An Adaptive Feature Extraction Method and System for Quantitative Detection of Fluorescent Immunochromatograms

Abstract

This invention discloses an adaptive feature extraction method and system for quantitative fluorescence immunoassay, relating to the field of fluorescence immunoassay technology. The method includes: acquiring a sequence of fluorescence speckle images; calculating the image gradient difference between adjacent frames; constructing a three-dimensional interferometric perturbation tensor; and obtaining the principal perturbation mode map through sparse principal component analysis. Based on the principal perturbation mode map, a self-supervised triplet training sample is constructed, and the perturbation value is used as a pixel-level weighting factor to train an image patch encoder. The trained encoder is used to extract image patch features, which are then matched with a response feature dictionary to generate a response probability map. The perturbation spectral entropy of each time-series map is calculated, and its reciprocal is used for weighted fusion to generate a fused response probability map. The result is output through mask screening and response integral calculation, using a fitting method. This invention effectively enhances the local immune response recognition capability, improves the adaptability of feature extraction under perturbation, and enhances the stability of response fusion, achieving high-precision quantitative fluorescence immunoassay.

Description

Technical Field

[0001] This invention relates to the field of fluorescence immunoassay technology, and in particular to an adaptive fluorescence immunoassay quantitative detection feature extraction method and system. Background Technology

[0002] With the rapid development of biomedical detection technology, fluorescence immunoassay has been widely used in many fields such as disease diagnosis, food safety, and environmental monitoring due to its advantages such as high sensitivity, strong specificity and simple operation. In practical applications, the use of fluorescently labeled antibodies to induce a specific immune reaction with the target antigen and the analysis of fluorescence signal intensity to achieve quantitative detection of the target substance has become one of the mainstream methods. Especially on new detection platforms such as microfluidic chips and point-of-care testing, immunoassay systems based on fluorescence signals are gradually becoming miniaturized, integrated and automated.

[0003] However, existing quantitative detection methods for fluorescence immunoassay lack in-depth modeling of the dynamic changes in the immune response process, and cannot capture subtle perturbation features in the image sequence, which limits the accuracy of identification of the immune layer response region and the robustness of quantitative output. In addition, current methods generally do not consider the differences in sensitivity of different response regions to feature extraction in the image, resulting in lag and error accumulation in the identification of response regions, making it difficult to meet the actual needs of high-precision quantitative detection. Summary of the Invention

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

[0005] Therefore, this invention provides an adaptive fluorescent immunoassay feature extraction method to solve the problems of existing methods being unable to capture dynamic perturbation features and having low accuracy in identifying image response regions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an adaptive method for quantitative detection of fluorescent immunoassay features, comprising: acquiring a sequence of fluorescent speckle images, calculating the image gradient difference between adjacent frames, constructing a three-dimensional interferometric perturbation tensor, and obtaining a principal perturbation mode map through sparse principal component analysis.

[0008] The image region is divided into response regions based on the main perturbation mode map, self-supervised triplet training samples are constructed, and the perturbation value is used as a pixel-level weighting factor to train the image block encoder.

[0009] The trained image patch encoder is used to perform sliding window processing on the fluorescent speckle image to extract the spatial features of each image patch. The spatial features of each image patch are then matched with a pre-constructed response feature dictionary to generate a response probability map for each pixel location.

[0010] Based on the response probability maps at different time points, the perturbation spectral entropy value of each response probability map is calculated. The reciprocal of each perturbation spectral entropy value is used as a weighting coefficient to perform weighted fusion of the response probability maps to generate a fused response probability map.

[0011] A mask screening threshold is set on the fusion response probability map to obtain the mask of the immune response region, and the immune response index of the immune response region is statistically analyzed. The concentration value of the fluorescence immune layer is calculated and output by the concentration fitting method.

[0012] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, the specific steps of constructing a three-dimensional interferometric perturbation tensor and obtaining the principal perturbation mode map through sparse principal component analysis are as follows:

[0013] Collect fluorescent speckle image frames at continuous time points throughout the entire fluorescent immunoassay process to obtain a sequence of fluorescent speckle image frames.

[0014] Spatial gradient calculation is performed on adjacent fluorescent speckle image frames in the fluorescent speckle image frame sequence, and the gradient difference is taken to obtain the fluorescent speckle image gradient difference sequence. The sequence is then arranged in chronological order along the time frame direction to form a three-dimensional interference perturbation tensor.

[0015] The pixel position is expanded on the three-dimensional interference perturbation tensor to form two-dimensional perturbation characteristic moments, and the principal perturbation response value is obtained by solving the optimization objective through a sparse principal component analysis model.

[0016] The principal perturbation response values ​​in the sparse principal component matrix are mapped back to the fluorescence speckle image frame to obtain the principal perturbation mode map.

[0017] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, the specific steps of constructing self-supervised triplet training samples and using perturbation values ​​as pixel-level weighting factors to train the image patch encoder are as follows:

[0018] The main perturbation mode map is normalized, a response threshold is set, and the pixel positions in the main perturbation mode map are divided into high response region, intermediate transition region and low response region according to the perturbation intensity based on the response threshold.

[0019] The pixel blocks in the high response region and the pixel blocks in the low response region are used as positive sample image blocks, anchor sample image blocks and negative sample image blocks as candidate sets of triplet samples;

[0020] In the normalized principal perturbation mode map, positive sample image blocks, anchor sample image blocks, and negative sample image blocks are extracted by a sliding window to form self-supervised triplet training samples.

[0021] The perturbation value at each pixel position in the self-supervised triplet training samples is sampled to generate a pixel-level perturbation weighting factor matrix;

[0022] The self-supervised triplet training samples and the pixel-level perturbation weighting factor matrix are input into the image patch encoder to calculate the contrastive learning loss function and update the internal training parameters of the image patch encoder.

[0023] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, the specific steps for generating the response probability map of each pixel location are as follows:

[0024] A sliding window process is applied to the fluorescence speckle image to generate an image patch sliding window sequence. Each image patch in the image patch sliding window sequence is then input into the trained image patch encoder to obtain the spatial features of the image patch.

[0025] A response feature dictionary is constructed using manually labeled high-response region image patches;

[0026] The similarity scores between the spatial features of the image patch and each spatial feature in the response feature dictionary are calculated sequentially, and the response probability value of the image patch is obtained by the Top-K average similarity.

[0027] The response probability values ​​are mapped according to the starting position of the sliding window in the fluorescence speckle image to construct a response probability map for each pixel location.

[0028] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, the specific steps for calculating the perturbation spectral entropy value of each response probability spectrum based on response probability spectra at different time points are as follows:

[0029] The response probability map of each frame is normalized, and a pixel grayscale histogram is constructed based on each normalized response probability map to obtain the probability distribution histogram.

[0030] The perturbation spectral entropy of the response probability map is calculated using the probability distribution histogram.

[0031] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, wherein: the step of using the reciprocal of the entropy value of each perturbation spectrum as a weighting coefficient to perform weighted fusion of the response probability spectrum to generate a fused response probability spectrum, specifically includes:

[0032] Calculate the reciprocal of each perturbation spectral entropy value to obtain the perturbation spectral entropy reciprocal weight set;

[0033] The set of weighted inverses of the perturbation spectral entropy is normalized to generate a set of weighted coefficients of the normalized inverses of the perturbation spectral entropy. The response probability map is then weighted and summed to obtain the fusion response probability value. This fusion response probability value is then mapped onto the fluorescence speckle image to generate the fusion response probability map.

[0034] As a preferred embodiment of the adaptive fluorescence immunoassay feature extraction method of the present invention, the specific steps for calculating and outputting the fluorescence immunoassay concentration value are as follows:

[0035] Local window statistics are performed on the fusion response probability map. The mask selection threshold is calculated based on the pixel mean and standard deviation within the sliding window. The mask selection threshold mapping value is then fused into the response probability map to generate a mask selection threshold map.

[0036] The pixel locations in the fusion response probability map where the fusion response probability value is greater than the mask screening threshold are marked as immune response regions, and a mask for the immune response regions is generated.

[0037] Immune response indicators are statistically analyzed within the mask of the immune response region to form a set of immune response indicators;

[0038] The concentration value of the fluorescence immunoassay layer was obtained by calculating the response integral using a concentration fitting method.

[0039] In a second aspect, the present invention provides an adaptive fluorescence immunoassay feature extraction system, comprising a perturbation extraction module for acquiring a fluorescence speckle image sequence, calculating the image gradient difference between adjacent frames, constructing a three-dimensional interference perturbation tensor, and acquiring the main perturbation mode map through sparse principal component analysis.

[0040] The encoding training module is used to divide the image region into response regions based on the main perturbation mode map, construct self-supervised triplet training samples, and use the perturbation value as a pixel-level weighting factor to train the image block encoder.

[0041] The feature matching module is used to perform sliding window processing on the fluorescent speckle image using the trained image patch encoder, extract the spatial features of each image patch, and match the spatial features of each image patch with the pre-built response feature dictionary to generate the response probability map of each pixel location.

[0042] The graph fusion module is used to calculate the perturbation spectral entropy value of each response probability graph based on response probability graphs at different time points, and to use the reciprocal of each perturbation spectral entropy value as a weighting coefficient to perform weighted fusion of the response probability graphs to generate a fused response probability graph.

[0043] The immune parameter output module is used to set the mask screening threshold on the fusion response probability map, obtain the mask of the immune response region, and count the immune response index of the immune response region. It also calculates and outputs the quantitative detection concentration value of the fluorescent immune layer through the concentration fitting method.

[0044] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the adaptive fluorescence immunoassay feature extraction method as described in the first aspect of the present invention.

[0045] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the adaptive fluorescence immunoassay feature extraction method as described in the first aspect of the present invention.

[0046] The beneficial effects of this invention are as follows: by constructing a three-dimensional interferometric perturbation tensor and using sparse principal component analysis, the ability to identify early local response regions is enhanced; by using self-supervised triplet samples and using perturbation values ​​as weighting coefficients for training image block encoders, the spatial recognition accuracy of response regions is improved; and by using weighted perturbation spectral entropy, local noise interference is effectively suppressed, and the stability and robustness of the response spectrum are improved. Attached Figure Description

[0047] 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.

[0048] Figure 1 This is a flowchart of an adaptive fluorescence immunoassay feature extraction method.

[0049] Figure 2 This is a schematic diagram of an adaptive fluorescence immunoassay feature extraction system.

[0050] Figure 3 The flowchart shows the training process of the image block encoder for the main perturbation mode map.

[0051] Figure 4 The flowchart for generating probability maps and outputting concentrations in response. Detailed Implementation

[0052] 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.

[0053] 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.

[0054] 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.

[0055] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an adaptive fluorescent immunoassay feature extraction method, comprising the following steps:

[0056] S1: Obtain the fluorescence speckle image sequence, calculate the image gradient difference between adjacent frames, construct the three-dimensional interferometric perturbation tensor, and obtain the principal perturbation mode map through sparse principal component analysis.

[0057] Fluorescent speckle image frames were acquired at continuous time points throughout the entire fluorescence immunoassay process using a time-series imaging device, resulting in a sequence of fluorescent speckle image frames.

[0058] Furthermore, on the detection slide, fluorescent labeling and fixation of antigens or antibodies are completed, and the immune binding reaction is initiated; a fluorescence imaging timing device with short exposure response capability is used, and a stable wavelength excitation source is set; the observation field is fixed, the time-series imaging function is activated, the acquisition frequency is set, and multiple frame image sequences are continuously acquired throughout the entire immune reaction process; the total acquisition time is set, and the fluorescence speckle image frame sequence is saved. Where I represents the fluorescence speckle image frame sequence, t represents the frame number, (x,y) represents the spatial coordinates of the fluorescence speckle image, and T is the total number of frames.

[0059] Spatial gradient calculations are performed on adjacent fluorescent speckle image frames in the fluorescent speckle image frame sequence, and the gradient difference is taken to obtain the fluorescent speckle image gradient difference sequence.

[0060] Furthermore, for each frame of the fluorescence speckle image frame sequence, I... t Applying the gradient operator to (x,y) calculates the spatial gradient value and generates a spatial gradient map, represented as:

[0061]

[0062] in, Let represent the spatial gradient value of the t-th frame image at the fluorescence speckle image spatial coordinates (x, y). Represents fluorescent speckle image I t The partial derivative in the x-direction, Represents fluorescent speckle image I t Partial derivative in the y-direction;

[0063] For every two adjacent frames t and t-1, the absolute value of the spatial gradient difference is calculated to form a sequence of gradient differences under fluorescence speckle, which is expressed as:

[0064]

[0065] Among them, D t (x,y) represents the interference value at the spatial coordinates (x,y) in the fluorescence speckle image of frame t. Image I representing the fluorescence speckle pattern at frame number t-1 t Spatial gradient value at spatial coordinates (x, y) in the fluorescent speckle image.

[0066] The gradient difference sequence of the fluorescence speckle image is arranged in chronological order along the time frame direction to form a three-dimensional interferometric perturbation tensor.

[0067] Furthermore, the gradient difference sequence of fluorescence speckle is arranged along the time dimension to construct a three-dimensional interferometric perturbation tensor, which is represented as:

[0068] D(x,y,t)={D2(x,y),D3(x,y),...,D T (x,y)};

[0069] Where D(x,y,t) represents the three-dimensional interference perturbation tensor, with a three-dimensional structure of H×W×(T-1); where H and W are the image height and width of the perturbation map, respectively.

[0070] The pixel positions of the three-dimensional interference perturbation tensor are expanded to form a two-dimensional perturbation feature matrix.

[0071] Furthermore, the three-dimensional interferometric perturbation tensor D(x,y,t) is expanded into a two-dimensional sample matrix.

[0072] The two-dimensional perturbation feature matrix is ​​input into the sparse principal component analysis model, and the sparse principal component matrix of the perturbation direction is extracted based on the sparse constraint conditions.

[0073] Furthermore, we construct a sparse PCA optimization problem, with the optimization objective expressed as:

[0074] max v Var(Xv)st‖v‖2=1,‖v‖1≤λ1;

[0075] in, This represents the matrix obtained by expanding the three-dimensional interferometric perturbation tensor D(x,y,t) according to the pixel rows and time as columns. Let VX represent the projection vector, Var(Xv) represent the variance of the projection result, and λ1 be the coefficient constraint coefficient.

[0076] Based on the optimization objective, the sparse PCA optimization problem of the three-dimensional interferometric perturbation tensor is solved to obtain the principal perturbation response value.

[0077] It should be noted that constructing a sparse principal component analysis model involves adding an L1 regularization term to the traditional PCA objective, thereby making the solution sparse.

[0078] The principal perturbation response values ​​in the sparse principal component matrix are mapped back to the fluorescence speckle image frame to obtain the principal perturbation mode map.

[0079] Furthermore, the principal component projection results are reshaped into a two-dimensional image size to obtain the principal perturbation mode map F1(x,y); where the principal perturbation mode map F1(x,y) represents the principal perturbation feature intensity of each pixel position in the time dimension, and the value corresponding to each pixel position (x,y) in the principal perturbation mode map is defined as the perturbation value.

[0080] S2: Based on the main perturbation mode map, the response region of the fluorescent speckle image is divided, a self-supervised triplet training sample is constructed, and the perturbation value is used as a pixel-level weighting factor to train the image block encoder.

[0081] The minimum and maximum values ​​of the main perturbation mode map are normalized to generate the normalized main perturbation mode map F2(x,y).

[0082] A response threshold is set for the normalized main perturbation mode map, and the pixel positions are divided into high response regions, intermediate transition regions, and low response regions according to the perturbation intensity.

[0083] Furthermore, based on the typical value of the response of the region where the immune binding reaction is located on the normalized principal perturbation mode map, and based on the false perturbation caused by lens or background shaking, the response threshold is set to [ξ0,ξ1].

[0084] Based on the response threshold, the normalized principal perturbation mode map is divided into a high response region, an intermediate transition region, and a low response region. Specifically, if F2(x,y)≥ξ1, it is classified as a high response region; if ξ0<F2(x,y)<ξ1, it is classified as an intermediate transition region; and if F2(x,y)≤ξ0, it is classified as a low response region.

[0085] The positive sample image block, anchor sample image block, and negative sample image block are selected as candidate sets of triplet samples by using pixel blocks in the high response region and pixel blocks in the low response region.

[0086] Furthermore, pixel blocks in high-response regions and low-response regions are extracted. Pixel blocks in the same high-response region but in different frames are used as positive sample image blocks and anchor sample image blocks in the candidate set of triplet samples, while pixel blocks in low-response regions are used as negative sample image blocks in the candidate set of triplet samples.

[0087] In the normalized principal perturbation mode map, positive sample image blocks, anchor sample image blocks, and negative sample image blocks are extracted by a sliding window to form self-supervised triplet training samples.

[0088] Furthermore, the normalized principal perturbation mode map is divided into image blocks according to an a×a pixel sliding window, and the perturbation value of the center pixel of each image block is used as a response reference index. The generation rule of the self-supervised triple is as follows:

[0089] Let image block P i The location center is (x i ,y i If the image block P is... i The disturbance value is F2(x) i ,y i If F2(x) is satisfied i ,y i If ξ ≥ 1, then image block P i Two snapshots at different time frames t1 and t2 They are constructed as positive sample image patches and anchor sample image patches; from different positions, they satisfy F2(x i ,y i Image blocks ≤ ξ0 Construct as negative sample image patches;

[0090] Furthermore, the self-supervised triplet training samples are represented as follows:

[0091]

[0092] in, This represents the training samples of the self-supervised triplet. Represents the positive sample image patch in the i-th triplet. Represents the anchor sample image block in the i-th triplet. Let N represent the negative sample image block in the i-th triplet, and N represent the total number of triplets.

[0093] The perturbation values ​​of the normalized principal perturbation mode map at each pixel location in the positive sample image block, anchor sample image block, and negative sample image block are sampled to generate a pixel-level perturbation weighting factor matrix.

[0094] Furthermore, the interference values ​​in the main perturbation mode map are normalized and sampled. The normalized interference value of the center pixel of the image block is used as the pixel-level perturbation weighting factor, expressed as:

[0095] w i =F2(x i ,y i )∈[0,1];

[0096] Among them, w i Let represent the weighting coefficient of the i-th triplet training sample, with a value range of [0,1]; construct the contrastive loss function of the pixel-level perturbation weighting factor, expressed as:

[0097]

[0098] in, The contrastive loss function represents the pixel-level perturbation weighting factor. This represents the traditional contrastive loss function.

[0099] The self-supervised triplet training samples and the pixel-level perturbation weighting factor matrix are input into the image patch encoder to calculate the contrastive learning loss function and update the internal training parameters of the image patch encoder.

[0100] Furthermore, the image patch encoder performs iterative training by using a contrastive loss function weighted by perturbation values. After each round of updates, the average similarity change between positive sample image patches and anchor sample image patches is evaluated to monitor training convergence. The model parameters with the minimum loss on the validation set are selected as the parameter configuration of the final model.

[0101] S3: Use the trained image patch encoder to perform sliding window processing on the fluorescent speckle image, extract the spatial features of each image patch, and match the spatial features of each image patch with the pre-constructed response feature dictionary to generate the response probability map of each pixel location.

[0102] The fluorescent speckle image is processed by sliding windowing according to a set step size and image block size to generate an image block sliding window sequence.

[0103] Furthermore, a sliding window process is applied to the fluorescence speckle image, with a step size of b pixels and an image patch size of c×c pixels, to generate an image patch sliding window sequence.

[0104] Each image patch in the image patch sliding window sequence is input into the trained image patch encoder to obtain the spatial features of the image patch.

[0105] A response feature dictionary is constructed using manually labeled high-response region image patches.

[0106] The similarity scores are calculated sequentially between the image patch spatial features and each response spatial feature in the response feature dictionary to form a similarity score vector of the image patch features.

[0107] Furthermore, high-response region image patches are selected from the training dataset through manual annotation. Reference spatial features of these high-response region image patches are then extracted using an image encoder to construct a response feature dictionary, represented as follows:

[0108]

[0109] in, The response feature dictionary consists of spatial features from M representative high-response image patches. This represents the k-th reference space feature in the response dictionary.

[0110] Furthermore, the similarity between the spatial features of each image patch and all the reference spatial features in the response dictionary is calculated to form a similarity score vector of the image patch features;

[0111] The similarity score vectors of image patch features are subjected to Top-K average similarity analysis to obtain the response probability value of the image patch.

[0112] Furthermore, the response probability value of an image patch is defined as the maximum similarity value. The response probability value of an image patch is obtained by using the Top-K average similarity, and is expressed as:

[0113]

[0114] in, v represents the response probability value of the image patch at pixel space coordinates (x, y) in frame t. (t) This represents the spatial characteristics of the image block in spatial coordinates at frame t.

[0115] The response probability values ​​are mapped according to the starting position of the sliding window in the fluorescence speckle image to construct a response probability map for each pixel location.

[0116] Furthermore, the response probability values ​​of all image patches are remapped according to the original coordinates to fill a two-dimensional response probability map.

[0117] S4: Based on the response probability maps at different time points, calculate the perturbation spectral entropy value of each response probability map, and use the reciprocal of each perturbation spectral entropy value as a weighting coefficient to perform weighted fusion of the response probability maps to generate a fused response probability map.

[0118] The response probability map of each frame is normalized to generate a set of normalized response probability maps.

[0119] Based on each normalized response probability map, construct a pixel grayscale histogram to obtain the probability distribution histogram.

[0120] Furthermore, based on the image response feature accuracy requirements, a fixed number of bins is set for the normalized response probability map. For each normalized response probability map, all pixels are traversed to determine the bin range into which each pixel value falls. The number of pixels in each bin is counted to generate a pixel grayscale histogram. The statistical results are then used to calculate the ratio of the total number of pixels in the image to obtain the probability corresponding to each bin, forming a probability distribution histogram.

[0121] The perturbation spectral entropy of the response probability map is calculated using the probability distribution histogram.

[0122] Furthermore, for each time frame, the perturbation spectral entropy is calculated based on the distribution of the probability distribution histogram of the normalized response probability map, and is expressed as:

[0123]

[0124] in, Let represent the perturbation spectral entropy of the fluorescence speckle image in frame t, and B represent the number of bins in the pixel grayscale histogram. Let represent the probability histogram distribution in the b-th gray-level interval of the t-th frame of the fluorescence speckle image, and ε represent a minimum constant.

[0125] Calculate the reciprocal of each perturbation spectral entropy value to obtain the set of perturbation spectral entropy reciprocal weights.

[0126] The set of weights corresponding to the inverse of the perturbation spectral entropy is normalized to generate a set of weighted coefficients corresponding to the normalized inverse of the perturbation spectral entropy.

[0127] Based on the set of weighted coefficients of the normalized reciprocal of the perturbation spectral entropy, the response probability spectra at the corresponding time points are weighted and summed to obtain the fusion response probability value, and the fusion response probability value is mapped onto the fluorescence speckle image to generate the fusion response probability spectrum.

[0128] The value corresponding to each pixel position on the fusion response probability map is the fusion response probability value.

[0129] Furthermore, the fusion response probability map is calculated and represented as:

[0130]

[0131] Among them, F f (x,y) represents the value of the fused response probability map at pixel position (x,y).

[0132] S5: Set a mask screening threshold on the fusion response map, obtain the mask of the immune response region, and count the immune response index of the immune response region. Calculate and output the quantitative detection concentration value of the fluorescent immune layer through the concentration fitting method.

[0133] Local window statistics are performed on the fused response probability map. The mask selection threshold is calculated based on the pixel mean and standard deviation within the sliding window. The mask selection threshold is then mapped into the fused response probability map to generate a mask selection threshold map.

[0134] Furthermore, the pixel mean μ and standard deviation σ of the fused response map are calculated, and a mask screening threshold is set, expressed as:

[0135] Y = μ + λ²·σ;

[0136] Where Y represents the mask screening threshold, and λ2 represents the adjustable hyperparameter, which is usually set to 0.5 to 1.0.

[0137] Pixels with fusion response probability values ​​greater than the mask screening threshold in the fusion response probability map are marked as immune response regions, and masks for the immune response regions are generated.

[0138] Furthermore, based on the mask screening threshold, the fusion response probability value in the fusion response map is compared with the mask screening threshold, and the pixel positions that are greater than the mask screening threshold are marked as immune response regions.

[0139] The mask value M'(x,y) at the pixel location of the mask image of the immune response region is represented as:

[0140]

[0141] If the value of M'(x,y) is 1, it means that the current pixel position belongs to the immune response region and needs to be marked; if the value of M'(x,y) is 0, it means that the current pixel position belongs to the non-immune response region and does not need to be marked.

[0142] Immune response indicators are statistically analyzed within the mask of the immune response region to form an immune response indicator set; the immune response indicators include response area, response integral, and average response intensity.

[0143] Furthermore, the response area A is expressed as:

[0144] A = ∑ x,y M'(x,y);

[0145] The response integral S is expressed as:

[0146] S=∑ x,y F f (x,y)·M'(x,y);

[0147] Average response strength Represented as:

[0148]

[0149] The response integral is calculated using a concentration fitting method, and the concentration value of the fluorescence immunoassay layer is output as a quantitative detection value.

[0150] Furthermore, the concentration fitting method specifically involves: collecting multiple samples with known concentrations, and using a nonlinear least squares method to estimate and fit the values ​​of four parameters: A1, A2, C0, and z, as follows:

[0151]

[0152] Where C represents the quantitative detection concentration value of the fluorescent immunoassay layer, A1 represents the upper limit concentration response of the quantitative detection concentration value of the fluorescent immunoassay layer, A2 represents the lower limit concentration response of the quantitative detection concentration value of the fluorescent immunoassay layer, C0 represents the inflection point of the response curve, and z represents the curve steepness control parameter.

[0153] A1 is the upper limit of the response integral when the concentration of the quantitative detection of the fluorescent immunoassay layer approaches 0, A2 is the lower limit of the response integral when the concentration of the quantitative detection of the fluorescent immunoassay layer approaches infinity, C0 is the response integral value corresponding to the inflection point of the response curve, and z is the steepness of the curve.

[0154] This embodiment also provides an adaptive fluorescence immunoassay feature extraction system, including:

[0155] The perturbation extraction module is used to acquire fluorescent speckle image sequences, calculate the image gradient difference between adjacent frames, construct a three-dimensional interferometric perturbation tensor, and obtain the main perturbation mode map through sparse principal component analysis.

[0156] The encoding training module is used to divide the image region into response regions based on the main perturbation mode map, construct self-supervised triplet training samples, and use the perturbation value as a pixel-level weighting factor to train the image block encoder.

[0157] The feature matching module is used to perform sliding window processing on the fluorescent speckle image using the trained image patch encoder, extract the spatial features of each image patch, and match the spatial features of each image patch with the pre-built response feature dictionary to generate the response probability map of each pixel location.

[0158] The graph fusion module is used to calculate the perturbation spectral entropy value of each response probability graph based on response probability graphs at different time points, and to use the reciprocal of each perturbation spectral entropy value as a weighting coefficient to perform weighted fusion of the response probability graphs to generate a fused response probability graph.

[0159] The immune parameter output module is used to set the mask screening threshold on the fusion response probability map, obtain the mask of the immune response region, and count the immune response index of the immune response region. It also calculates and outputs the quantitative detection concentration value of the fluorescent immune layer through the concentration fitting method.

[0160] This embodiment also provides a computer device applicable to the adaptive fluorescence immunoassay feature extraction method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the adaptive fluorescence immunoassay feature extraction method proposed in the above embodiment.

[0161] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0162] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the adaptive fluorescence immunoassay feature extraction method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0163] In summary, this invention enhances the early local response region recognition capability by constructing a three-dimensional interferometric perturbation tensor and employing sparse principal component analysis; improves the spatial recognition accuracy of the response region by using self-supervised triplet samples and applying perturbation values ​​as weighting coefficients for training the image block encoder; and effectively suppresses local noise interference and improves the stability and robustness of the response map by using weighted perturbation spectral entropy.

[0164] 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 extracting features for quantitatively detecting a sample using adaptive fluorescence immunoassay, the method comprising: The application relates to a fluorescence immunolayer quantitative detection method based on immune response region extraction and response probability spectrum generation. ​ The method comprises the following steps of, acquiring a fluorescence speckle image sequence, calculating image gradient difference values between adjacent frames, constructing a three-dimensional interference disturbance tensor, and acquiring a main disturbance mode atlas through sparse principal component analysis; based on the main disturbance mode atlas, dividing an image region into response regions, constructing a self-supervised triple training sample, and taking the disturbance value as a pixel-level weighting factor to train an image block encoder; using the trained image block encoder to perform sliding window processing on the fluorescence speckle image, extracting spatial features of each image block, and matching the spatial features of each image block with a pre-constructed response feature dictionary to generate a response probability atlas of each pixel position; based on the response probability atlases at different time points, calculating disturbance spectral entropy values of each response probability atlas, weighting and fusing the response probability atlases by taking the reciprocal of each disturbance spectral entropy value as a weighting coefficient to generate a fused response probability atlas; 2. The adaptive fluorescent immunoassay quantitative detection feature extraction method of claim 1, wherein: setting a mask screening threshold on the fused response probability atlas, acquiring a mask of an immune response region, and counting an immune response index of the immune response region, and calculating and outputting a fluorescence immunolayer quantitative detection concentration value through a concentration fitting method. The method comprises the following steps of, acquiring a fluorescence speckle image sequence, calculating image gradient difference values between adjacent frames, constructing a three-dimensional interference disturbance tensor, and acquiring a main disturbance mode atlas through sparse principal component analysis; acquiring fluorescence speckle image frames at continuous time points in the whole process of fluorescence immunoreaction, and acquiring a fluorescence speckle image frame sequence; calculating the spatial gradient of adjacent fluorescence speckle image frames in the fluorescence speckle image frame sequence and taking the gradient difference value to acquire a fluorescence speckle image gradient difference value sequence, and arranging the fluorescence speckle image gradient difference value sequence in time sequence along the time frame direction to form a three-dimensional interference disturbance tensor; 3. The method of claim 2, wherein the method further comprises: determining the intensity of the fluorescent signal of the sample; and determining the intensity of the fluorescent signal of the control. pixel position unfolding is performed on the three-dimensional interference disturbance tensor to form a two-dimensional disturbance feature matrix, and a main disturbance response value is acquired through a sparse principal component analysis model based on an optimization target solution; the main disturbance response value in the sparse principal component matrix is mapped back to the fluorescence speckle image frame to acquire a main disturbance mode atlas. The method comprises the following steps of, normalizing the main disturbance mode atlas, setting a response threshold, and dividing the pixel positions in the main disturbance mode atlas into a high-response region, an intermediate transition region and a low-response region according to the disturbance intensity according to the response threshold; taking the pixel blocks in the high-response region and the pixel blocks in the low-response region as positive sample image blocks, anchor sample image blocks and negative sample image blocks of a triple sample candidate set; extracting the positive sample image blocks, the anchor sample image blocks and the negative sample image blocks from the normalized main disturbance mode atlas through a sliding window to form a self-supervised triple training sample; 4. The method of claim 3, wherein the method further comprises: sampling the disturbance values of each pixel position in the self-supervised triple training sample to generate a pixel-level disturbance weighting factor matrix; inputting the self-supervised triple training sample and the pixel-level disturbance weighting factor matrix into the image block encoder to calculate a contrast learning loss function and update the training parameters in the image block encoder. The method comprises the following steps of, ​ The speckle image is subjected to sliding window processing to generate a sliding window sequence of image blocks, each image block in the sliding window sequence of image blocks is input into the trained image block encoder, and image block spatial features are obtained; A response feature dictionary is constructed based on the manually labeled high response region image blocks; The image block spatial features and each response spatial feature in the response feature dictionary are sequentially calculated for a similarity score, and a response probability value of the image block is obtained through Top-K average similarity; The response probability value is mapped according to the sliding window starting position in the speckle image to construct a response probability atlas of each pixel position.

5. The adaptive fluorescent immunoassay quantitative detection feature extraction method of claim 4, wherein: The response probability atlas at different time points is used to calculate the perturbation spectral entropy value of each response probability atlas, and the specific steps are, Each response probability atlas is normalized, a pixel gray histogram is constructed based on each normalized response probability atlas, and a probability distribution histogram is obtained; The probability distribution histogram is used to calculate the perturbation spectral entropy value of the response probability atlas.

6. The adaptive fluorescent immunoassay quantitative detection feature extraction method of claim 5, wherein: Each perturbation spectral entropy value is used as a weighting coefficient to weight and fuse the response probability atlas to generate a fused response probability atlas, and the specific steps are, The reciprocal of each perturbation spectral entropy value is calculated to obtain a perturbation spectral entropy reciprocal weight set; The perturbation spectral entropy reciprocal weight set is normalized to generate a perturbation spectral entropy normalized reciprocal weighting coefficient set, the response probability atlas is weighted and summed to obtain a fused response probability value, and the fused response probability value is mapped to the speckle image to generate a fused response probability atlas.

7. The adaptive fluorescent immunoassay test strip quantitative detection feature extraction method of claim 6, wherein: The concentration value of the fluorescent immunolayer quantitative detection is calculated and output, and the specific steps are, The fused response probability atlas is subjected to local window statistics, a mask screening threshold is calculated based on the mean and standard deviation of the pixels in the sliding window, and the mask screening threshold is mapped to the fused response probability atlas to generate a mask screening threshold map; The pixel positions with a fused response probability value greater than the mask screening threshold in the fused response probability atlas are marked as immune response regions to generate a mask of the immune response regions; The immune response index is counted in the mask of the immune response regions to form an immune response index set; The response integral is calculated by a concentration fitting method to obtain the concentration value of the fluorescent immunolayer quantitative detection.

8. A self-adapting fluorescent immunoassay layer quantitative detection feature extraction system based on the self-adapting fluorescent immunoassay layer quantitative detection feature extraction method of any one of claims 1-7, characterized in that: It comprises, A perturbation extraction module is configured to obtain a sequence of speckle images, calculate image gradient difference values between adjacent frames, construct a three-dimensional interference perturbation tensor, and obtain a main perturbation mode atlas through sparse principal component analysis; An encoding training module is configured to divide image regions into response regions based on the main perturbation mode atlas, construct a self-supervised triple training sample, and train an image block encoder by taking the perturbation value as a pixel-level weighting factor; A feature matching module is configured to use the trained image block encoder to perform sliding window processing on the speckle image, extract spatial features of each image block, and match the spatial features of each image block with a pre-constructed response feature dictionary to generate a response probability atlas of each pixel position; An atlas fusion module is configured to calculate the perturbation spectral entropy value of each response probability atlas based on the response probability atlas at different time points, use the reciprocal of each perturbation spectral entropy value as a weighting coefficient to weight and fuse the response probability atlas, and generate a fused response probability atlas. An immune parameter output module is configured to set a mask screening threshold on a fusion response probability map, obtain a mask of an immune response region, and count an immune response index of the immune response region, and calculate and output a fluorescence immunoassay layer quantitative detection concentration value through a concentration fitting method. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The computer program, when executed by the processor, implements the steps of the adaptive fluorescence immunoassay layer quantitative detection feature extraction method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program, when executed by the processor, implements the steps of the adaptive fluorescence immunoassay layer quantitative detection feature extraction method of any one of claims 1-7.

Images