A multi-antigen detection test paper result interpretation method and system based on artificial intelligence visual recognition

Abstract

This invention discloses a method and system for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition, belonging to the field of artificial intelligence detection technology. The method includes acquiring continuous images of radial multi-antigen test strips, performing quality assessment on the images, and selecting target analysis images; locating the center point of the test strip card based on the target analysis images, identifying the direction and number of each detection arm, establishing a polar coordinate system, and performing geometric correction on the images according to the polar coordinate system; segmenting the fan-shaped regions of each detection arm according to the polar coordinate system, locating the control line and test line in each detection arm, and obtaining a set of regions of interest for the strips; extracting the visual features of the set of regions of interest for the strips, and normalizing the test line features based on the corresponding features of the control line in the same detection arm. This invention constructs a polar coordinate correction and same-arm normalization mechanism, integrating artificial intelligence interpretation and a time-series stabilization mechanism to achieve high-precision and interpretable intelligent interpretation of radial multi-antigen test strips.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence detection technology, and in particular to a method and system for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition. Background Technology

[0002] In recent years, lateral tomography immunoassay technology has been widely used in infectious disease detection, food safety monitoring, and chronic disease management due to its advantages such as requiring no complex equipment, fast detection speed, and suitability for rapid on-site screening. With the increasing demand for multi-indicator joint detection, traditional single-detection-line structures are gradually evolving towards multi-antigen, multi-channel detection. Among these, radial multi-arm structure test strips have attracted attention because they can integrate multiple detection units within a limited space. The continuous improvement of image acquisition capabilities and computing performance of intelligent terminals has driven the rapid development of computer vision-based automatic interpretation technology. Related research is gradually shifting from traditional methods based on threshold segmentation and grayscale statistics to intelligent interpretation that integrates feature engineering and deep learning models. However, because radial structures break through the geometric assumptions of traditional linear or regular arrangements, the irregular circular layout makes it difficult to directly transfer existing visual analysis frameworks based on rectangular regions of interest or fixed template matching. Fundamental technological gaps exist in key aspects such as robustness of center positioning, adaptive recognition of detection arm directions, and comparability of cross-arm features, severely restricting the high-precision intelligent interpretation of multi-antigen test strips in real-world complex scenarios.

[0003] Existing test strip interpretation technologies generally suffer from the following limitations: First, they lack the ability to analyze multi-arm structures in images. Current methods often rely on fixed templates or simple geometric assumptions, making it difficult to adaptively model the directional distribution, scale variations, and imaging distortion of radial multi-arm detection structures. This results in large band localization deviations and high segmentation failure rates. Second, they lack the ability to uniformly model multiple antigen features. Existing technologies typically process each detection line independently, lacking a cross-channel normalization mechanism based on quality control lines and a multi-antigen joint analysis strategy. This makes it difficult to effectively eliminate errors between different detection arms, affecting overall interpretation consistency and result interpretability. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition, which solves the problems of insufficient image analysis capability of multi-arm structures and insufficient unified modeling capability of multi-antigen features.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for interpreting the results of a multi-antigen test strip based on artificial intelligence visual recognition, comprising, Acquire continuous images of the radial multiantigen test strip, and perform image quality assessment and screen target analysis images; Based on the target analysis image, locate the center point of the test strip card, identify the direction and number of each detection arm, establish a polar coordinate system, and perform geometric correction on the image according to the polar coordinate system; The fan-shaped regions of each detection arm are divided according to the polar coordinate system, and the quality control line and detection line in each detection arm are located to obtain the set of regions of interest of the strips; Visual features of the region of interest set of the strips are extracted, and the features of the detection line are normalized based on the features corresponding to the quality control line in the same detection arm to obtain a normalized multi-antigen feature matrix. The normalized multi-antigen feature matrix is ​​input into the pre-trained artificial intelligence model, and the effectiveness is judged by combining the corresponding features of the quality control line. The qualitative interpretation results and confidence levels of each antigen are output. Based on the qualitative interpretation results of each antigen, a combined feature vector is constructed, and the combined feature vector is input into a decision tree model to output risk level, probability score and explanatory information. A time series is constructed based on multiple interpretation results and dynamically updated. The final interpretation result is output when the preset stability condition is met.

[0007] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the method for acquiring continuous images of the radial multi-antigen test strip includes: The test strip is photographed by a camera device on a smart terminal placed on a flat background, and multiple frames of images are continuously acquired based on the sampling frequency to form a time series.

[0008] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the step of quality assessment and screening of target analysis images includes: Sharpness, brightness, contrast, and framing integrity quality evaluation parameters are extracted from each image frame in the image sequence, and a quality evaluation value is obtained through weighted fusion. The quality evaluation value is compared with a preset threshold to filter candidate processing objects. After enhancing the candidate processing objects, the comprehensive quality score is recalculated and sorted. The image with the highest score is selected as the target analysis image.

[0009] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the step of locating the center point of the test strip card based on the target analysis image, identifying the direction and number of each detection arm, establishing a polar coordinate system, and performing geometric correction on the image according to the polar coordinate system includes: Based on the target analysis image, the Hough circle transform is used to detect near-circular structures and determine the center point and radius of the central disk of the test strip card. Radial sampling is performed within a preset angle range with the center point as the origin. A radial response function is constructed by integrating the image intensity in each direction. Then, the direction of each detection arm is identified and the number of detection arms is determined by peak detection. A polar coordinate system is established with the center point as the origin of the polar coordinate system and the direction of the first detection arm as the angular reference. Based on the deviation and scale relationship between the direction of the detection arm and the standard template, the image is sequentially rotated, scaled, and perspective corrected to obtain the geometrically corrected image.

[0010] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the step of dividing the fan-shaped regions of each detection arm according to the polar coordinate system, locating the control line and the detection line in each detection arm, and obtaining the set of regions of interest for the bands includes: Based on the polar coordinate system and the direction information of the detection arm, the geometric correction image is segmented into a sector region to obtain the sector image region corresponding to each detection arm; The positions of the quality control line and the detection line in each detection arm are located by combining radial grayscale projection with peak detection and a preset strip position template. Further extract the corresponding regions to form the regions of interest for quality control lines and detection lines, and summarize them to obtain the set of regions of interest for the strips.

[0011] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the method involves: extracting visual features of the set of regions of interest in the bands, and normalizing the test line features based on the corresponding features of the control lines in the same test arm to obtain a normalized multi-antigen feature matrix, including: Visual features are extracted from the region of interest set of the strips; and the visual features of the control lines in the same detection arm are used as a benchmark to normalize the detection line features and construct a normalized multi-antigen feature matrix.

[0012] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the step of inputting the normalized multi-antigen feature matrix into a pre-trained artificial intelligence model, and combining it with the corresponding features of the quality control line to perform validity judgment, and outputting the qualitative interpretation results and confidence levels of each antigen includes: The feature vectors of each normalized detection line in the normalized multi-antigen feature matrix are input into the pre-trained artificial intelligence model to obtain the response value corresponding to each antigen. The validity of the detection arm is determined based on the quality control line characteristics. Only for the detection arm that is determined to be valid, the qualitative interpretation result of each antigen is output according to the comparison result of the model response value and the corresponding judgment threshold. Simultaneously, the confidence level of the interpretation is calculated based on the relative relationship between the model response value and the judgment threshold, and a set of multi-antigen interpretation results is constructed.

[0013] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the method involves constructing a combined feature vector based on the qualitative interpretation results of each antigen, inputting the combined feature vector into a decision tree model, and outputting risk level, probability score, and explanatory information, including: Based on the set of multi-antigen interpretation results, the qualitative results of each antigen are extracted and a combined feature vector is constructed. The combined feature vector is then input into a pre-trained decision tree model to obtain the risk level and probability score. Explanatory information is generated based on the classification path of the decision tree model, thus forming the results of a single multi-antigen combined analysis.

[0014] As a preferred embodiment of the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition described in this invention, the step of constructing a time series based on multiple interpretation results and dynamically updating it, and outputting the final interpretation result when a preset stability condition is met includes: Within the reaction time window of the test strip, a time series of multiple interpretation results is constructed and updated, and the change in interpretation results between adjacent time points is calculated; The system determines stability based on the change in response value and outputs the final result when the preset stability conditions are met.

[0015] Secondly, this invention provides a multi-antigen test strip result interpretation system based on artificial intelligence visual recognition, comprising, The image acquisition and quality screening module is used to continuously capture test strip images via a mobile device and screen target analysis images based on multidimensional quality evaluation parameters. The geometric correction and region segmentation module is used to locate the center of the test strip and the direction of the detection arm, establish a polar coordinate system and complete geometric correction, and then segment the control line and detection line regions of each detection arm. The feature extraction and normalization module is used to extract the grayscale, color, width and edge features of the strip region, and perform normalization processing based on the same arm quality control line to obtain the multi-antigen feature matrix; The AI ​​interpretation and validity verification module is used to input normalized features into a pre-trained AI model for interpretation, combine the strength of the quality control line to judge the detection validity, and output the qualitative results and confidence level of each antigen. The risk assessment and dynamic decision-making module is used to perform joint analysis based on the results of multi-antigen interpretation, output risk level, probability score and explanatory information, and achieve stable interpretation by combining time series.

[0016] The beneficial effects of this invention are as follows: By constructing a polar coordinate system and implementing triple geometric correction, the center positioning, arm orientation recognition, and strip segmentation of radial test strips are achieved, significantly improving the robustness of multi-arm structure image analysis. A normalization mechanism based on the same-arm quality control line is introduced to construct a normalized multi-antigen feature matrix, eliminating errors between different detection arms and achieving consistent interpretation of cross-channel features. Furthermore, by combining artificial intelligence model interpretation, quality control validity verification, multi-antigen combination decision-making, and a time-series dynamic stabilization mechanism, not only are high-confidence qualitative results output, but also risk level, probability score, and interpretability information are provided. In complex home environments, multi-frame time-series convergence ensures the reliability of the final interpretation, comprehensively improving the intelligence level and clinical applicability of simultaneous multi-indicator detection. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for interpreting the results of a multi-antigen test strip based on artificial intelligence visual recognition, as described in Example 1.

[0019] Figure 2 This is a structural diagram of a multi-antigen test strip result interpretation system based on artificial intelligence visual recognition in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figures 1-2This is the first embodiment of the present invention, which provides a method for interpreting the results of a multi-antigen test strip based on artificial intelligence visual recognition, including the following steps: S1. Acquire continuous images of the radial multiantigen test strip, evaluate the quality of the images, and select target images for analysis.

[0024] S1.1: The test strip placed on a flat background is photographed by the camera device of the smart terminal, and multiple frames of images are continuously acquired based on the sampling frequency to form a time series.

[0025] Specifically, use a smartphone, tablet, or dedicated testing device to take a picture of the radial multi-antigen test strip using its built-in camera. Place the radial multiantigen test strip on a flat background surface and adjust the device position so that the entire test strip is within the field of view of the camera device; Control the camera device at the sampling frequency Multiple frames of images are acquired continuously to form a time-continuous image sequence. :

[0026] in, Indicates at time The captured image frames, This indicates the total number of frames captured.

[0027] S1.2: Extract the sharpness, brightness, contrast and framing integrity quality evaluation parameters from each image frame in the image sequence, and obtain the quality evaluation value through weighted fusion.

[0028] Specifically, for each image frame in the image sequence Extract the quality evaluation parameters separately; The quality evaluation parameters include sharpness parameters. Brightness parameters Contrast parameters and framing integrity parameters ; The various quality evaluation parameters are weighted and fused to obtain the image frame. Quality evaluation value :

[0029] in, These are the weighting coefficients for the corresponding quality evaluation parameters; It should be noted that, The value range of is [0,1], and satisfies The optimal value can be determined through existing technologies such as grid search, cross-validation, or analytic hierarchy process based on expert experience. The specific process involved recruiting 200 ordinary users to collect approximately 60,000 real test strip images in their home environments. Using professional physician interpretation as the gold standard, end-to-end accuracy was evaluated for different combinations of quality evaluation weights. Experimental results showed that when the weights of the four indicators—clarity, brightness, contrast, and image completeness—were respectively set to... At the same time, it can optimize the accuracy of artificial intelligence interpretation, which is significantly better than other configurations, and therefore it has been identified as the preferred implementation parameter.

[0030] S1.3: Compare the quality evaluation value with the preset threshold to filter candidate processing objects, perform enhancement processing on the candidate processing objects, recalculate the comprehensive quality score and sort them, and select the image with the highest score as the target analysis image.

[0031] Specifically, the quality evaluation value of each image frame is... Compared with the preset quality threshold Comparison: When satisfied When, the image frame is determined. Low-quality image frames are discarded. When satisfied When this happens, the image frame is retained. As a candidate for processing; It should be noted that the quality threshold The preset threshold, ranging from 0.55 to 0.70, can be determined using existing image screening techniques that employ image quality-interpretation performance correlation analysis and end-to-end accuracy verification. Specifically, at least 200 ordinary users are recruited in a typical home environment, and continuous image sequences of the radial multi-antigen test strip are collected using mainstream smartphones. A normalized quality evaluation value is calculated for each frame of the image. The consistency between the corresponding AI interpretation results and the clinical gold standard was used as a performance indicator, and different statistical analyses were conducted. The accuracy of image interpretation and the success rate of single-shot capture are considered. Based on a comprehensive trade-off between interpretation performance and efficiency, a threshold range is selected that ensures high interpretation accuracy while also considering image throughput. Furthermore, through a trade-off analysis of false rejection rate and false use rate, the optimal operating point is determined, with the preferred value being [value missing]. The optimal threshold was determined through user image data modeling, effective interpretation event statistics, and triple verification of accuracy and usability under real home multi-antigen testing scenarios. This ensures that images that can be reliably retained for high-precision AI interpretation are retained, while effectively eliminating low-quality interference frames caused by blurring, uneven lighting, or occlusion. Image enhancement processing is performed on the candidate object, and the enhanced image frame can be represented as follows:

[0032] in, This indicates the original image frame at the pixel level. Pixel value at that location, This represents the enhanced pixel value. Indicates the gain coefficient. Indicates the bias coefficient; It should be noted that the gain coefficient and bias coefficient These are preset parameters, with value ranges of [missing information]. , This can be determined using existing image enhancement techniques with light-adaptive calibration and contrast optimization. Specifically, at least 100 sets of images of radial multi-antigen test strips under different lighting conditions in a typical home environment, including strong sidelight, weak backlight, direct desk lamp light, and natural diffused light, are collected. For each set of images, manually annotate whether the control lines and test lines in the interpretable area are clearly visible. This process is repeated for each set of images. and The combination of parameters is used to perform linear enhancement on the original image, and then input into a subsequent artificial intelligence interpretation model to evaluate the accuracy of stripe recognition. The parameter combination that can effectively enhance the contrast of the stripe region and meet the interpretation requirements is selected, with the preferred value being... , The optimal parameters were determined through triple verification in real home testing scenarios, including multi-light condition modeling, effective color signal enhancement verification, and over-enhancement distortion suppression. This ensures that the visibility of weak color bands can be effectively improved while avoiding saturation of bright areas or amplification of background noise, thereby guaranteeing the robustness and accuracy of multi-antigen result interpretation. The overall quality score is recalculated for the enhanced images corresponding to the candidate processing objects, and the images are sorted from high to low according to the overall quality score. The image with the highest score is selected as the target analysis image. :

[0033] in, This refers to the overall quality score corresponding to the enhanced image frame; The comprehensive quality score corresponding to the enhanced image frame Use the same quality evaluation value as the original image frame. The same evaluation parameter system and weighting method were used for calculation.

[0034] S2. Based on the target analysis image, locate the center point of the test strip card, identify the direction and number of each detection arm, establish a polar coordinate system, and perform geometric correction on the image according to the polar coordinate system.

[0035] S2.1: Based on the target analysis image, the Hough circle transform is used to detect near-circular structures and determine the center point of the test strip card and the radius of the central disk.

[0036] Specifically, the Hough circle transform method is used to analyze the target image. A search is performed on near-circular structures to obtain a set of candidate circular parameters. :

[0037] in, Indicates the coordinates of the candidate circle center. Indicates the radius of the candidate circle; Based on the positional and dimensional relationships between the candidate circles and the main outline of the test strip, the optimal candidate circle is selected, and its center is determined as the center point of the test strip. :

[0038] in, Indicate the coordinates of the center point of the test strip card; Meanwhile, the radius of the optimal candidate circle is denoted as the radius of the central disk. .

[0039] S2.2: Radial sampling is performed within a preset angle range with the center point as the origin. A radial response function is constructed by integrating the image intensity in each direction. Then, the direction of each detection arm is identified and the number of detection arms is determined by peak detection.

[0040] Specifically, with the center point With the origin as the reference point, within the angle range Radial sampling is performed with a fixed step size for each angle. , along the radial interval Integrating the image intensity yields the radial response value. :

[0041] in, The starting radius of the radial integral. The terminator radius of the radial integral. Refers to radial variable, Point along the angle Direction, distance from center point for sampling points coordinate, Point along the angle Direction, distance from center point for sampling points coordinate, This refers to performing a line integral of the image intensity function along the ray path. The length of an infinitesimal line segment along the ray direction; After traversing all angles, the complete radial response function is obtained. :

[0042] radial response function Peak detection is performed, and the local extremum analysis method of one-dimensional signal is used to screen out the local maxima points that are relatively separated and representative in the angular space. The angle corresponding to the peak value that meets the conditions is determined as the detection arm direction angle, forming a set of detection arm directions. :

[0043] in, Indicates the first Detection arm direction, , Number of detection arms; Arrange them in ascending order of angle size to obtain an ordered set of direction angles. :

[0044] in, Indicates the direction of the first detection arm. Indicates the direction of the second detection arm. Refers to the first Detection arm direction.

[0045] S2.3: Establish a polar coordinate system with the center point as the origin of the polar coordinate system and the direction of the first detection arm as the angular reference. Based on the deviation and scale relationship between the direction of the detection arm and the standard template, perform rotation correction, scale correction and perspective correction on the image in sequence to obtain the geometrically corrected image.

[0046] Specifically, with the center point With the origin of the polar coordinates as the reference point, and the direction of the first detection arm as the reference point... Construct a polar coordinate system using the zero point of angle. ; For any pixel in the image Polar coordinates are represented as:

[0047]

[0048] Geometric correction of an image using polar coordinates includes the following steps: Rotational correction: Set the reference detection arm direction in the standard template as... Based on the actual detected direction of the first detection arm Calculate the rotation correction angle :

[0049] With the center point Using the rotation center, perform rotation transformation on the image; Scale correction: based on the radius of the central disk With standard template radius Calculate the scaling transformation coefficients :

[0050] The rotated image is then isotropically scaled. It should be noted that, The typical value range is 0.8 to 1.2, which can be determined through scale robustness modeling and end-to-end interpretation performance verification techniques based on real user images. The specific process is as follows: recruit no fewer than 200 ordinary users in a typical home environment, and use mainstream smartphones to collect continuous image sequences of the radial multi-antigen test strip. For each frame of the image, calculate the radial distance of each test line region with the center of the test strip as the reference. and divided by the standard radius specified in the laboratory. get Then, using the consistency between the AI ​​interpretation results and the clinical gold standard as a performance indicator, statistical analysis was conducted on different... The accuracy and positional error of strip recognition under offset are analyzed, and a scale transformation coefficient is selected that can ensure the positioning accuracy of the strip area and maintain stable interpretation performance. The tolerance range was determined by combining the scale distribution histogram caused by changes in the phone's focal length and shooting height during actual shooting. Finally, the optimal working range was determined through a trade-off analysis of false rejection rate and false negative rate. The preferred range is [0.85~1.15]. This was verified by user data modeling, effective event statistics and accuracy-availability dual constraints in real home multi-antigen testing scenarios to ensure robust adaptability to common shooting distance changes. Perspective correction: The boundary points of the central disk and the starting and ending points of each detection arm are selected as control points, and the perspective transformation matrix is ​​solved using a direct linear transformation algorithm. And transform the scale-corrected image:

[0051] in, The corrected pixel coordinates; Complete the analysis of the target image After rotation, scaling, and perspective triple geometric correction, a geometrically corrected image is generated and output. .

[0052] S3. Divide the sector region of each detection arm according to the polar coordinate system, locate the quality control line and detection line in each detection arm, and obtain the set of regions of interest of the strip.

[0053] S3.1: Based on the polar coordinate system and the direction information of the detection arm, the geometric correction image is segmented into a sector region to obtain the sector image region corresponding to each detection arm.

[0054] Specifically, based on geometrically corrected images Number of detection arms and the sorted set of detection arm orientation angles Determine the angle boundary range corresponding to each detection arm; Based on the angle boundary intervals corresponding to each detection arm, in the polar coordinate system Below the geometrically corrected image Perform sector-shaped region segmentation to obtain the sector-shaped image region corresponding to each detection arm. :

[0055] in, and These represent the inner and outer diameter boundaries of the detection arm, respectively. This is a preset constant, calibrated by the physical dimensions of the test strip card. and They represent the first The starting and ending boundaries of the angles of each detection arm. It is determined by the midpoint of the direction angle of adjacent detection arms.

[0056] S3.2: The positions of the quality control line and the detection line in each detection arm are located by combining radial grayscale projection with peak detection and a preset strip position template.

[0057] Specifically, for each sector region Perform grayscale projection along the radial direction to obtain the first... Radial response curves of each detection arm :

[0058] in, The geometrically corrected image is in polar coordinates with a radius of... And the angle is The pixel intensity value of the sampling point, Refers to the geometrically corrected image at a fixed radius Place, along the first The angle range corresponding to each detection arm Perform line integration. The infinitesimal angular increment in the direction of the angle; According to the radial response curve The local peak positions in the data, combined with the preset strip position template, are used to determine the first... The position of the quality control line in each testing arm and the position of the detection line ,in, , For the first The number of detection lines in each detection arm; The preset strip position template refers to a geometric model or set of parameters that is predefined in the AI ​​image analysis of radial or linear immunochromatographic test strips and is used to locate the theoretical positions of each functional strip in the image coordinate system.

[0059] S3.3: Further extract the regions of interest corresponding to the quality control line and the detection line, and summarize them to obtain the set of regions of interest for the strips.

[0060] Specifically, based on the position of the quality control line Centered on a target area, the corresponding image region is cropped according to preset radial and angular widths to obtain the region of interest for the quality control line. :

[0061] in, Radial coordinates in polar coordinates. Angular coordinates in polar coordinate system The radial half-width parameter is obtained through test paper structure calibration. Based on the position of the detection line Centered on a target area, the corresponding image region is cropped according to a preset radial and angular width to obtain the region of interest for the detection line. :

[0062] in, Refers to the testing line number; By combining the regions of interest (ROIs) of the control lines and the detection lines in all detection arms, a set of ROIs for the strips is obtained. : .

[0063] S4. Extract the visual features of the region of interest set of the strips, and normalize the features of the detection line based on the features corresponding to the quality control line in the same detection arm to obtain the normalized multi-antigen feature matrix.

[0064] S4.1: Extract visual features from the set of regions of interest of the strips.

[0065] Specifically, the region of interest of the strip is set. Input the feature extraction module to the regions of interest for each quality control line. and the region of interest of the detection line Extract visual features separately; The visual features include grayscale mean features, color intensity features, stripe width features, and edge gradient features; For the The first of the detection arms Visual features of a detection line Represented as:

[0066] in, Refers to the grayscale mean characteristic. Refers to color intensity characteristics, The characteristics of the strip width. Refers to edge gradient features; For the Visual features of control lines in a single inspection arm Represented as: .

[0067] S4.2: The detection line features are normalized based on the visual features of the control lines in the same detection arm to construct a normalized multi-antigen feature matrix.

[0068] Specifically, visual characteristics of control lines within the same inspection arm. Based on this, visual features of each detection line within the detection arm are analyzed. Normalization is performed to obtain normalized detection line features. :

[0069] in, This refers to preventing tiny positive numbers with a denominator of zero. ; In component form, normalized detection line features Represented as:

[0070] The normalized detection line features in all detection arms are arranged according to the detection arm number and the detection line number to construct a normalized multi-antigen feature matrix. : .

[0071] S5. Input the normalized multi-antigen feature matrix into the pre-trained artificial intelligence model, and combine it with the corresponding features of the quality control line to make an effectiveness judgment, and output the qualitative interpretation results and confidence level of each antigen.

[0072] S5.1: Input the feature vectors of each normalized detection line in the normalized multi-antigen feature matrix into the pre-trained artificial intelligence model to obtain the response value corresponding to each antigen.

[0073] Specifically, the normalized multi-antigen feature matrix The normalized detection line feature vectors in Each antigen is input into a pre-trained artificial intelligence model to obtain its corresponding response value. :

[0074] in, Pre-trained artificial intelligence models; The pre-trained artificial intelligence model includes an input layer, a fusion layer, and an output layer; The input layer is used to receive the normalized detection line feature vector. The input dimension is 4×1, and the input features are subjected to linear projection and standardization. The fusion layer consists of one or more fully connected hidden layers, each containing linear transformation, batch normalization, and the nonlinear activation function ReLU, used to deeply mine the high-order interaction relationship between the four dimensions of grayscale mean, color intensity, strip width, and edge gradient, and to construct a nonlinear response mapping sensitive to antigen concentration. The output layer is either a single-neuron fully connected layer with no activation function or a Sigmoid activation function, outputting a scalar value. ; Model training steps: Construct a dataset of test strip images containing a large number of known antigen concentrations or positive / negative labels; extract the four-dimensional visual features of each test line and normalize them with the control lines on the same arm to form input-label pairs. The sample is input into a fully connected network, and the predicted response value is calculated through forward propagation. The mean squared error or Huber loss function is used to measure the deviation between the predicted value and the true label. The network parameters are updated by backpropagation using the Adam optimizer. After training, a pre-trained artificial intelligence model for antigen response value prediction is obtained.

[0075] S5.2: Based on the quality control line characteristics, the validity of the detection arm is determined. Only for the detection arm determined to be valid, the qualitative interpretation result of each antigen is output according to the comparison result of the model response value and the corresponding judgment threshold.

[0076] Specifically, based on the visual features of the quality control line For the The effectiveness of each detection arm is judged: When the quality control line characteristics meet the preset validity conditions, the first... Each detection arm is effective; When the quality control line characteristics do not meet the preset validity conditions, the first... One detection arm failed to detect; The validity condition of the quality control line can be expressed as:

[0077] refer to , refer to ; when The reading result from the detection arm is invalid and will not be included in the final output. It should be noted that the quality control line validity threshold The preset threshold value is typically a range of positive real numbers in the image grayscale or normalized intensity domain. It can be determined through experimental statistical analysis of valid and invalid test strip samples. Specifically, under various typical detection conditions, known valid negative / positive test strip image samples are collected, and simultaneously, invalid test strip image samples (those without sample, those that have dried out, and those with abnormal color development) are collected. The average signal intensity of the control line region in both types of samples is then calculated. Distribution, on the effective test strip High-density zone and invalid test strips Between low-density intervals, the boundary point that maximizes the effective read rate and minimizes the false negative rate is selected as... Preferred The determination was made through experimental statistical analysis of typical valid and invalid test strip images: A large number of images of valid and various invalid test strips were collected in actual home and clinical application scenarios, and the average intensity distribution of the control line area was calculated. The results showed that the valid test strips… The concentration is between 80 and 200, while invalid test strips are mostly below 50. It can achieve a reasonable balance between the quality control detection rate and the false judgment rate, and is the optimal value that achieves the best balance between high reliability and strong robustness; For validity markers The detection arm, based on the model response value With the corresponding antigen determination threshold The comparison results will output the qualitative interpretation results of each antigen. :

[0078] in, This indicates that the corresponding antigen is positive. This indicates that the corresponding antigen is negative; It should be noted that the threshold The preset threshold, with a value range of positive real numbers in the model's response range, can be determined through experimental statistical analysis of a large number of known positive and negative samples. Specifically, the method involves acquiring test strip images of known positive and negative samples, extracting and normalizing features, and then inputting them into the pre-trained model to obtain the response values ​​corresponding to each antigen. The distribution is determined by selecting the boundary point that maximizes the accuracy of interpretation and minimizes the false positive and false negative rates between the high-density interval of positive samples and the low-density interval of negative samples. Preferred Based on experimental statistics, in both home and clinical settings, the effective positive response to a certain antigen is concentrated in [specific locations]. The negative samples were mostly below 0.30. This can keep the risks of false positives and false negatives within a controllable range, achieving the best balance between high reliability and strong robustness.

[0079] S5.3: Simultaneously calculate the interpretation confidence based on the relative relationship between the model response value and the judgment threshold, and construct a set of multi-antigen interpretation results.

[0080] Specifically, it outputs response values ​​based on an artificial intelligence model. With the judgment threshold The relative distance mapping yields the interpretation confidence level for each antigen. Among them, confidence level The greater the distance, the higher the confidence level. Combining the antigen qualitative interpretation results from all valid detection arms and corresponding confidence level A set of multi-antigen interpretation results was obtained. : .

[0081] S6. Construct a combined feature vector based on the qualitative interpretation results of each antigen, and input the combined feature vector into the decision tree model to output the risk level, probability score and explanatory information.

[0082] S6.1: Extract the qualitative results of each antigen based on the set of multi-antigen interpretation results and construct a combined feature vector. Input the combined feature vector into a pre-trained decision tree model to obtain the risk level and probability score.

[0083] Specifically, from the set of multi-antigen interpretation results Qualitative interpretation results of each antigen extracted They are arranged according to a preset antigen order to construct a combined feature vector. :

[0084] in, Refers to all antigens involved in the joint analysis; Combine feature vectors Input a pre-trained decision tree model to obtain the corresponding risk level. and probability score :

[0085] in, Refers to the decision tree model.

[0086] S6.2: Generate explanatory information based on the classification path of the decision tree model, and then form the results of a single multi-antigen combined analysis.

[0087] Specifically, based on the combined feature vectors in the decision tree model The classification path, extraction and risk level Generate explanation information based on the corresponding node rule information. ; Among them, explanatory information This is used to describe the positive antigen combination relationships that lead to the current risk level determination, key determination points, and corresponding risk warnings; Risk level Probability scoring and explanatory information With the corresponding set of multi-antigen interpretation results Together they constitute the results of a single multi-antigen combined analysis. .

[0088] S7. Construct a time series based on multiple interpretation results and update it dynamically. Output the final interpretation result when the preset stability condition is met.

[0089] S7.1: Construct and update a time series of multiple interpretation results within the test strip reaction time window, and calculate the change in interpretation results between adjacent time points.

[0090] Specifically, within the reaction time window of the test strip, the process of image quality assessment, target analysis image screening, geometric correction, band localization, feature extraction, artificial intelligence interpretation, and joint analysis is repeatedly executed to obtain multiple multi-antigen joint analysis results at different sampling times. ; The results of multi-antigen combined analysis at each sampling time Sort by time and construct a time series. :

[0091] in, , and This indicates the sampling times within the reaction time window; Using a sliding window method for time series Perform dynamic updates and calculate the change in interpretation results between adjacent time points. .

[0092] S7.2: Based on the change in response value, a stability determination is made, and the final judgment result is output when the preset stability conditions are met.

[0093] Specifically, when the model response value is used as the criterion for stability determination, the change is defined as:

[0094] in, Indicates the first An antigen at time The model response value; When satisfied When this condition is met, the current interpretation result is determined to satisfy a preset stability condition, where, This indicates the preset convergence threshold; It should be noted that the threshold The preset threshold, with a value range of positive real numbers within the model response range, can be determined through time-series modeling and stability analysis of the test strip colorimetric kinetics process. Specifically, in typical home and clinical testing environments, images are continuously acquired at a fixed frequency for a large number of valid test strip samples within a reaction time window (e.g., 0–30 minutes), and the entire interpretation process is executed. The curves of the model response values ​​corresponding to each antigen over time are recorded. When the signal tends to stabilize, the distribution of response value changes at adjacent time points is statistically analyzed. A threshold value representing the convergence state of the interpretation result and capable of suppressing image shake and light fluctuations is selected as the stability judgment threshold ε. A balance is achieved between interpretation stability and response efficiency. When the change value is below the threshold, the current result is considered to have reached a stable state. =0.025 can ensure the high stability of the interpretation results while taking into account the user's operating efficiency. It is the optimal value that achieves the best balance between reliability and practicality. After satisfying the preset stability conditions, extract the set of multi-antigen interpretation results corresponding to the current time. Results of combined analysis of multiple antigens This will be output as the final interpretation result. The final interpretation results include the qualitative interpretation results of each antigen. Confidence level Risk level Probability scoring and explanatory information .

[0095] This embodiment also provides a multi-antigen test strip result interpretation system based on artificial intelligence visual recognition, including, The image acquisition and quality screening module is used to continuously capture test strip images via a mobile device and screen target analysis images based on multidimensional quality evaluation parameters. The geometric correction and region segmentation module is used to locate the center of the test strip and the direction of the detection arm, establish a polar coordinate system and complete geometric correction, and then segment the control line and detection line regions of each detection arm. The feature extraction and normalization module is used to extract the grayscale, color, width and edge features of the strip region, and perform normalization processing based on the same arm quality control line to obtain the multi-antigen feature matrix; The AI ​​interpretation and validity verification module is used to input normalized features into a pre-trained AI model for interpretation, combine the strength of the quality control line to judge the detection validity, and output the qualitative results and confidence level of each antigen. The risk assessment and dynamic decision-making module is used to perform joint analysis based on the results of multi-antigen interpretation, output risk level, probability score and explanatory information, and achieve stable interpretation by combining time series.

[0096] In summary, this invention significantly improves the robustness of multi-arm structure image analysis by constructing a polar coordinate system and implementing triple geometric correction, thereby achieving center positioning, arm orientation recognition, and strip segmentation of radial test strips. A normalization mechanism based on the same-arm quality control line is introduced to construct a normalized multi-antigen feature matrix, eliminating errors between different detection arms and achieving consistent interpretation of cross-channel features. Furthermore, by combining artificial intelligence model interpretation, quality control validity verification, multi-antigen combination decision-making, and a time-series dynamic stabilization mechanism, it not only outputs high-confidence qualitative results but also provides risk level, probability score, and interpretability information. In complex home environments, multi-frame time-series convergence ensures the reliability of the final interpretation, comprehensively improving the intelligence level and clinical applicability of simultaneous multi-indicator detection.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition, characterized in that: include, Acquire continuous images of the radial multiantigen test strip, and perform image quality assessment and screen target analysis images; Based on the target analysis image, locate the center point of the test strip card, identify the direction and number of each detection arm, establish a polar coordinate system, and perform geometric correction on the image according to the polar coordinate system; The fan-shaped regions of each detection arm are divided according to the polar coordinate system, and the quality control line and detection line in each detection arm are located to obtain the set of regions of interest of the strips; Visual features of the region of interest set of the strips are extracted, and the features of the detection line are normalized based on the features corresponding to the quality control line in the same detection arm to obtain a normalized multi-antigen feature matrix. The normalized multi-antigen feature matrix is ​​input into the pre-trained artificial intelligence model, and the effectiveness is judged by combining the corresponding features of the quality control line. The qualitative interpretation results and confidence levels of each antigen are output. Based on the qualitative interpretation results of each antigen, a combined feature vector is constructed, and the combined feature vector is input into a decision tree model to output risk level, probability score and explanatory information. A time series is constructed based on multiple interpretation results and dynamically updated. The final interpretation result is output when the preset stability condition is met.

2. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 1, characterized in that: The continuous images acquired from the radial multi-antigen test strip include: The test strip is photographed by a camera device on a smart terminal placed on a flat background, and multiple frames of images are continuously acquired based on the sampling frequency to form a time series.

3. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 2, characterized in that: The process of quality assessment and selection of target images for analysis includes: Sharpness, brightness, contrast, and framing integrity quality evaluation parameters are extracted from each image frame in the image sequence, and a quality evaluation value is obtained through weighted fusion. The quality evaluation value is compared with a preset threshold to filter candidate processing objects. After enhancing the candidate processing objects, the comprehensive quality score is recalculated and sorted. The image with the highest score is selected as the target analysis image.

4. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 3, characterized in that: The process of locating the center point of the test strip card based on the target analysis image, identifying the direction and number of each detection arm, establishing a polar coordinate system, and performing geometric correction on the image based on the polar coordinate system includes: Based on the target analysis image, the Hough circle transform is used to detect near-circular structures and determine the center point and radius of the central disk of the test strip card. Radial sampling is performed within a preset angle range with the center point as the origin. A radial response function is constructed by integrating the image intensity in each direction. Then, the direction of each detection arm is identified and the number of detection arms is determined by peak detection. A polar coordinate system is established with the center point as the origin of the polar coordinate system and the direction of the first detection arm as the angular reference. Based on the deviation and scale relationship between the direction of the detection arm and the standard template, the image is sequentially rotated, scaled, and perspective corrected to obtain the geometrically corrected image.

5. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 4, characterized in that: The process of dividing the fan-shaped regions of each detection arm according to the polar coordinate system, locating the control line and detection line in each detection arm, and obtaining the set of regions of interest for the strips includes: Based on the polar coordinate system and the direction information of the detection arm, the geometric correction image is segmented into a sector region to obtain the sector image region corresponding to each detection arm; The positions of the quality control line and the detection line in each detection arm are located by combining radial grayscale projection with peak detection and a preset strip position template. Further extract the corresponding regions to form the regions of interest for quality control lines and detection lines, and summarize them to obtain the set of regions of interest for the strips.

6. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 5, characterized in that: The visual features of the extracted region of interest set of the strips are then normalized using the features corresponding to the control lines in the same detection arm as a benchmark, resulting in a normalized multi-antigen feature matrix including: Visual features are extracted from the region of interest set of the strips; and the visual features of the control lines in the same detection arm are used as a benchmark to normalize the detection line features and construct a normalized multi-antigen feature matrix.

7. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 6, characterized in that: The process of inputting the normalized multi-antigen feature matrix into a pre-trained artificial intelligence model, and combining it with the corresponding features of the quality control line to determine effectiveness, outputting the qualitative interpretation results and confidence levels of each antigen, includes: The feature vectors of each normalized detection line in the normalized multi-antigen feature matrix are input into the pre-trained artificial intelligence model to obtain the response value corresponding to each antigen. The validity of the detection arm is determined based on the quality control line characteristics. Only for the detection arm that is determined to be valid, the qualitative interpretation result of each antigen is output according to the comparison result of the model response value and the corresponding judgment threshold. Simultaneously, the confidence level of the interpretation is calculated based on the relative relationship between the model response value and the judgment threshold, and a set of multi-antigen interpretation results is constructed.

8. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 7, characterized in that: The combined feature vector is constructed based on the qualitative interpretation results of each antigen, and then input into the decision tree model to output risk level, probability score, and explanatory information, including: Based on the set of multi-antigen interpretation results, the qualitative results of each antigen are extracted and a combined feature vector is constructed. The combined feature vector is then input into a pre-trained decision tree model to obtain the risk level and probability score. Explanatory information is generated based on the classification path of the decision tree model, thus forming the results of a single multi-antigen combined analysis.

9. The method for interpreting the results of multi-antigen test strips based on artificial intelligence visual recognition as described in claim 8, characterized in that: The step of constructing a time series based on multiple interpretation results and dynamically updating it, and outputting the final interpretation result when a preset stability condition is met, includes: Within the reaction time window of the test strip, a time series of multiple interpretation results is constructed and updated, and the change in interpretation results between adjacent time points is calculated; and stability is determined based on the change in response value, and the final interpretation result is output when the preset stability condition is met.

10. A multi-antigen test strip result interpretation system based on artificial intelligence visual recognition, based on the multi-antigen test strip result interpretation method based on artificial intelligence visual recognition according to any one of claims 1 to 9, characterized in that: include, The image acquisition and quality screening module is used to continuously capture test strip images via a mobile device and screen target analysis images based on multidimensional quality evaluation parameters. The geometric correction and region segmentation module is used to locate the center of the test strip and the direction of the detection arm, establish a polar coordinate system and complete geometric correction, and then segment the control line and detection line regions of each detection arm. The feature extraction and normalization module is used to extract the grayscale, color, width and edge features of the strip region, and perform normalization processing based on the same arm quality control line to obtain the multi-antigen feature matrix; The AI ​​interpretation and validity verification module is used to input normalized features into a pre-trained AI model for interpretation, combine the strength of the quality control line to judge the detection validity, and output the qualitative results and confidence level of each antigen. The risk assessment and dynamic decision-making module is used to perform joint analysis based on the results of multi-antigen interpretation, output risk level, probability score and explanatory information, and achieve stable interpretation by combining time series.

Images