Digital biomarker tracking method
By extracting multidimensional features from biomarker response images and compensating for environmental factors, a nonlinear mapping model was established, which solved the problem that biomarker detection is susceptible to environmental interference and enabled rapid and accurate personalized detection.
Patent Information
- Application Number
- CN202511492671.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2026-01-23
AI Technical Summary
Existing biomarker detection methods are susceptible to interference from environmental factors, lack accuracy and stability, and lack multidimensional image feature analysis and personalized health assessment methods, making it difficult to achieve rapid, convenient, and accurate detection.
By acquiring reaction images of biomarkers combined with reagents, image preprocessing and feature extraction are performed. Nonlinear reaction intensity functions and linear regression models are established, and combined with environmental compensation, personalized test reports are generated.
It significantly improves the accuracy and stability of detection, enabling rapid and convenient detection and dynamic monitoring of biomarker concentrations, and supporting personalized health assessments.
Smart Images
Figure CN121390102A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and particularly relates to a digital biomarker tracking method. BACKGROUND
[0002] As an important indicator for disease diagnosis, treatment monitoring and health management, biomarkers have received extensive attention in the medical and life science fields in recent years. Traditional biomarker detection relies on complex laboratory instruments and equipment, with long detection cycles and complex operations, which cannot meet the needs of on-site rapid detection and real-time monitoring. Especially in emergency, chronic disease management and personalized medicine scenarios, how to achieve convenient, accurate and real-time biomarker concentration detection has become a technical problem to be solved.
[0003] In the prior art, biomarker detection often relies on a single optical signal or chemical reaction result, ignoring the influence of environmental factors such as temperature, humidity and air pressure on detection accuracy, resulting in detection results being easily disturbed, and insufficient accuracy and repeatability. In addition, the lack of multi-dimensional feature analysis and environmental compensation of image information limits the application range and reliability of the detection method. SUMMARY
[0004] In order to make up for the above shortcomings, the present application provides a digital biomarker tracking method, aiming to improve the shortcomings of the prior art that the detection process is easily disturbed by environmental factors, the accuracy and stability of the detection results are insufficient, and there is a lack of multi-dimensional image feature comprehensive analysis and personalized health evaluation means, to realize rapid, accurate and portable biomarker concentration detection and dynamic monitoring.
[0005] The present application provides the following technical scheme, a digital biomarker tracking method, comprising: obtaining a reaction image of biomarker and reagent combination, and transmitting the image to a server; using a statistical method to screen and remove outliers in the image, using a least squares method to preliminarily correct the remaining image data, and then using an image algorithm for secondary correction; extracting a region of interest by an image segmentation algorithm, removing a background region irrelevant to the reaction, and extracting color, light intensity, turbidity and texture features of the preprocessed image; based on the compensated image features, using an infinite series expansion to establish a nonlinear reaction intensity function between the image features and the biomarker concentration, and using the function to establish a linear regression model to fit the curve of the concentration change over time.
[0006] Further, a two-dimensional code is arranged on the detection card corresponding to the reaction image, the two-dimensional code encodes product information including but not limited to reagent variety, batch number, production information and channel information; in the image uploading or processing process, the two-dimensional code content is parsed and associated with the detection data to realize the corresponding tracking of the detection result and the product source.
[0007] Further, the step of screening and removing outliers in the image is: detecting outliers in the image pixel or feature data by using statistical methods; identifying outliers deviating from the normal range by setting a threshold or based on distribution characteristics; removing or replacing the identified outliers with adjacent normal data.
[0008] Further, the steps of preliminary correction and secondary correction are: performing preliminary correction on the image data after removing outliers by using the least square method to fit and correct the basic color or light intensity parameters of the image; performing geometric distortion correction to correct the deformation of the image caused by the shooting device or the environment; adopting image registration technology to accurately align multiple images to ensure consistency in subsequent analysis; applying Gaussian filter denoising algorithm to preliminarily remove noise in the image; implementing motion blur repair by using image restoration algorithm to improve the blur caused by motion.
[0009] Further, the step of removing the background area unrelated to the reaction is: adopting image segmentation algorithm to analyze the preprocessed image, the image segmentation algorithm including but not limited to threshold segmentation, edge detection or deep learning model; determining the boundary of the reaction area, extracting the region of interest, and excluding the image background and irrelevant area; performing morphological processing on the extracted region of interest, such as erosion and dilation operation, to further refine the region boundary; and performing numerical processing on the detection signal corresponding to the extracted region, including subtracting the signal value of the negative group from the signal values of the experimental group and the positive group to eliminate the chemical reaction background noise, and then calculating the ratio of the signal values of the experimental group and the positive group to correct the influence of reagent attenuation.
[0010] Further, the step of extracting features from the received reaction image is: obtaining color features by HSV space histogram statistics and dominant color clustering; obtaining light intensity features by region gray mean, variance calculation and Gamma correction compensation; Obtain turbidity characteristics by scattered light spot area measurement and edge sharpness analysis; Obtain texture characteristics by energy value and entropy value calculation of gray level co-occurrence matrix.
[0011] Further, the step of establishing the nonlinear reaction intensity function by infinite technology is as follows: Taking the image feature vector as input, the multiple orthogonal polynomial basis function is used for infinite series expansion; The number of expansion terms is determined by small amount control before convergence and basis function expansion; The series expansion coefficient is solved by least square method, and the nonlinear reaction intensity function of image features to biomarker concentration is established.
[0012] Further, the step of establishing the linear regression model is as follows: Taking biomarker concentration as dependent variable and reaction intensity function as independent variable, linear regression equation is established by using historical test database; The fitting effect and reliability are verified by residual analysis and confidence interval calculation.
[0013] Further, the method further comprises: collecting user area meteorological data by GPS positioning, combining temperature, humidity and air pressure sensor data to compensate the regression model for environment; the step of environment compensation is as follows: Meteorological data of the area where the user is located is obtained by GPS positioning, including temperature, humidity and air pressure; The parameters of the regression model are dynamically adjusted by using the temperature, humidity and air pressure data collected by the environmental sensor in real time; According to the pre-established environment compensation model, the influence of environmental factors on biomarker concentration measurement is corrected.
[0014] Further, the method further comprises: comparing the results with the feature database, generating a personalized detection report and sending it to the user; the step of comparing with the feature database is as follows: The biomarker concentration obtained by mapping is compared with the standard reference range in the feature database, to determine whether the detection result is within the normal interval; Retrieving user historical detection data, analyzing the time variation trend and fluctuation mode of biomarker concentration; According to the personal information of the user, the corresponding personalized reference standard is matched, and the statistical data of the same group of people are compared horizontally; Based on the comparison result, a personalized detection report containing concentration value, trend analysis, risk assessment and health suggestion is generated.
[0015] The present application has the following beneficial effects: 1、The present application extracts multi-dimensional features from the biomarker-reagent reaction image, covering color, light intensity, turbidity, texture and other aspects, significantly improving the utilization rate of image information and the accuracy of detection. Real-time compensation is combined with environmental sensor data to effectively offset the influence of environmental factors such as temperature, humidity and air pressure on the detection results, thereby ensuring the stability and reliability of the detection data.
[0016] 2、The present application uses the infinite series expansion method to establish a nonlinear mapping model between image features and biomarker concentration, which can more accurately capture complex feature relationships and improve the accuracy of concentration prediction. At the same time, based on the concentration data obtained by mapping, a linear regression model is used for time series fitting to dynamically reflect the trend of the signs. BRIEF DESCRIPTION OF DRAWINGS
[0017] Figure 1 A flowchart of a digital biomarker tracking method proposed by the present application; Figure 2 An application schematic diagram of a digital biomarker tracking method proposed by the present application. DETAILED DESCRIPTION
[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0019] Embodiment one In the first embodiment of the present application, the present application provides a digital biomarker tracking method, as shown in Figure 1 The method comprises the following steps: A digital biomarker tracking method, characterized in that it comprises: S100: Obtain the reaction image of the biomarker combined with the reagent, and transmit the image to the server; Further, a two-dimensional code is arranged on the detection card corresponding to the reaction image, and the two-dimensional code encoding contains product information, including but not limited to reagent variety, batch number, production information and channel information; In the image uploading or processing process, the two-dimensional code content is parsed and associated with the detection data to realize the corresponding tracking of the detection result and the product source.
[0020] By arranging a two-dimensional code containing reagent variety, batch number, production information and channel information on the detection card, and parsing the two-dimensional code content in the image uploading or processing process, the detection result can be accurately bound to the specific product source, realizing the traceability of the whole detection process.
[0021] S200: screening and removing outliers in the image using statistical methods, using least squares method to preliminarily correct the remaining image data, and then performing secondary correction through image algorithm; Further, the step of screening and removing outliers in the image is: detecting outliers in the image pixel or feature data using statistical methods; identifying outliers deviating from the normal range by setting a threshold or based on distribution characteristics; removing or replacing the identified outliers with adjacent normal data.
[0022] Specifically, first, the acquired reaction image is converted into a pixel matrix or feature data such as color value, brightness value, texture parameter, etc. is extracted; then, statistical methods are used to detect outliers in the above-mentioned pixel or feature data, which can include mean-standard deviation method, box plot method (IQR method), Z-Score detection method, etc. For example, using the Z-Score method, the difference between each data point and the overall mean is calculated and divided by the overall standard deviation, and when the absolute value is greater than a set threshold (such as 3), it is determined to be an outlier. Subsequently, the threshold can be adjusted adaptively according to the historical sample distribution or real-time data fluctuation to improve the robustness of the detection. After identifying outliers, the outliers can be directly removed, or the mean or median of adjacent normal data can be used for replacement, thereby avoiding interference of null values on subsequent processing.
[0023] By using statistical methods to effectively identify and remove outliers in the reaction image, the interference of abnormal pixels or feature data on subsequent processing can be significantly reduced.
[0024] Further, the steps of preliminary correction and secondary correction are: using least squares method to preliminarily correct the image data after removing outliers to fit and correct the basic color or light intensity parameters of the image; performing geometric distortion correction to correct the deformation of the image caused by the shooting device or the environment; using image registration technology to accurately align multiple images to ensure consistency in subsequent analysis; applying Gaussian filter denoising algorithm to preliminarily remove noise in the image; implementing motion blur repair to improve the blur caused by motion through image restoration algorithm.
[0025] Specifically, after removing outliers, the effective image data is input into the least squares correction module. Least squares method is used to fit the color or light intensity distribution curve of the image, by calculating the least square error between the fitted curve and the original data, the color gain coefficient or light intensity correction factor is obtained, and it is applied to the whole image, realizing the preliminary unification of the whole color or brightness.
[0026] According to the lens parameters of the shooting device or the data collected by the calibration board, a geometric distortion model such as a barrel distortion or a pincushion distortion model is established, and the image pixel positions are re-mapped through a reverse mapping algorithm to restore the true geometric shape.
[0027] Then, image registration technology is used to align the positions of multiple images. The registration method can include feature point-based matching (such as SIFT, ORB algorithm) or gray-based mutual information matching, coordinate transformation after calculating translation, rotation and scaling parameters to ensure the consistency of the images in spatial position.
[0028] Next, a Gaussian filter algorithm is used to perform preliminary denoising processing on the image, and the pixels are weighted and averaged through a convolution kernel to smooth random noise while trying to preserve edge details. The size and standard deviation of the Gaussian kernel can be adaptively selected according to the noise intensity.
[0029] Finally, according to the blur direction and blur kernel length of the image, image restoration methods such as Wiener filtering, Lucy-Richardson algorithm, etc. are used to perform deconvolution processing on the blurred image to improve the image blur phenomenon caused by camera shake or movement of the photographed object, thereby obtaining an image with higher clarity.
[0030] Through the combined processing of least squares correction, geometric distortion correction, image registration, denoising and motion blur repair, the geometric precision, color consistency and clarity of the image can be significantly improved, and the interference of shooting conditions, equipment defects and motion factors on the detection data can be effectively reduced.
[0031] S300: Extract the region of interest by image segmentation algorithm, remove the background area irrelevant to the reaction, and extract the color, light intensity, turbidity and texture features of the preprocessed image; Further, the step of removing the background area irrelevant to the reaction is: An image segmentation algorithm is used to analyze the preprocessed image, which includes but is not limited to threshold segmentation, edge detection or deep learning model; Determine the boundary of the reaction area, extract the region of interest, and exclude the image background and irrelevant areas; Perform morphological processing on the extracted region of interest, such as erosion and dilation operations, to further refine the region boundary; And perform numerical processing on the detection signal corresponding to the extracted region, including subtracting the signal value of the negative group from the signal values of the experimental group and the positive group to eliminate the background noise of the chemical reaction, and then calculating the ratio of the signal values of the experimental group and the positive group to correct the influence of reagent attenuation.
[0032] Specifically, the pre-processed image is input into an image segmentation algorithm, which can be threshold segmentation, edge detection, or a deep learning model based on a convolutional neural network (CNN). Threshold segmentation separates the reaction region from the background region by setting a pixel intensity threshold; edge detection identifies the region boundary using gradient changes; and the deep learning model can be trained with labeled samples to automatically identify the reaction region.
[0033] According to the segmentation result, the boundary of the reaction region is determined, and a region of interest (ROI) is extracted, and all background parts irrelevant to the reaction are removed, to ensure that subsequent calculations are only performed on the effective reaction region.
[0034] Morphological operations, including erosion and dilation operations, are performed on the extracted region of interest to eliminate small isolated noise points and smooth the boundaries, thereby improving the accuracy and integrity of the region boundary.
[0035] The background noise of the detection signal corresponding to the region of interest is subtracted and the ratio is corrected. First, the signal values of the experimental group and the positive group are respectively subtracted from the signal values of the negative group to subtract the background noise of the chemical reaction, and then the ratio of the signal values of the experimental group and the positive group is calculated to eliminate the influence of reagent decay and other non-reactive factors on the signal intensity.
[0036] Through the above steps, the reaction region data with background interference removed, boundary accurate and signal corrected is finally obtained, providing high-quality input for subsequent image feature extraction and concentration calculation.
[0037] Further, the steps of feature extraction on the received reaction image are as follows: Color features are obtained by HSV space histogram statistics and dominant color clustering; Light intensity features are obtained by region gray mean, variance calculation and Gamma correction compensation; Turbidity features are obtained by scatter spot area measurement and edge sharpness analysis; Texture features are obtained by energy value and entropy value calculation of the gray level co-occurrence matrix.
[0038] Specifically, the input image is converted from the RGB color space to the HSV color space to improve the stability of color perception. By counting the histogram distribution of the hue channel of each pixel in the image, the color histogram feature reflecting the main color composition of the image is obtained. To further extract the main color information of the image, K-means clustering analysis is performed on the pixels in the HSV space, and the main color cluster centers in the image are extracted as the dominant color feature vector, which is used to quantitatively represent the color composition.
[0039] After the image is grayed, the reaction area is selected for pixel statistics, and the gray mean and gray variance of the area are calculated, which reflect the light intensity and distribution uniformity respectively. In order to overcome the influence of exposure difference or light change, the image is processed by Gamma correction, and the image brightness distribution is adjusted according to the preset gamma value, so that the images collected under different conditions have consistent contrast basis. After correction, the gray mean and variance are extracted again as the normalized light intensity features.
[0040] The scattered light spots appearing in the reaction area are extracted by image enhancement and threshold segmentation method, and the area of high-light area in the binary image is measured to estimate the light spot coverage area corresponding to the turbidity. At the same time, the edge detection (such as Sobel or Canny operator) is used to identify the clarity of the light spot edge in the image, and the edge gradient change rate is calculated to quantify the edge sharpness feature, which indirectly reflects the change of turbidity.
[0041] A gray level co-occurrence matrix is constructed for the gray level image of the reaction area to count the spatial relationship between the gray values in the image. The energy value and entropy value are extracted from the gray level co-occurrence matrix, where the energy value reflects the uniformity of the image texture, and the entropy value reflects the complexity of the image texture. The above texture features are used to assist in judging the structural details and reactant distribution state of the reaction area.
[0042] Through the above methods, the image multi-modal feature vector is constructed from four dimensions of color, brightness, turbidity and texture, which provides high-precision and multi-angle input data support for subsequent environment compensation and concentration mapping modeling.
[0043] S400: Based on the compensated image features, an infinite series expansion is used to establish a non-linear reaction intensity function between the image features and the biomarker concentration, and a linear regression model is established based on the function to fit the curve of the concentration change over time; Further, the step of establishing a non-linear reaction intensity function by using infinite technology is: The image feature vector is taken as the input, and a multivariate orthogonal polynomial basis function is used for infinite series expansion; The number of expansion terms is determined by a small amount of control before convergence and basis function expansion; The least square method is used to solve the series expansion coefficient to establish a non-linear reaction intensity function from image features to biomarker concentration.
[0044] Specifically, the whole analysis system is regarded as a mathematical function mapping model wherein, represents the image feature vector (including color, light intensity, turbidity, texture, etc.) after environment compensation, represents the corresponding biomarker concentration value (target output), represents an unknown nonlinear mapping function, which needs to be approximated and modeled by training data.
[0045] To approximate , an infinite series expansion will be performed on the definition domain : ; where each term represents a component composed of a set of multivariate orthogonal polynomial basis functions. For example, in a two-dimensional feature space, the expansion can take the following format (using Legendre orthogonal polynomials as an example): ; where and represent nonlinear transformations of the original features, represent the importance coefficients of these transformations in predicting the target. Finally, by superimposing a finite number of terms (ignoring high-order terms when they tend to zero), the original nonlinear function can be accurately approximated within a small error range. In practical engineering implementation, considering function convergence and computational efficiency, the expansion can be simplified by combining the monotone convergence theorem. That is, within the interval , when each term is a monotonic function, the condition is satisfied, where is the negligible error threshold. Therefore, when the contribution of a certain term is less than 1%, the expansion can be stopped. Usually, only the first 4-6 orders need to be expanded (depending on the complexity of the data), and convergence can be achieved.
[0046] Using the small quantity control method to dynamically determine the maximum number of expansion terms , the change in the output of the calculation model is calculated for each order. If , it is considered to be convergent, and the expansion is terminated. If the condition is not met, the expansion order is increased until the convergence requirement is met or the maximum limited order is reached.
[0047] By inputting the training sample set , the least squares method is used to solve the coefficients of each term in the expansion , that is, by minimizing the error function to obtain the weight coefficients of each series term, thereby constructing a complete nonlinear mapping model.
[0048] The infinite series expansion modeling technique described in this embodiment not only has mathematical convergence and interpretability, but also can achieve high-precision nonlinear function approximation under limited computing resources, ensuring the stability and accuracy of the mapping between image features and biomarker concentrations, and effectively improving the intelligent level of the detection system.
[0049] Further, the step of establishing a linear regression model is: A linear regression equation is established using a historical test database, with biomarker concentration as the dependent variable and reaction intensity function as the independent variable; The fitting effect and reliability are verified by residual analysis and confidence interval calculation.
[0050] Specifically, the historical test database contains n sets of data, each set of data consisting of reaction intensity value and corresponding biomarker concentration. Data is processed for missing values and outliers are removed to ensure data quality.
[0051] A linear regression equation is established using the least squares method: ; Wherein, represents the biomarker concentration, represents the reaction intensity function value, represents the intercept term, represents the regression coefficient, represents the random error. The least squares estimation formula is: ; ; Wherein, represents the mean value of the reaction intensity function, represents the biomarker concentration.
[0052] Calculate the residual error , draw a residual distribution chart to determine whether the residual error is randomly distributed to test the fitting applicability of the model. At a significance level , the confidence interval of the regression coefficient is calculated according to the standard error. The reliability of the regression model is judged by residual analysis and confidence interval results. If the model fitting degree is high and the parameters are significantly not zero within the confidence interval, the model can be used for subsequent trend prediction of physical changes and personalized report generation.
[0053] The linear regression model established by the above steps can establish a stable and interpretable mathematical relationship between the reaction intensity function and the biomarker concentration, effectively reducing the influence of single-point anomalies and measurement noise on concentration prediction.
[0054] S500: Collect user regional meteorological data using GPS positioning, and combine temperature, humidity and barometric pressure sensor data to compensate for the regression model; Further, the step of compensating the regression model is: Obtain the meteorological data of the user's region through GPS positioning, including temperature, humidity and barometric pressure; The temperature, humidity and air pressure data collected in real time by the environmental sensor are used to dynamically adjust the parameters of the regression model. According to the pre-established environmental compensation model, the influence of environmental factors on the concentration measurement of biomarkers is corrected.
[0055] Specifically, the geographical location of the user is obtained through the GPS module, and the real-time weather data of the area, including temperature T, relative humidity H and atmospheric pressure P, are further obtained from the networked weather service interface. At the same time, the temperature, humidity and air pressure data of the experimental site are collected in real time by the built-in environmental sensor as the basis for compensation.
[0056] According to a large number of pre-collected experimental data, a model of the influence of environmental factors on the concentration measurement of biomarkers is established by using multiple linear regression method. The environmental compensation model can be expressed as: ; Wherein, represents the concentration value calculated by the regression model, represents the concentration value after environmental compensation, , , respectively represent the standard environmental temperature, humidity and air pressure, , , respectively represent the compensation coefficients of temperature, humidity and air pressure on the concentration measurement, which are obtained by fitting experimental data. In actual application, the environmental data collected in real time are substituted into the above compensation formula, the biomarker concentration value output by the regression model is adjusted, the deviation caused by environmental variables is eliminated, and the accuracy and stability of the measurement result are improved.
[0057] S600: Compare the results with the feature database, generate a personalized detection report and send it to the user.
[0058] Further, the step of comparing with the feature database is: Compare the biomarker concentration obtained by mapping with the standard reference range in the feature database to determine whether the detection result is within the normal range; Retrieve the user's historical detection data and analyze the time variation trend and fluctuation pattern of the biomarker concentration; Match the corresponding personalized reference standard according to the user's personal information, and make horizontal comparison and analysis with the statistical data of similar people; Generate a personalized detection report containing concentration value, trend analysis, risk assessment and health suggestion based on the comparison result.
[0059] Specifically, the current detected biomarker concentration value is compared with the standard reference range of the corresponding marker in the feature database. The standard reference range is determined by authoritative medical data and clinical trial results, which is used to determine whether the test result is in the normal or abnormal interval, so as to preliminarily evaluate the user's health status.
[0060] The system automatically retrieves the user's historical detection records, extracts the concentration data within a period of time, analyzes the time series trend and fluctuation pattern, including the rising, falling or stable situation of the concentration, and identifies abnormal fluctuations or potential risks.
[0061] Combined with the user's personal information (such as age, gender, medical history, etc.), the most suitable personalized reference standard for the user is matched from the database. By comparing the user's data with the statistical characteristics of the same population, a more accurate health status evaluation is achieved.
[0062] According to the comparison and analysis results, the system automatically generates a detection report containing biomarker concentration values, time trend, risk assessment level and corresponding health suggestions. The report is presented in the form of charts and text, which is easy for users to understand and for medical professionals to refer to. The personalized detection report is sent to the mobile device or cloud account through the user-authorized channel, supporting the user to view at any time and long-term health management.
[0063] This embodiment ensures the scientific and reasonable interpretation of the test results, and realizes dynamic and personalized health monitoring and risk warning functions by combining individual differences of users.
[0064] Embodiment two: In the emergency medical scene, due to limited conditions, the traditional blood marker detection equipment is large in size, complex in operation and long in detection period, which is difficult to meet the rapid and portable detection requirements, resulting in technical bottlenecks in obtaining patient blood marker concentration information in time, affecting the efficiency of first aid decision-making.
[0065] To solve the above problems, a digital biomarker tracking method provided by the present application is adopted, and a schematic diagram thereof is shown in Figure 2 The specific implementation process of the method is as follows: The user at the scene scans the two-dimensional code on the reagent kit using a mobile device, and the two-dimensional code encodes product data including reagent variety, batch number, production information and channel information. After analyzing the two-dimensional code content, the system stores it in association with the subsequent collected detection data, realizes accurate binding of the detection result and the product source. The user can select or take a reaction image of the biomarker combined with the reagent on the system page, and upload the image file to the remote server for subsequent data processing and analysis.
[0066] The server adopts a statistical method to detect and remove outliers of received image data, removes abnormal pixels or feature points; then uses a least square method to preliminarily correct the remaining data, and completes secondary correction through image processing algorithms such as geometric distortion correction, image registration, Gaussian filter denoising and motion blur repair, to ensure that the image quality meets the requirements of subsequent analysis.
[0067] The reaction area is extracted by an image segmentation algorithm (such as threshold segmentation or a deep learning model), and irrelevant backgrounds are excluded; the signal of the extracted area is numerically processed, including subtracting the negative group signal from the experimental group and the positive group signal to eliminate the chemical reaction background noise, and then calculating the signal ratio of the experimental group and the positive group to correct the reagent attenuation and environmental influence.
[0068] Color features, light intensity features, turbidity features and texture features are extracted from the preprocessed reaction area image, and an infinite series expansion method is used to establish a nonlinear mapping of the image features to the biomarker reaction intensity function.
[0069] Based on the reaction intensity function obtained by mapping, a linear regression model is established by using the least square method to fit the biomarker concentration signature curve combined with the historical test database.
[0070] Through GPS positioning, the meteorological data of the area where the user is located is obtained, and the temperature, humidity and air pressure monitored by the built-in sensor in real time are combined to dynamically compensate the parameters of the regression model, eliminating the influence of environmental factors on the detection results.
[0071] The compensated concentration data are compared with the standard reference range in the feature database, combined with the user's historical detection data and personalized information, the trend of the signs and the risk level are analyzed, a personalized detection report containing concentration values, trend analysis, risk assessment and health suggestions is generated, and is sent to the user's mobile terminal in real time to assist emergency decision-making.
[0072] Through the embodiment, the digital biomarker tracking method realizes rapid, portable and efficient blood marker detection in emergency medical field, significantly improves the timeliness and accuracy of detection, and overcomes the defects of large size and complex operation of traditional equipment, providing strong data support for emergency medical treatment.
[0073] Finally, it should be noted that: the above only describes the preferred embodiments of the present application and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A method of digital biomarker tracking, the method comprising: The method comprises the following steps: Obtain the reaction image of the biomarker combined with the reagent, and transmit the image to a server; Screen and remove outliers in the image by statistical methods, use the least square method to preliminarily correct the retained image data, and then perform secondary correction through an image algorithm; Extract the region of interest through an image segmentation algorithm, remove the background region irrelevant to the reaction, and extract the color, light intensity, turbidity and texture features of the preprocessed image; Based on the compensated image features, a non-linear reaction intensity function between the image features and the concentration of the biomarker is established by using infinite series expansion, and a linear regression model is established by using the function to fit the curve of the concentration of the biomarker changing with time.
2. The method of claim 1, wherein, A two-dimensional code is arranged on the detection card corresponding to the reaction image, and the two-dimensional code encodes product information including but not limited to reagent variety, batch number, production information and channel information; in the image uploading or processing process, the content of the two-dimensional code is parsed and associated with the detection data to realize the corresponding tracking of the detection result and the product source.
3. The method of claim 1, wherein, The steps of screening and removing outliers in the image are as follows: Abnormal value detection is performed on the image pixel or feature data by using statistical methods; Outliers deviating from the normal range are identified by setting a threshold or based on distribution characteristics; The identified outliers are removed or replaced with adjacent normal data.
4. The method of claim 1, wherein, The steps of preliminary correction and secondary correction are as follows: The image data after removing outliers is preliminarily corrected by using the least square method to fit and correct the basic color or light intensity parameters of the image; Geometric distortion correction is performed to correct the deformation of the image caused by the shooting device or environment; Image registration technology is used to accurately align multiple images to ensure consistency in subsequent analysis; Gaussian filter denoising algorithm is applied to preliminarily remove noise in the image; Motion blur repair is implemented to improve the blur caused by motion through image restoration algorithm.
5. The method of claim 1, wherein, The steps of removing the background region irrelevant to the reaction are as follows: An image segmentation algorithm is used to analyze the preprocessed image, and the image segmentation algorithm includes but is not limited to threshold segmentation, edge detection or deep learning model; The boundary of the reaction region is determined, the region of interest is extracted, and the image background and irrelevant regions are excluded; Morphological processing such as erosion and dilation operation is performed on the extracted region of interest to further refine the region boundary; And the detection signal corresponding to the extracted region is numerically processed, including subtracting the signal value of the negative group from the signal values of the experimental group and the positive group to eliminate the background noise of the chemical reaction, and then calculating the ratio of the signal values of the experimental group and the positive group to correct the influence of reagent attenuation.
6. The method of claim 1, wherein, The steps of feature extraction on the received reaction image are as follows: Color features are obtained through HSV space histogram statistics and dominant color clustering; Light intensity features are obtained through regional gray mean value, variance calculation and Gamma correction compensation; Turbidity features are obtained through scattered light spot area measurement and edge sharpness analysis; Texture features are obtained through energy value and entropy value calculation of the gray level co-occurrence matrix.
7. The method of claim 1, wherein, The steps of establishing a non-linear reaction intensity function by using infinite technology are as follows: The image feature vector is used as input, and a multivariate orthogonal polynomial basis function is used for infinite series expansion; The number of expansion terms is determined through small amount control before convergence and basis function expansion; The least square method is used to solve the series expansion coefficients, and a nonlinear response intensity function of the image features to the biomarker concentration is established.
8. The method of claim 1, wherein, The steps of establishing the linear regression model are: Taking the biomarker concentration as the dependent variable and the response intensity function as the independent variable, a linear regression equation is established by using the historical test database; The fitting effect and reliability are verified by residual analysis and confidence interval calculation.
9. The method of claim 1, wherein, The method further comprises: collecting the user's regional meteorological data by GPS positioning, and combining the temperature, humidity and air pressure sensor data to perform environmental compensation on the regression model; the steps of environmental compensation are: Obtaining the meteorological data of the user's region by GPS positioning, including temperature, humidity and air pressure; Using the temperature, humidity and air pressure data collected by the environmental sensor in real time to dynamically adjust the parameters of the regression model; According to the pre-established environmental compensation model, the influence of environmental factors on the measurement of biomarker concentration is corrected.
10. The method of claim 1, wherein, The method further comprises: comparing the results with the feature database to generate a personalized detection report and send it to the user; the steps of comparing with the feature database are: Comparing the mapped biomarker concentration with the standard reference range in the feature database to determine whether the detection result is within the normal interval; Retrieving the user's historical detection data to analyze the time variation trend and fluctuation pattern of the biomarker concentration; Matching the corresponding personalized reference standard according to the user's personal information, and performing horizontal comparison analysis with the statistical data of the same population; Based on the comparison result, a personalized detection report containing concentration value, trend analysis, risk assessment and health suggestion is generated.