum detection device is 400-800 nm. If the wavelength of the excitation light used is less than 300 nm, the main peak of the fluorescence spectrum of the excited fluorescence emission signal is located in the band <400 nm, making it difficult for the imaging spectrum detection device to obtain a high-quality effective fluorescence spectrum signal. If the wavelength of the excitation light used exceeds 390 nm, the excitation light itself will be visible light, and the spectral signal of the excitation light will be superimposed on the fluorescence spectral signal, making it difficult to extract an effective fluorescence spectral signal without interference from the excitation light. Glucose in venous blood vessels absorbs ultraviolet light in the wavelength range of 300-390 nanometers and then emits a fluorescent emission signal in the visible light band of 400-800 nm, which is within the effective response range of the imaging spectral detector. The characteristic spectral intensity of this fluorescent emission signal is positively correlated with the glucose concentration, resulting in higher fluorescence excitation efficiency.

[0040] Spectrum acquisition step: based on the grayscale distribution of the pixel points in the first image, divide the imaging area into an area where venous blood vessels are located and an area other than the area where venous blood vessels are located, select a detection point from the position of the area where venous blood vessels are located corresponding to the second image, select a reference point from the position of the area where non-venous blood vessels are located corresponding to the second image, and respectively obtain spectral data of the detection point and the reference point in the second image. Specifically, based on the grayscale values ​​of the pixel points in the second image, select one pixel point having a grayscale value that meets a predetermined requirement as the detection point from the area where venous blood vessels are located, or select a combination of this pixel point and adjacent pixel points as the detection point, and select one pixel point whose grayscale value of the selected detection point is within a predetermined deviation range from the area other than the area where venous blood vessels are located, or select a combination of this pixel point and multiple adjacent pixel points as reference points, and calculate and obtain fluorescence spectral data of the detection point and the reference point in the second image. The spectral data can be selected from a single pixel point of the detection point, the reference point, or an average of a combination of multiple pixel points, and can be appropriately selected based on the width of the blood vessel. Averaging a combination of multiple pixel points improves the signal-to-noise ratio but is limited by the width of the blood vessel and avoids acquiring data from areas outside the blood vessel. Selecting a single pixel point provides high spatial resolution and is suitable for situations with thin blood vessels, but has a lower signal-to-noise ratio. As a preset requirement for grayscale values, it is possible to use the point with the smallest grayscale value as the detection point, but this application is not limited to this. The calculation results are shown in Figure 5, where the horizontal axis is wavelength (unit: nm) and the vertical axis is relative radiance (unit: W / nm). The solid line represents the spectral data of the detection point, and the dotted line represents the spectral data of the reference point.Here, the reason why the grayscale value of the reference point and the grayscale value of the selected detection point are within the predetermined deviation range is that the skin in the imaging area has influencing factors such as skin color, blemishes, and cosmetics, which may directly affect the spectral data of the reference point. However, the first image does not distinguish the areas of these influencing factors, so by setting a predetermined deviation range of the grayscale value, these influencing factors can be effectively excluded. Furthermore, because the grayscale value and the grayscale value of the selected detection point are within the predetermined deviation range, it is guaranteed that a reference point close to the detection point, such as the edge of a venous blood vessel, will be selected. This ensures that, excluding the blood vessels, the parameters of the remaining epidermis, dermis, and subcutaneous tissue, such as color and thickness, are closest. This makes it possible to exclude as much as possible the influence of non-analyte objects due to the deviation between the spectral data of the detection point and the spectral data of the reference point.

[0041] In addition to the spectral reconstruction algorithm, the method of obtaining the spectral data is to form the radiation calibration coefficient through pre-radiation calibration, and then calculate the grayscale value * radiation calibration coefficient to obtain the spectral line.

[0042] When a combination of multiple pixel points is selected from the detection point, the fluorescence spectrum data of the detection point can be the average value of the fluorescence spectrum data of these pixel points. At the same time, the number of detection points and reference points can be one or more. When the number of detection points and reference points is multiple, the average value of the fluorescence spectrum data of all the detection points and the average value of the fluorescence spectrum data of all the reference points can be calculated respectively.

[0043] Analysis stage: The acquired spectral data of the detection points and reference points is preprocessed and then input into a trained detection model, which outputs the glucose concentration or an intermediate result showing the correlation between glucose and the spectral data. When training the detection model, it is necessary to simultaneously acquire the subject's spectral data and accurate test results such as blood test results, and use the spectral data as the input of the detection model and the blood test results as the output of the detection model to train the detection model.

[0044] The detection model may employ a convolutional neural network model, which sequentially includes an input layer, at least two convolutional layers, at least two activation function layers, a flattened layer, a fully connected layer, and an output layer, where the convolutional layers and the activation function layers are spaced apart, and the activation function used in the activation function layer is a Relu function.

[0045] Here, the size of the convolution kernel of each layer in the convolutional neural network model is 1, the number of convolution kernels in the first convolutional layer is 32, and the number of convolution kernels in the second layer is 64. Both are used to extract blood glucose features and nonlinearly transform the output of the convolutional layer through an activation function. The Flatten layer flattens the output of the convolutional layer into a one-dimensional vector to facilitate connection to the subsequent fully connected layer, and the final output dimension is 1. In the model training process, the Adam optimizer is used to train the model, and the mean squared error is used as the loss function, and the mean absolute error is simultaneously calculated as the performance indicator for model evaluation.

[0046] If the output result of the detection model is a glucose concentration, the training for obtaining the detection model is stopped when the error between the output result and the measured standard glucose concentration value satisfies a preset condition. If the output result of the detection model is an intermediate result correlating glucose with spectral data, such as an intermediate neuron result, the training for obtaining the detection model is stopped when the error between the output result and the intermediate neuron result satisfies a preset condition. The intermediate neuron result is further processed by model correction to obtain the glucose concentration.

[0047] As shown in Figure 4, the input layer is a spectral data input layer obtained by preprocessing the original spectral data. The hidden layer is an intermediate hidden layer that uses deep learning convolutional operations to combine features and output the final predicted blood glucose concentration value as an output layer. Deep learning convolutional operations can also output one neuron, Output1, as an intermediate result after combining features. The intermediate result, Output1, and the two infrared IR feature intensity values ​​are then used for further model training to further correct the blood glucose prediction error and output the final predicted blood glucose concentration value, Output2. The degree of training of the detection model can be determined by setting different parameters as needed. The extracted glucose feature values ​​are continuously trained according to different parameter settings. When the error between the output result and the standard glucose value of the above label value meets the requirements, the training for obtaining the detection model is stopped.

[0048] Through multiple repeated training, the neurons can learn the corresponding change rules between different glucose concentrations and glucose spectrum features of different samplers, thereby improving the generality of the detection model and achieving the goal of predicting the glucose concentrations of different users.

[0049] The entire glucose detection process does not require blood sampling or skin puncture or skin implantation, but rather obtains the subject's spectral information based on the fluorescence spectrum, and then obtains the subject's glucose detection result based on the spectral information, thereby avoiding pain and discomfort and improving the discomfort and convenience of detection. This method allows for detailed distinction between the spectral signals at the blood vessel site and the skin site, allowing for accurate extraction of the subsequent glucose signal, and at the same time, the intensity of the spectral signal and the glucose concentration are closely correlated, achieving accurate measurement of glucose concentration, more accurate detection results, and more convenient processing.

[0050] Figure 6 shows a schematic diagram of the experimental results of the trained detection model. The horizontal axis represents the reference blood glucose concentration (unit: mmol / L) collected by the blood glucose meter, and the vertical axis represents the blood glucose concentration (unit: mmol / L) predicted by the patented method. The total number of samples collected by the subjects was 2,037, of which 1,537 samples were in the training set and 500 samples were in the prediction set. The figure shows the distribution of the detection results of the detection model. The MARD value of the predicted samples was 11.32%, and they were classified into region A and region B. The samples in region A accounted for 87.03%, and the samples in region B accounted for 12.77%, indicating that the detection accuracy of the detection model is relatively high.

[0051] Example 3 This example is based on Example 2, and replaces infrared light with visible light to provide a non-invasive glucose detection method, which includes the following steps:

[0052] Imaging step: Visible light is irradiated onto the skin where veins are located, such as on the wrist or back of the hand, to capture a first image of the imaging area, and ultraviolet light within a second wavelength range of 300 to 390 nanometers is irradiated onto the same location to capture a second image of the imaging area.

[0053] In the first image, the color of the area where the venous blood vessels are located is different from the color of the area other than the area where the venous blood vessels are located, so the imaging area can be easily divided into the area where the venous blood vessels are located and the area other than the area where the venous blood vessels are located.

[0054] In the second image, it is difficult to distinguish between areas where venous blood vessels are located and areas where they are not, so it is necessary to distinguish between the first image. The purpose of using excitation light in the second wavelength range of 300-390 nanometers is to obtain a high-quality effective fluorescence spectral signal. This is because the main response band of the imaging spectral detection device is 400-800 nm. If the wavelength of the excitation light used is less than 300 nm, the main peak of the fluorescence spectrum of the excited fluorescence emission signal is located in the band <400 nm, making it difficult for the imaging spectral detection device to obtain a high-quality effective fluorescence spectral signal. If the wavelength of the excitation light used is greater than 390 nm, the excitation light itself is visible light, and the spectral signal of the excitation light is superimposed on the fluorescence spectral signal, making it difficult to extract the effective fluorescence spectral signal without interference from the spectral signal of the excitation light. Glucose in the venous blood vessels absorbs ultraviolet light in the wavelength range of 300 to 390 nanometers, and then emits fluorescent radiation signals in the visible light band of 400 to 800 nm, which is located within the effective response range of the imaging spectrum detection device. The characteristic spectral intensity of the fluorescent radiation signals is positively correlated with the glucose concentration and has higher fluorescence excitation efficiency.

[0055] Spectrum acquisition step: based on the grayscale distribution of the pixel points in the first image, divide the imaging area into an area where venous blood vessels are located and an area other than the area where venous blood vessels are located, select a detection point in the second image from the area where venous blood vessels are located, select a reference point from the second image corresponding to the area other than the area where venous blood vessels are located, and obtain spectral data of the detection point and the reference point, respectively. Specifically, based on the grayscale values ​​of the pixel points in the second image, select one pixel point having a grayscale value that meets a predetermined requirement from the area where venous blood vessels are located, or select a combination of this pixel point and its adjacent pixel points as the detection point, select one pixel point from the area other than the area where venous blood vessels is located, or select a combination of this pixel point and multiple adjacent pixel points as reference points, and calculate and obtain fluorescence spectral data of the detection point and the reference point. Here, the reason why the grayscale value of the reference point and the grayscale value of the selected detection point are within a preset deviation range is that the skin in the imaging area has influencing factors such as skin color, blemishes, cosmetics, etc., which may directly affect the spectral data of the reference point. However, in the first image, the areas of all influencing factors are not distinguished at the same time, and by setting a preset deviation range of the grayscale values, these influencing factors can be effectively excluded.

[0056] When a combination of multiple pixel points is selected from the detection point, the fluorescence spectrum data of the detection point can be the average value of the fluorescence spectrum data of these pixel points. At the same time, the number of detection points and reference points can be one or more. When the number of detection points and reference points is multiple, the average value of the fluorescence spectrum data of all the detection points and the average value of the fluorescence spectrum data of all the reference points can be calculated respectively.

[0057] Analysis stage: The acquired spectral data of the detection point and reference point are preprocessed and then input into a trained detection model to output the glucose concentration. When training the detection model, it is necessary to simultaneously acquire the subject's spectral data and accurate test results such as blood test results, use the spectral data as the input of the detection model, and use the blood test results as the output of the detection model to train the detection model.

[0058] The detection model may employ a convolutional neural network model, which sequentially includes an input layer, at least two convolutional layers, at least two activation function layers, a flattened layer, a fully connected layer, and an output layer, where the convolutional layers and the activation function layers are spaced apart, and the activation function used in the activation function layer is a Relu function.

[0059] Here, the size of the convolution kernel of each layer in the convolutional neural network model is 1, the number of convolution kernels in the first convolutional layer is 32, and the number of convolution kernels in the second layer is 64. Both are used to extract blood glucose features and nonlinearly transform the output of the convolutional layer through an activation function. The Flatten layer flattens the output of the convolutional layer into a one-dimensional vector to facilitate connection to the subsequent fully connected layer, and the final output dimension is 1. In the model training process, the Adam optimizer is used to train the model, and the mean squared error is used as the loss function, and the mean absolute error is simultaneously calculated as the performance indicator for model evaluation.

[0060] When the error between the output result of the detection model and the standard glucose value satisfies a preset condition, the training for obtaining the detection model is stopped.

[0061] The degree of training of the detection model needs to set different parameters as needed, and the extracted multiple glucose feature values ​​are continuously learned according to different parameter settings. When the error between the output result and the standard glucose value of the above label value meets the requirement, the training to obtain the detection model is stopped.

[0062] Through multiple repeated training, the neurons can learn the corresponding change rules between different glucose concentrations and glucose spectrum features of different samplers, thereby improving the generality of the detection model and achieving the goal of predicting the glucose concentrations of different users.

[0063] The entire glucose detection process does not require blood sampling or skin puncture, but rather obtains the subject's spectral information based on the fluorescence spectrum, and then obtains the subject's glucose detection result based on the spectral information, thereby avoiding pain and discomfort and improving the discomfort and convenience of detection. This method can finely distinguish the spectral signals from the blood vessel site and the skin site, allowing the subsequent glucose signal to be accurately extracted, and at the same time, the intensity of the spectral signal and the glucose concentration are closely correlated, achieving accurate measurement of the glucose concentration, more accurate detection results, and more convenient processing.

[0064] Example 4 This embodiment provides an analyte detection system, which can be realized by performing the process steps of the analyte detection method, i.e., those skilled in the art can understand the analyte detection method as a preferred embodiment of the analyte detection system. The analyte detection system includes:

[0065] The imaging module includes a light source that provides light within a predetermined wavelength range to illuminate a first region, and an imaging spectrum detector that captures the first region to obtain an image of the imaging region. By irradiating the imaging region with light within the predetermined wavelength range, the image can reflect distribution data and spectral data of the reflected or excited signals generated by the analyte when the light is irradiated in the imaging region. Because different wavelength ranges are required to obtain distribution data and spectral data of the analyte, the light may be light of two corresponding wavelength ranges, or one wavelength range may be wider and cover the two required wavelength ranges. When two types of light are used, two images are obtained, and for ease of processing, the imaging regions of the two images typically need to be the same.

[0066] In the present application, the analyte may be glucose, ketones, alcohol, lactate, oxygen, hemoglobin A1C, acetylcholine, amylase, bilirubin, cholesterol, chorionic gonadotropin, creatine kinase (e.g., CK-MB), creatine, creatinine, DNA, fructosamine, glutamine, growth hormone, hormones, peroxide, prostate-specific antigen, prothrombin, RNA, thyroid-stimulating hormone, troponin, or drugs such as antibiotics (e.g., gentamicin, vancomycin, etc.), digitoxin, digoxin, drugs of abuse, theophylline, and warfarin in an animal's blood vessels. In embodiments where more than one analyte is detected, the analytes can be monitored at the same or different times. In other embodiments, the analyte may be other substances in the fluid.

[0067] Spectral acquisition module: Acquires spectral data from the image, which reflects the non-uniform distribution of reflected or excited signals generated by the object to be analyzed when the imaging area is irradiated with light via the imaging spectrum detection device. Specifically, the imaging area can be divided based on different distribution status data, and positions for acquiring spectral data from different sections can be selected.

[0068] Analysis module: Based on the acquired spectral data, the analysis module acquires information about the analyte in the imaging area, including information about the analyte correlated with the spectral data. Because the distribution of the analyte in different sections is different, the reflected or excited signals generated by the analyte when irradiated with light are also different. By utilizing this characteristic, the difference between the spectral data can be obtained, and the information about the analyte correlated with the spectral data, such as the concentration of the analyte, can be accurately reflected.

[0069] As shown in Figure 9, a schematic diagram of image processing in analyte detection is shown. Specifically, in the spectrum acquisition module, this embodiment provides an image processing system in analyte detection for segmentation, and the image processing system specifically includes the following modules:

[0070] Image acquisition module: acquires an infrared grayscale image obtained by imaging the first region. Segmentation module: distinguishes between dark and bright points according to the pixel gradient, and segments the infrared grayscale image into dark and bright areas. Point selection module: Calculates the grayscale gradient of pixel point combinations consisting of each pixel point and surrounding pixel points in the dark point area, and selects a predetermined number of pixel point combinations with large grayscale gradients as the first target candidate points.

[0071] The point selection module: The method further includes calculating the grayscale gradient of pixel point combinations consisting of each pixel point and surrounding pixel points in the bright point region, and selecting a predetermined number of pixel point combinations with small grayscale gradients as second target candidate points.

[0072] Screening module: performs outlier detection on the first target candidate point, and screens and obtains the first target candidate point after screening using a preset multiple of the standard deviation, and determines the position of the first target candidate point after screening as the position where the first target exists.

[0073] The screening conditions of the screening module are: |X-μ|≦T*σ, Here, X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a preset multiple, and σ represents the standard deviation.

[0074] The screening module comprises: The method includes performing outlier detection on the second target candidate points, screening and acquiring the second target candidate points after screening using a predetermined multiple of standard deviation, and determining the positions of the second target candidate points after screening as positions where the second target is present.

[0075] The blood vessels are accurately located by the following modules: Module 1 focuses on the main sub-regions containing skin and blood vessels in the infrared grayscale image and smooths them to reduce the influence of noise. Module 2 distinguishes dark points from bright points based on pixel gradients, where dark points represent blood vessel regions and bright points represent skin regions. First, it calculates the grayscale gradient of a 2x2 period around each pixel point in the subregion, then finds the peaks of the subregion, and selects 20 consecutive maximum values ​​for both maxima (dark points) and minima (bright points) as candidate locations for dark points. Module 3 performs outlier detection on candidate scotomas, sets a threshold based on the standard deviation, and performs filtering and screening to remove abnormal values ​​such as skin dullness and fluorescence. The threshold determination condition is |X-μ|≦T*σ, where μ represents the mean value of scotomas or bright spots, σ ​​represents the standard deviation of scotomas or bright spots, and T sets the threshold, usually a multiple of the standard deviation σ. Here, a threshold of 1 is set to determine whether a selected data point is an outlier (abnormal value), resulting in a more accurate coordinate location of the scotoma, i.e., the coordinate location of the blood vessel.

[0076] Those skilled in the art will recognize that in addition to realizing the system provided by the present invention and its respective devices, modules, and units purely in the form of computer-readable program codes, the system provided by the present invention and its respective devices, modules, and units can also realize similar functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. by performing logic programming in method steps. Therefore, the system provided by the present invention and its various devices, modules, and units can be considered as hardware components, and the devices, modules, and units included therein that realize various functions can also be considered as structures within the hardware components, and the devices, modules, and units for realizing various functions can also be considered as both software modules for realizing methods or structures within the hardware components.

[0077] Example 5 Shown in Figure 7 is an electronic device of this embodiment, specifically an analyte detection device 200. The detection device 200 is a portable, non-invasive detection device for the human body, which can be used as a standalone detection device or integrated into a wristwatch or mobile phone, realizing convenient and rapid detection of analytes on the body surface.

[0078] The detection device 200 includes a light source 201, an imaging spectrum detector 202, a controller 203, a first bandpass filter 204, a second bandpass filter 206, and a lens 205. The controller 203 establishes an electrical or communication connection with the light source 201 and the imaging spectrum detector 202, respectively.

[0079] The light source 201 can provide light within a predetermined wavelength range. Since light of different wavelength ranges is required to obtain distribution data and spectral data of an analyte, two methods can be used: one light source that can provide light of a wider wavelength range, or two light sources that provide light of narrower wavelength ranges. When a single light source is used, the wavelength range of the light provided by the light source must simultaneously cover the wavelength range for obtaining distribution data of the analyte, such as a halogen lamp, and the wavelength range for obtaining spectral data of the analyte. When two light sources are used, the two light sources provide different light beams, with one light beam having a wavelength covering the wavelength range for obtaining distribution data of the analyte, and the other light beam having a wavelength covering the wavelength range for obtaining spectral data of the analyte, such as a combination of infrared and ultraviolet light, or a combination of visible light and ultraviolet light.

[0080] To uniformly illuminate the imaging area 100, a ring-shaped light source can be used, which has multiple light-emitting modules uniformly distributed on the same circumference. When there are two types of light sources, the light-emitting modules of the two types of light sources are arranged relative to each other.

[0081] The imaging spectral detector 202 can image the imaging region 100 to obtain a corresponding image upon command, and can also obtain corresponding spectral data upon command. The imaging spectral detector 202 includes a sensor and a periodic pixel-level optical filter structure disposed on a surface of the sensor. The periodic pixel-level optical filter structure is used to perform spectral modulation on an incident optical signal, thereby enabling the sensor to generate an image containing the spectral information of interest.

[0082] The periodic pixel-level optical filter structure includes a plurality of optical filter pixel channels with different shapes of pixel-level structures, each having the same specifications and size, uniformly arranged, and each having a length and width that is an integer multiple of the pixel point size of the pixel sensor. The optical filter pixel channels of the different shapes of the pixel-level optical filter structures correspond to different spectral wave filter coefficients, and the pixel-level optical filter structures with different spectral wave filter coefficients are periodically arranged after being combined in a fixed order. The sensor modulates the received first detected light through the periodic pixel-level optical filter structures arranged on its surface to form a mosaic image containing spectral information, and then reconstructs the spectral data using an algorithm.

[0083] The controller 203 is configured to control the light source to provide light within a predetermined wavelength range to illuminate a first region, control the imaging spectrum detector to image the first region, and acquire an image of the imaging region. The controller 203 is configured to control the imaging spectrum detector to acquire spectral data from the image, the spectral data reflecting the non-uniform distribution of reflected or excited signals generated by the analyte when the light is irradiated in the imaging region. Based on the acquired spectral data, information about the analyte in the imaging region is acquired, where the information about the analyte correlates with the spectral data. When the light source 201 is one type, one image is acquired; when the light source 201 is two types, two images are acquired. When the first light source is turned on, the second light source is turned off; similarly, when the second light source is turned on, the first light source is turned off, so that the two light sources do not interfere with each other.

[0084] The first bandpass filter 204 is located between the light source 201 and the imaging area 100, and its function is to pass light within a predetermined wavelength range and block light outside the predetermined wavelength range, thereby reducing the influence of other external light on the detection results.

[0085] The second bandpass filter 206 is located between the imaging spectrum detector 202 and the lens 205, and its function is to pass light in the wavelength range in which the reflection signal or excitation signal generated by the analyte when irradiated with light is located, and to block light in other wavelength ranges, thereby reducing the influence of reflection signals or excitation signals other than those of the analyte on the detection results.

[0086] The lens 205 can be used for focusing to obtain a clear image. In another embodiment, the second bandpass filter 206 can be located on one side of the lens 205, away from the imaging spectrum detector 202, but the present invention is not limited thereto.

[0087] According to the above description, FIG. 10 shows a wristwatch for detecting an analyte provided by this embodiment. The front of the wristwatch is a display, and as shown in FIG. 11, the back of the wristwatch is a light-transmitting window with a built-in detection device 200. As shown in FIG. 12, the light source 201 and the first band-pass filter 204 are all annular structures. The light-emitting modules of the light source 201 are distributed in a ring shape, and the emitted light is optically filtered by the first band-pass filter 204 to output light of a required wavelength, which is then irradiated onto the human body through the light-transmitting window on the back of the wristwatch. The reflected signal or excitation signal from the human body enters the light-transmitting window, passes through the central hollow part of the light source 201 and the first band-pass filter 204, and enters the second band-pass filter 206 through the lens 205. After being optically filtered by the second band-pass filter 206, it enters the imaging spectrum detection device 202. The imaging spectrum detector 202 is mounted on a circuit board 207, and a controller 203 (not shown) of the detector device 200 is also mounted on the circuit board 207. As shown in Figure 12, the watch can be worn on the inside of the wrist to more accurately identify the location of the venous blood vessels. As shown in Figure 13, the watch can be worn on the inside of the wrist to more accurately identify the location of the venous blood vessels.

[0088] Example 6 FIG. 8 is a structural schematic diagram of an electronic device provided by an embodiment of the present application, which includes at least one processor 501, as shown in FIG. 8, and a memory 502 communicatively connected to the at least one processor 501, wherein the memory 502 stores instructions executable by the at least one processor 501, the instructions are executed by the at least one processor 501, and the at least one processor 501 can perform the above-mentioned analyte detection method.

[0089] Here, the memory 502 and the processor 501 are connected in a bus manner, which may include any number of interconnected buses and bridges, connecting various circuits of one or more processors 501 and memories 502. The bus may also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be further described in the present invention. A bus interface provides an interface between the bus and a transceiver. The transceiver may be a single component or multiple components, for example, multiple receivers and transmitters, providing a unit for communicating with various other devices via a transmission medium. Data processed by the processor 501 is transmitted over a wireless medium via an antenna, which receives data and transmits data to the processor 501.

[0090] The processor 501 is responsible for bus management and general processing, and may also provide a variety of functions including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 502 may be used to store data used by the processor 501 when performing operations.

[0091] The present invention further provides a computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the above-described method for detecting an analyte.

[0092] That is, those skilled in the art will understand that all or some of the steps of implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a storage medium containing some instructions that enable a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or some of the steps of the methods described in each embodiment of the present application. The storage medium includes various media that can store program code, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0093] Those skilled in the art will understand that the above embodiments are specific examples for realizing the present invention, and that in actual applications, various changes in form and details are possible without departing from the spirit and scope of the present invention.

[0094] The above describes specific examples of the present invention. The present invention is not limited to the above specific embodiments, and it should be understood that those skilled in the art can make various changes or modifications within the scope of the claims without affecting the essential content of the present invention. Unless inconsistent, the examples and features of the examples in this application can be combined with each other in any way. [Explanation of symbols]

[0095] 100: Imaging area 200:Detection device 201: Light source 202: Imaging spectrum detector 203: Controller 204: 1st bandpass filter 205: Lens 206: Second bandpass filter 207: Circuit board 501: Processor 502: Memory

Claims

1. 1. An image processing method for analyte detection, comprising: an image capturing step of capturing an infrared grayscale image of the first region; a segmentation step of distinguishing between dark spots and bright spots according to pixel gradients and segmenting the infrared grayscale image into dark spot regions and bright spot regions; a point selection step of calculating a grayscale gradient of a pixel point combination consisting of each pixel point in the dark point region and a surrounding pixel point, and selecting a predetermined number of pixel point combinations having a large grayscale gradient as first target candidate points; a screening step of detecting outliers for the first target candidate points, screening and acquiring the first target candidate points after screening using a standard deviation of a preset multiple, and determining the positions of the first target candidate points after screening as positions where the first targets are present.

2. The screening conditions of the screening step are: |X−μ|≦T*σ, where X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a predetermined multiple, and σ represents the standard deviation. The image processing method for detecting an analyte according to claim 1 .

3. The point selection step comprises: Further comprising: calculating a grayscale gradient of a pixel point combination consisting of each pixel point and a surrounding pixel point in the bright point region; and selecting a predetermined number of pixel point combinations having a small grayscale gradient as second target candidate points; The screening step comprises: The method further includes: detecting outliers for the second target candidate points; screening and acquiring the second target candidate points after screening using a standard deviation of a preset multiple; and determining the positions of the second target candidate points after screening as positions where the second target exists. The image processing method for detecting an analyte according to claim 2.

4. 1. A method for detecting an analyte, comprising: The image processing method for detecting an analyte according to any one of claims 1 to 3 is included. Method for detecting analytes.

5. 1. An image processing system for analyte detection, comprising: an image acquisition module for acquiring an infrared grayscale image of the first region; a segmentation module for distinguishing between dark spots and bright spots according to pixel gradients and segmenting the infrared grayscale image into dark spot regions and bright spot regions; a point selection module that calculates grayscale gradients of pixel point combinations consisting of each pixel point and surrounding pixel points in the dark point region, and selects pixel point combinations having a predetermined number of large grayscale gradients as first target candidate points; a screening module that performs outlier detection on first target candidate points, screens and acquires first target candidate points after screening using a standard deviation that is a preset multiple, and sets the positions of the first target candidate points after screening as positions where the first target is present.

6. The screening conditions of the screening module are: |X−μ|≦T*σ, where X represents the grayscale gradient of the pixel point combination, μ represents the average grayscale gradient, T represents a predetermined multiple, and σ represents the standard deviation.

6. An image processing system for detecting an analyte according to claim 5.

7. The point selection module: Further comprising: calculating a grayscale gradient of a pixel point combination consisting of each pixel point and a surrounding pixel point in the bright point region; and selecting a predetermined number of pixel point combinations having a small grayscale gradient as second target candidate points; The screening module includes: The method further includes: detecting outliers for the second target candidate points; screening and acquiring the second target candidate points after screening using a standard deviation of a preset multiple; and determining the positions of the second target candidate points after screening as positions where the second target exists. The image processing system for analyte detection according to claim 6.

8. 1. A system for detecting an analyte, comprising: A system for detecting an analyte, comprising the image processing system for detecting an analyte according to any one of claims 5 to 7.

9. A computer-readable storage medium having a computer program stored thereon, The computer-readable storage medium, characterized in that when the computer program is executed by a processor, the image processing step in the analyte detection according to any one of claims 1 to 3 is realized.

10. 1. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, The electronic device, characterized in that when the computer program is executed by a processor, the image processing step in the analyte detection according to any one of claims 1 to 3 is realized.

Images