Method, apparatus and storage medium for correction of digital pathology fluorescence images

Abstract

The application discloses a kind of correction method, equipment and storage medium of digital pathology fluorescence image, it is related to digital pathology technical field, the method comprises: the original fluorescence image is down-sampled to obtain sampling fluorescence image, and initial path parameter and initial quenching field parameter of sampling fluorescence image are solved according to joint optimization objective function;Initial path parameter is up-sampled and mapped to the image coordinate of original fluorescence image, to obtain original path parameter;Target quenching field of original fluorescence image is solved based on original path parameter, initial quenching field parameter and joint optimization objective function;Original fluorescence image is corrected according to target quenching field, to obtain target fluorescence image.The application solves the problem of initial path and quenching field parameter by multi-resolution joint optimization scheme, and then mapping to high resolution to solve target quenching field, solves the problem of pathological quantitative analysis result distortion, improves the iterative convergence speed, while giving consideration to correction accuracy and pathological signal fidelity effect.

Description

Technical Field

[0001] This application relates to the field of digital pathology technology, and in particular to a method, device and storage medium for correcting digital pathological fluorescence images. Background Technology

[0002] In the context of digital pathology whole-slice fluorescence scanning and precision medicine quantitative analysis, the accuracy of brightness correction and the fidelity of fluorescence signal quantification in fluorescence scan images are directly related to the quality of pathological imaging and the reliability of precision medicine research such as multiplex immunofluorescence analysis and spatial transcriptomics. The effective application of adaptive brightness gradient correction software algorithms that can explicitly model the causal relationship between the scanning process and fluorescence quenching in this context is the core foundation for improving the level of automated correction and quantitative analysis accuracy of fluorescence pathology images.

[0003] In related technologies, fluorescence scan images are processed by standard fluorescence reference slide hardware calibration, flat field correction, global or local histogram equalization, and reference region signal standardization. This method, based on a fixed reference or local statistical features of the image, performs forced normalization and equalization correction operations on the original pixel intensity of the fluorescence image, which can easily destroy the quantitative relationship of fluorescence intensity, thereby leading to distortion of the pathological quantitative analysis results.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide a method, device, and storage medium for correcting digital pathological fluorescence images, aiming to solve the technical problem of distortion in the results of quantitative pathological analysis.

[0006] To achieve the above objectives, this application proposes a method for correcting digital pathological fluorescence images, the method comprising: A sampled fluorescence image is obtained by downsampling the original fluorescence image, wherein the resolution of the sampled fluorescence image is lower than that of the original fluorescence image; The initial path parameters and initial quenching field parameters of the sampled fluorescence image are solved based on the initial scan path parameters and the joint optimization objective function, wherein the joint optimization objective function aims to minimize the objective function value. The initial path parameters are upsampled and mapped to the image coordinates of the original fluorescence image to obtain the original path parameters; The target quenching field of the original fluorescence image is solved based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function. The original fluorescence image is corrected according to the target quenching field to obtain the target fluorescence image.

[0007] In one embodiment, a scan path parameterization model is initialized based on the pixel coordinate system and image size of the sampled fluorescence image to obtain a scan path framework adapted to the sampled fluorescence image; Based on the scanning path framework, the corresponding scanning relative time is matched for each pixel in the two-dimensional pixel coordinate system of the sampled fluorescence image, and the mapping relationship between the scanning path and the scanning relative time is constructed. The scanning sequence field is obtained by matching the scanning time index corresponding to the pixel in the sampled fluorescence image with the mapping relationship.

[0008] In one embodiment, the fluorescence intensity of each pixel in the sampled fluorescence image is correlated with the scanning time sequence in the scanning sequence field to obtain a correlation dataset of fluorescence intensity and scanning time sequence; Based on the associated dataset and the spatially stationary statistical distribution of fluorescence signals from similar biological tissues, a data fidelity term is constructed to measure the local statistical uniformity of the corrected image. Based on the physical decay law of fluorescence quenching and the associated dataset, constraints are applied to the spatial gradient of the quenching decay field, and a model regularization term is constructed to constrain the smoothness and monotonicity of the quenching decay field. The data fidelity term and the model regularization term are weighted and fused to obtain the joint optimization objective function used to simultaneously optimize the scan path and the quenching attenuation field.

[0009] In one embodiment, based on the spatially stationary statistical distribution of fluorescence signals from the same type of biological tissue, the sampled fluorescence image is uniformly divided into blocks to obtain local image blocks; For each local image block, the average value of the corrected fluorescence intensity of all pixels within the local image block is calculated to obtain the local average intensity. The fluorescence intensity variance of the local image patch is calculated based on the difference between the average value of each fluorescence intensity and the corresponding local intensity mean. The average value of the fluorescence intensity variance of all the local image patches is used to obtain the data fidelity term.

[0010] In one embodiment, a pixel-level temporal correlation set is constructed by matching the relative scanning time of each pixel in the sampled fluorescence image according to the initialization of the scanning path parameters; The pixel-level temporal correlation set is iteratively processed by the joint optimization objective function, and the path parameters and quenching field parameters are updated round by round, and the corresponding objective function values ​​are calculated. By comparing the objective function values ​​obtained through continuous iterations, once the objective function value satisfies the convergence condition, the path parameters and quenching field parameters corresponding to the objective function value are determined as the initial path parameters and initial quenching field parameters.

[0011] In one embodiment, the original path parameters and the initial quenching field parameters are transformed by coordinate mapping and matched to the pixel coordinate system of the original fluorescence image to generate the original scanning sequence field; The original fluorescence image is segmented based on the original scanning sequence field to obtain image regions of homogeneous tissue type; Using image regions of homogeneous tissue types as constraint units, the original scanning sequence field is optimized by region division according to the constraint units to obtain the target scanning sequence field; Based on the target scanning sequence field and the physical attenuation law of fluorescence quenching, the target quenching field is obtained by solving the joint optimization objective function.

[0012] In one embodiment, based on the pixel intensity information of the original fluorescence image, the quenching attenuation coefficient corresponding to the same spatial location in the target quenching field is matched; Based on the principle of reverse compensation for fluorescence quenching and the quenching attenuation coefficient, the pixel intensity information of the original fluorescence image is reversely corrected to obtain the corrected pixel intensity. By subjecting the corrected pixel intensity to a reasonable range constraint, the distortion of biofluorescence signals and loss of local details caused by overcorrection are eliminated, and the target pixel intensity is obtained. The target fluorescence image is obtained by reconstructing the global image by summarizing the intensities of all the target pixels.

[0013] In one embodiment, based on the spatially stationary statistical distribution of fluorescence signals from similar biological tissues and the target quenching field, the local statistical uniformity, biological signal restoration degree, and quenching artifact residue of the target fluorescence image are calculated. Based on the local statistical uniformity, the biosignal fidelity, and the residual amount of quenching artifacts, the problem region of the target fluorescence image is located; Match the weight correlation between the problem region and the data fidelity term and model regularization term in the joint optimization objective function to determine the optimization direction and adjustment boundary of the weight parameters; Based on the optimization direction and adjustment boundary of the weight parameters, and combined with the organizational type characteristics and corresponding scanning time sequence characteristics of the problem region, the weighting coefficients of the data fidelity term and the model regularization term in the joint optimization objective function are iteratively adjusted to generate an optimized set of weight parameters. The optimized set of weight parameters is updated into the joint optimization objective function of the digital pathology fluorescence image correction process to complete the parameter self-iterative optimization of the correction process.

[0014] Furthermore, to achieve the above objectives, this application also proposes a digital pathological fluorescence image correction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the digital pathological fluorescence image correction method described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the digital pathological fluorescence image correction method as described above.

[0016] This application provides a correction method for digital pathological fluorescence images, which includes: first, downsampling the original fluorescence image to obtain a sampled fluorescence image with a lower resolution; then, solving for the initial path parameters and initial quenching field parameters corresponding to the sampled fluorescence image based on initial scan path parameters and a joint optimization objective function with minimizing the objective function value as the optimization objective; subsequently, upsampling the initial path parameters and mapping them to the image coordinates of the original fluorescence image to obtain the original path parameters; then, solving for the target quenching field corresponding to the original fluorescence image based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function; and finally, performing correction processing on the original fluorescence image based on the target quenching field to obtain a multi-resolution progressive joint optimization correction scheme for the target fluorescence image. This method solves the problem in existing digital pathological fluorescence image quenching correction techniques that directly apply high resolution scan path parameters to the sampled fluorescence image. Global parameter solving on high-resolution full-slice images incurs high computational overhead and low iterative convergence efficiency. Furthermore, it fails to consider the spatially stationary statistical characteristics of fluorescence signals in similar biological tissues and the physical attenuation laws of fluorescence quenching. This results in residual quenching artifacts, pathological signal distortion, and loss of diagnostic details in the corrected images, failing to balance correction efficiency and accuracy. This paper reduces the computational cost of quenching correction for high-resolution pathological fluorescence images, improves the convergence speed and global optimality of solving scanning path temporal parameters and quenching attenuation fields, effectively eliminates scanning time-related fluorescence quenching brightness unevenness artifacts, fully preserves the true fluorescence signal of pathological tissues and key clinical diagnostic details, and improves the signal fidelity, image quality, and clinical diagnostic suitability of the corrected fluorescence images, achieving a synergistic optimization of correction efficiency and accuracy.

[0017] In summary, this application solves the problem of low efficiency in high-resolution fluorescence image correction process by first solving the initial path and quenching field parameters at a coarse scale, and then mapping to a high-resolution fine solution for the target quenching field through a multi-resolution progressive joint optimization scheme. This improves the iterative convergence speed while taking into account both correction accuracy and pathological signal fidelity. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of the digital pathological fluorescence image correction method of this application; Figure 2 This is a flowchart for the correction process of this application; Figure 3 This is a flowchart illustrating the fourth embodiment of the digital pathological fluorescence image correction method of this application; Figure 4 This is a flowchart illustrating the eighth embodiment of the digital pathological fluorescence image correction method of this application; Figure 5 This is a schematic diagram of the structure of the correction device for digital pathological fluorescence images in this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] In related technologies, fluorescence scan images are processed by standard fluorescence reference slide hardware calibration, flat field correction, global or local histogram equalization, and reference region signal standardization. This method, based on a fixed reference or local statistical features of the image, performs forced normalization and equalization correction operations on the original pixel intensity of the fluorescence image, which can easily destroy the quantitative relationship of fluorescence intensity, thereby leading to distortion of the pathological quantitative analysis results.

[0024] This application provides a solution: First, the original fluorescence image is downsampled to obtain a sampled fluorescence image, the resolution of which is lower than that of the original fluorescence image. Then, the initial path parameters and initial quenching field parameters of the sampled fluorescence image are solved based on the initial scan path parameters and the joint optimization objective function, the joint optimization objective function having the objective function value as the optimization objective. Next, the initial path parameters are upsampled and mapped to the image coordinates of the original fluorescence image to obtain the original path parameters. Then, the target quenching field of the original fluorescence image is solved based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function. Finally, the original fluorescence image is corrected based on the target quenching field to obtain the target fluorescence image.

[0025] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as a digital pathology fluorescence image correction device. The following description uses a digital pathology fluorescence image correction device as an example to illustrate this embodiment and the subsequent embodiments.

[0026] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0027] This application provides a method for correcting digital pathological fluorescence images, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the digital pathological fluorescence image correction method of this application.

[0028] In this embodiment, the correction method for the digital pathological fluorescence image includes steps S10 to S50: Step S10: The original fluorescence image is downsampled to obtain a sampled fluorescence image, wherein the resolution of the sampled fluorescence image is lower than that of the original fluorescence image.

[0029] The original fluorescence image is a high-resolution whole-slice immunofluorescence image acquired by a digital pathology scanner. It is the core input data source for this quenching correction process, fully preserving the true fluorescence emission signal of the pathological tissue and the characteristics of quenching artifacts generated during scanning. Examples include full-field digital pathology fluorescence images acquired with a 40x objective lens, single-channel original scan images of multicolor immunofluorescence, original images of frozen section fluorescence scans, and original frame images of live-cell dynamic fluorescence imaging. The sampled fluorescence image is a coarse-scale global fluorescence image with a lower spatial resolution than the original image, obtained by downsampling the original fluorescence image. It fully preserves the global fluorescence brightness distribution, tissue contour features, and brightness gradient information related to the scanning direction of the original image, serving as the core input for subsequent global correction parameter pre-solution. Examples include 2x downsampled coarse-scale fluorescence images, 4x downsampled global feature images, multi-scale pyramid low-level fluorescence images, and tissue foreground global downsampled feature images.

[0030] In this embodiment, the downsampling process of the original fluorescence image can be initiated in six ways. First, it is automatically triggered upon image acquisition completion. After the digital pathology scanner completes the full-field fluorescence image acquisition of a single pathological slide and writes the original image completely into the system storage, the downsampling process is automatically initiated to generate a sampled fluorescence image for subsequent calibration parameter pre-solution. Second, it is triggered by a calibration task instruction. When the system receives a user-initiated digital pathology fluorescence image quenching calibration task instruction, it automatically reads the target original fluorescence image, initiates the downsampling process, and generates a sampled fluorescence image. Third, it is triggered by batch archiving preprocessing. When the system performs a batch archiving task for pathological slide images, it automatically initiates the downsampling process for each original fluorescence image to be archived, generating corresponding sampled fluorescence images in batches for subsequent batch calibration preprocessing. Fourth, it is triggered before AI analysis. When the system receives a pathological fluorescence image AI intelligent diagnostic analysis task, it first performs downsampling processing on the original fluorescence image to generate a sampled fluorescence image for pre-analysis and calibration parameter solving, ensuring the quality of the input image for subsequent AI analysis. Fifth, batch scanning triggering: After the scanner completes the acquisition of multiple pathological slide images of the same batch, with the same staining scheme and the same scanning parameters, the system automatically initiates batch downsampling processing for all original fluorescence images within the batch, generating corresponding sampled fluorescence images to provide parameter benchmarks for unified calibration within the same batch. Sixth, image quality anomaly triggering: When the system detects scan-time-related brightness unevenness or fluorescence quenching artifacts in the original fluorescence images through pre-calibration, it automatically initiates the downsampling processing flow to generate sampled fluorescence images for subsequent calibration parameter solving.

[0031] Once the digital pathology fluorescence image correction device receives the corresponding trigger command, it begins to read the metadata of the original fluorescence image, including image spatial resolution, pixel depth, scan path parameters, staining type information, and initiates the downsampling process.

[0032] For example, there are two methods for downsampling the original fluorescence image and generating the sampled fluorescence image. The first is global proportional downsampling without discrimination. Before the downsampling process officially starts, the full-size resolution of the original fluorescence image is read. Based on the preset fixed downsampling ratio, the target resolution of the sampled fluorescence image is calculated. Then, global proportional downsampling is performed on the entire image area of ​​the original fluorescence image using a bicubic interpolation algorithm. At the same time, the fluorescence intensity values ​​of the entire image are linearly normalized to completely preserve the global brightness distribution, tissue contour features, and brightness gradient information of the scanning direction of the original image. Finally, a sampled fluorescence image with the same aspect ratio as the original image but a lower resolution is generated. This method has a simple and stable downsampling logic, no regional discrimination processing, and the generated sampled image has good global consistency. It is not prone to feature loss and can provide uniform and stable input for subsequent global parameter optimization. It is suitable for the correction scenario of conventional pathological fluorescence images with uniform tissue distribution and single marker staining.

[0033] The second method is tissue-feature-guided differential weighted downsampling. Before the downsampling process officially starts, the original fluorescence image is first segmented into foreground and background regions to generate a foreground mask. This identifies foreground regions containing pathological tissue information and background regions lacking effective information. Differential downsampling weights are then assigned to different regions: for key foreground regions containing core diagnostic information such as tumor tissue and positive biomarkers, low-magnification downsampling is used to maximize the preservation of detailed features and fluorescence distribution information in the core diagnostic region. For non-key foreground regions such as normal tissue, medium-magnification downsampling is used to balance feature preservation and computational load. For background regions, high-magnification downsampling is used to reduce the computational load in invalid regions. Based on the differential downsampling weights for different regions, a region-adaptive interpolation algorithm is used to perform downsampling, while boundary smoothing fusion is performed on the sampling results of different regions. Finally, a sampled fluorescence image with a lower resolution than the original image but fully preserving the core diagnostic features is generated. This method employs tissue-feature-guided differential weight downsampling, which reduces image resolution and subsequent global optimization computation while maximizing the preservation of feature information in key areas of pathological diagnosis. It avoids the loss of core diagnostic details caused by global proportional downsampling, and solves the pain point of conventional downsampling methods where reducing computational load and preserving key features cannot be balanced. It is suitable for clinical diagnostic pathological fluorescence image correction scenarios involving multiple marker staining and complex tissue types.

[0034] In one alternative approach, the system's preset fixed downsampling ratio and interpolation algorithm are first invoked. Following a preset global proportional downsampling rule, the original fluorescence image is downsampled to directly generate a sampled fluorescence image of the corresponding resolution. This method relies on the system's preset standardized rules, offers fast execution speed, intuitive logic, and is well-suited for the rapid preprocessing needs of routine batch pathological images.

[0035] In another alternative approach, the staining type, scanning mode, and tissue complexity features of the original fluorescence image are first extracted. Simultaneously, a pathological image downsampling rule library corresponding to the staining type and scanning mode is retrieved. Based on this rule library, a differentiated downsampling strategy and interpolation algorithm adapted to the current image are dynamically generated. Then, downsampling processing is performed on the original fluorescence image based on the dynamically generated strategy to generate a sampled fluorescence image. The feature integrity of the generated sampled image is verified to ensure that no core tissue features are lost. This method can dynamically generate an adapted downsampling strategy based on the image's own features, resulting in higher adaptability and feature preservation. It can fully exploit the feature distribution characteristics of different pathological images and meet the high-precision preprocessing requirements of complex staining and scanning scenarios.

[0036] In an exemplary scheme for determining the sampling fluorescence image, the original high-resolution digital pathological fluorescence image to be corrected is first loaded, and the image's metadata, including image spatial resolution, pixel depth, scan path parameters, staining type, and objective magnification, is read simultaneously. Image preprocessing is then performed: Gaussian filtering is applied to the original fluorescence image to remove random noise, and Otsu's thresholding method is used to segment the tissue foreground and blank background, generating a tissue foreground mask to mark the core diagnostic region and non-critical regions. Next, a downsampling strategy is determined. Based on the tissue complexity of the image and the proportion of the core diagnostic region, a global downsampling baseline ratio of 4 is determined, while a differential downsampling weight of 2 is set for the core diagnostic region and an differential downsampling weight of 8 is set for the blank background region. Subsequently, adaptive downsampling of the region is performed. For the core diagnostic region, a bicubic interpolation algorithm is used to perform 2x downsampling to maximize the preservation of fluorescence intensity distribution and detailed features within the region. For normal tissue regions, a bilinear interpolation algorithm is used to perform 4x downsampling to balance feature preservation and computational cost. For blank background regions, a nearest neighbor interpolation algorithm is used to perform 8x downsampling to reduce unnecessary computation. Next, boundary smoothing fusion is performed on the downsampling results from different regions to eliminate abrupt boundary changes caused by resolution differences between regions. Simultaneously, global brightness normalization is performed on the fused image to fully preserve the brightness gradient information along the scanning direction of the original image. Finally, feature integrity verification is performed on the generated sampled fluorescence image to confirm that the global fluorescence distribution, tissue contours, and core diagnostic region features of the original image are fully preserved. After successful verification, the final sampled fluorescence image is generated for subsequent initial parameter solving.

[0037] It should be noted that in some special cases, the resolution of the original fluorescence image is lower than the preset sampling image resolution threshold, so there is no need to perform downsampling processing. In this case, the original fluorescence image is directly used as the sampled fluorescence image for subsequent initial parameter solving.

[0038] Step S20: Solve for the initial path parameters and initial quenching field parameters of the sampled fluorescence image based on the initial scan path parameters and the joint optimization objective function, wherein the joint optimization objective function aims to minimize the objective function value.

[0039] The initial scan path parameters are system-preset initial scan timing parameters that match the scanner's acquisition path of the original fluorescence image. They include the initial mapping rules for scan direction, scan line switching timing, and pixel-level scan relative time, and serve as the initial values ​​for iterative global parameter optimization. Examples include the scanner's default raster scan path parameters, serpentine scan path initial parameters, block scan path timing parameters, and custom scan path initial mapping parameters.

[0040] The joint optimization objective function is a bivariate collaborative optimization loss function used to simultaneously optimize the temporal parameters of the scan path and the physical parameters of the quenching attenuation field. It consists of a weighted sum of data fidelity terms and model regularization terms and is the core quantification criterion for parameter solving. Examples include objective functions constrained by the uniformity of signals from similar tissues, objective functions constrained by the physical laws of fluorescence quenching, bivariate collaborative global optimization objective functions, and multi-constraint weighted fusion objective functions.

[0041] The initial path parameters are globally optimal scanning path temporal parameters obtained through iterative solution of a joint optimization objective function in the sampled fluorescence image dimension. These parameters form the core foundation for subsequent upsampling mapping to a high-resolution dimension. Examples include coarse-scale globally optimal scanning temporal parameters, optimal scanning path mapping parameters in the sampled image dimension, global scanning relative time distribution parameters, and the low-resolution optimal path temporal matrix.

[0042] The initial quenching field parameters are coarse-scale quenching attenuation field parameters that conform to the physical laws of fluorescence quenching, obtained by co-solving with the initial path parameters in the dimension of the sampled fluorescence image. They serve as the initial benchmark for subsequent high-resolution target quenching field solutions. Examples include the coarse-scale global quenching attenuation coefficient matrix, fluorescence attenuation model parameters in the sampled image dimension, low-resolution global quenching field distribution parameters, and the scan time-related attenuation coefficient set.

[0043] The objective function value is the final calculated result of the joint optimization objective function. It represents the weighted sum of the total deviation of the correction effect and the total violation of the physical model. The smaller the value, the better the global optimization effect of the parameter combination. Examples include the calculated value of the global loss function, the loss value of bivariate collaborative optimization, the weighted sum of correction deviations, and the calculated value of the total constraint violation.

[0044] In this embodiment, the initial parameter solving process can be initiated in six ways. First, it can be automatically triggered after the sampled image is generated. After the sampled fluorescence image in step S10 is generated and feature verification is completed, the system automatically initiates the initial path parameter and initial quenching field parameter solving process without manual intervention. Second, it can be manually triggered by the user. After confirming the validity of the sampled fluorescence image through the system's interactive interface, the pathologist or operator manually initiates the parameter solving command, and the system initiates the initial parameter solving process upon receiving the command. Third, it can be triggered by a batch correction task. When the system performs a batch pathological fluorescence image correction task, after each image is downsampled and a sampled fluorescence image is generated, the initial parameter solving process for the corresponding image is automatically initiated in batches. Fourth, it can be triggered by AI pre-analysis. When the system detects obvious fluorescence quenching artifacts in the sampled fluorescence image through AI pre-analysis, the initial parameter solving process is automatically initiated to specifically solve for the appropriate correction parameters. Fifth, batch parameter benchmark triggering: When processing multiple pathological images from the same batch with the same scanning parameters, the system automatically starts the initial parameter solution process after sampling the first image. The solved parameters can be used as the initial benchmark for other images in the same batch. Sixth, calibration process restart triggering: When the system detects that the parameter solution of the previous calibration process has not reached the convergence condition, it automatically restarts the initial parameter solution process, updates the initial values, and executes the solution again.

[0045] After the digital pathology fluorescence image correction device completes the reading and preprocessing of the sampled fluorescence image, it begins to load the basic rules for initializing the scan path parameters and the joint optimization objective function, and starts the solution process for the initial path parameters and the initial quenching field parameters.

[0046] For example, there are two methods for solving the initial path parameters and initial quenching field parameters. The first method is a two-parameter alternating fixed-step-size iterative solution. Before the parameter solution process officially starts, the initial scan path parameters are loaded as the initial values ​​for iteration, and the initial quenching field parameters are initialized as an identity matrix of all 1s. A fixed iteration step size, a maximum number of iterations, and a convergence threshold are set. Then, the alternating iteration process is started. In the first round, the quenching field parameters are kept unchanged, and the goal is to minimize the objective function value. The scan path parameters are iteratively updated, and the updated objective function value is calculated. In the second round, the updated scan path parameters are kept unchanged, and the goal is to minimize the objective function value. The quenching field parameters are iteratively updated, and the updated objective function value is calculated. This process continues, with one set of parameters being fixed and the other set being optimized in each round. The parameter combination and objective function value of each iteration are recorded simultaneously. When the difference between the objective function values ​​of two consecutive iterations is less than the preset convergence threshold, or when the number of iterations reaches the maximum number of iterations, the iteration is terminated. The scan path parameters and quenching field parameters at the final convergence are used as the initial path parameters and initial quenching field parameters of the sampled fluorescence image, respectively. This method employs alternating fixed-step iteration with two parameters. The iteration logic is simple and stable, the convergence process is controllable, and it is not prone to iteration oscillation. It can quickly find the globally optimal parameter combination and is suitable for solving parameters in conventional pathological fluorescence images with uniform tissue distribution and simple artifact features.

[0047] The second approach is a bivariate collaborative adaptive step-size global optimization solution. Before the formal parameter solution process begins, tissue feature extraction is performed on the sampled fluorescence images to locate the core diagnostic region, normal tissue region, and blank background region. Differentiated weights are set for the two terms of the jointly optimized objective function: the weight of the data fidelity term in the core diagnostic region is increased to strengthen signal fidelity constraints; the weight of the model regularization term in the blank background region is increased to strengthen physical law constraints. Then, the initial scan path parameters are used as the initial values ​​for iteration, and the quenching field parameters are initialized as an initial matrix conforming to the monotonically decreasing fluorescence quenching law. Adaptive iteration step-size rules, maximum number of iterations, and convergence threshold are set. Simultaneously, a global optimization mechanism based on simulated annealing algorithm is constructed to avoid iteration getting trapped in local optima. The bivariate collaborative iteration process is then initiated. In each iteration, the two sets of variables, scan path parameters and quenching field parameters, are updated synchronously. Based on the change in the objective function value of the current iteration, the iteration step-size is adaptively adjusted: when the objective function value decreases rapidly, the iteration step-size is increased to accelerate convergence; when the decrease in the objective function value slows down, the iteration step-size is decreased to improve solution accuracy. Simultaneously, through simulated annealing, parameter combinations that slightly increase the objective function value are accepted with a certain probability, thus escaping local optima. The parameter combinations and objective function values ​​of each iteration are recorded synchronously. When the objective function value consistently falls below a preset convergence threshold for multiple iterations, or when the maximum number of iterations is reached, the iteration terminates. The scanning path parameters and quenching field parameters corresponding to the minimum objective function value during the entire iteration process are used as the initial path parameters and initial quenching field parameters for the sampled fluorescence image, respectively. This method employs bivariate collaborative adaptive step-size global optimization, combining the convergence efficiency advantage of adaptive step-size with the global optimization capability of simulated annealing. Furthermore, differentiated weight settings enhance the optimization effect in the core diagnostic region, addressing the pain points of conventional alternating iterative methods, such as being prone to getting trapped in local optima, failing to balance convergence speed and solution accuracy, and insufficient optimization of the core diagnostic region. It is suitable for solving pathological fluorescence image parameter problems involving multiple marker staining, complex artifact features, and high clinical diagnostic requirements.

[0048] In one alternative approach, the system first retrieves the pre-defined joint optimization objective function with fixed weight parameters and iteration rules. Using the initial scan path parameters as initial values, the scan path parameters and quenching field parameters are updated round by round according to the pre-defined alternating iteration rules. The goal is to minimize the objective function value. After iterative convergence, the initial path parameters and initial quenching field parameters are obtained. This method relies on the system's pre-defined standardized rules, has simple execution logic, fast convergence speed, and is suitable for the rapid parameter solving requirements of routine batch pathological images.

[0049] In another alternative approach, tissue type features, artifact features, and scanning mode features of the sampled fluorescence image are first extracted. Simultaneously, a historical optimized parameter library for the corresponding scene is retrieved. Based on the historically optimal parameters, a joint optimization objective function weight, iteration step size, and convergence rule adapted to the current image are dynamically generated. Then, based on the dynamically generated rules, a bivariate collaborative iterative solution is performed, while simultaneously verifying the parameter validity of each iteration to ensure that the solved quenching field parameters conform to the physical laws of fluorescence quenching. Finally, the initial path parameters and initial quenching field parameters are obtained. This method can dynamically generate suitable solution rules based on image features and historical data, resulting in higher adaptability and accuracy in parameter solving. It can fully exploit the artifact features of different pathological images and meet the high-precision parameter solving requirements of complex scenarios.

[0050] In an exemplary scheme for determining initial path parameters and initial quenching field parameters, the sampled fluorescence image generated in step S10, the corresponding tissue foreground mask, and the core diagnostic region marking results are first loaded. Simultaneously, the scanner initialization scan path parameters and the basic rules for jointly optimizing the objective function are read from the original image. Then, the objective function differential weight configuration is executed. Based on the division of the core diagnostic region, normal tissue region, and blank background region, differential weights are set for the data fidelity term and model regularization term of the jointly optimized objective function: the weight of the data fidelity term in the core diagnostic region is set to 0.7, and the weight of the model regularization term is set to 0.3. Both weights in the normal tissue region are set to 0.5. The weight of the data fidelity term in the blank background region is set to 0.2, and the weight of the model regularization term is set to 0.8. Next, iteration rules are set, using the initialized scan path parameters as the initial value for scan path iteration, and a matrix conforming to the monotonically decreasing fluorescence quenching law as the initial value for the quenching field parameters. The initial iteration step size is set to 0.1, with an adaptive step size adjustment rule, a maximum number of iterations of 100, and a convergence threshold of 1e-6. Simultaneously, a simulated annealing global optimization mechanism is enabled. The bivariate collaborative iterative solution process is then initiated. In each iteration, the scan path parameters and quenching field parameters are updated synchronously, the corresponding objective function value is calculated, and compared with the objective function value of the previous iteration. The iteration step size is adaptively adjusted for the next iteration. Simultaneously, a simulated annealing mechanism is used to determine whether to accept the current parameter combination, avoiding getting trapped in local optima. The parameter combinations and objective function values ​​of each iteration are recorded synchronously, and the optimal parameter combination with the smallest objective function value is marked. The iteration loop terminates when the difference in objective function values ​​for five consecutive iterations is less than a preset convergence threshold, or when the number of iterations reaches 100. Parameter validity is then verified to confirm that the initial path parameters obtained match the scan timing logic, and that the initial quenching field parameters conform to the physical law of monotonically decreasing fluorescence quenching, with no outliers. After successful verification, the scan path parameters corresponding to the minimum objective function value during the iteration process are determined as the initial path parameters, and the corresponding quenching field parameters are determined as the initial quenching field parameters, for subsequent upsampling mapping processing.

[0051] It should be noted that in some special cases, the objective function value may oscillate continuously during the iterative solution process, failing to meet the preset convergence condition. In such cases, the initial iteration values ​​will be automatically reset, the iteration step size and objective function weights will be adjusted, and the solution process will be restarted. If convergence still fails after multiple restarts, the initial scan path parameters will be used as the initial path parameters, and the initial quenching field matrix conforming to physical laws will be used as the initial quenching field parameters. At the same time, abnormal situations will be recorded for subsequent rule optimization.

[0052] Step S30: Upsample and map the initial path parameters to the image coordinates of the original fluorescence image to obtain the original path parameters.

[0053] Upsampling mapping is the process of matching the initial path parameters obtained by solving the dimensions of a low-resolution sampled image to the spatial mapping of the original high-resolution fluorescence image in the pixel coordinate system through proportional coordinate mapping and interpolation. It is the core step in realizing the transformation from coarse-scale parameters to fine-scale parameters. Examples include proportional coordinate upsampling mapping, tissue feature-guided interpolation mapping, pixel-level temporal parameter mapping, and multi-scale spatial coordinate transformation.

[0054] The original path parameters are high-resolution scan path temporal parameters that, after upsampling and mapping, perfectly match the pixel coordinate system of the original high-resolution fluorescence image. They contain the scan relative time mapping rules for each pixel in the original image and serve as the core temporal reference for subsequent high-resolution target quenching field solutions. Examples include high-resolution pixel-level scan temporal parameters, original image dimension-optimal scan path mapping parameters, full-image pixel-level scan relative time matrix, and high-resolution scan sequence field reference parameters.

[0055] In this embodiment, after the initial path parameters and initial quenching field parameters are solved and pass validity verification, the system automatically starts the upsampling mapping process for the initial path parameters. Alternatively, it can be triggered in six ways: First, automatic triggering after parameter solving: After the initial path parameters in step S20 are solved and pass validity verification, the system automatically starts the upsampling mapping process to generate the original path parameters. Second, manual triggering: After confirming the validity of the initial path parameters through the system's interactive interface, the operator manually initiates the upsampling mapping command, and the system starts the mapping process upon receiving the command. Third, batch task process triggering: When executing a batch correction task, after solving the initial parameters for a single image, the upsampling mapping process is automatically started in pipeline order to ensure continuous execution of batch tasks. Fourth, high-resolution solution pre-triggered triggering: When the system is about to start the target quenching field solution process in the original image dimension, it automatically checks whether there are matching original path parameters; if not, the upsampling mapping process is automatically started. Fifth, parameter reuse triggering within the same batch: when processing multiple pathological images of the same specifications within the same batch, after solving the initial path parameters for the first image, the upsampling mapping process is automatically started, and the generated original path parameters can be used as the mapping benchmark for other images in the same batch. Sixth, parameter optimization iteration triggering: after the initial path parameters in step S20 are iteratively optimized and updated, the system automatically restarts the upsampling mapping process, updates the corresponding original path parameters, and ensures parameter consistency.

[0056] After the digital pathology fluorescence image correction device completes the reading and validity verification of the initial path parameters, it begins to load the image coordinate information and resolution ratio parameters of the original fluorescence image and starts the upsampling mapping process.

[0057] For example, the upsampling mapping of the initial path parameters and the generation of the original path parameters include two methods. The first is a global proportional linear upsampling mapping. Before the upsampling mapping process officially starts, the resolution ratio coefficient between the sampled fluorescence image and the original high-resolution fluorescence image is calculated to determine the scaling factor of the coordinate mapping. Then, based on this scaling factor, global proportional linear upsampling is performed on the scan timing matrix in the initial path parameters. A bilinear interpolation algorithm is used to map the low-resolution timing parameters to the high-resolution pixel coordinate system of the original fluorescence image, generating a pixel-level scan timing parameter matrix that perfectly matches the size of the original image. At the same time, global linear correction is performed on the mapped timing parameters to ensure that the relative scanning time after mapping is completely consistent with the scanning timing logic of the original scanner, without timing misalignment or boundary abrupt changes, and finally generating original path parameters that perfectly match the image coordinates of the original fluorescence image. This method employs a global proportional linear upsampling mapping, which has a simple and stable mapping logic, fast calculation speed, good global consistency of temporal parameters, no temporal misalignment between regions, and can quickly generate high-resolution path parameters that match the original image. It is suitable for conventional pathological fluorescence image mapping scenarios with uniform tissue distribution and regular scanning paths.

[0058] The second method is nonlinear order-preserving upsampling mapping with tissue boundary constraints. Before the upsampling mapping process officially starts, the tissue segmentation results and tissue boundary masks of the original fluorescence image are read. At the same time, the resolution scaling factor between the sampled image and the original image is calculated to determine the base scaling factor. Then, differentiated upsampling is performed on the scan time series matrix in the initial path parameters: for homogeneous regions within the tissue, a bicubic interpolation algorithm is used to perform upsampling to ensure the smooth continuity of time series parameters within the region; for tissue boundary regions, a nonlinear order-preserving interpolation model with boundary constraints is constructed to strictly maintain the monotonic order of the scan time series during the mapping process, while strengthening the time series gradient characteristics of the boundary regions to avoid boundary time series blurring and tissue feature loss caused by linear interpolation; for blank background regions, a nearest neighbor interpolation algorithm is used to perform upsampling to reduce the amount of computation while ensuring time series consistency. Then, boundary smoothing fusion is performed on the mapping results of different regions, and global order-preserving verification is performed on the temporal parameters of the fused full-image to ensure that the relative scanning time after mapping completely conforms to the temporal logic of the original scanning path, without any order reversal or temporal abrupt change. Finally, original path parameters with image coordinates that perfectly match the original fluorescence image and accurate temporal features are generated. This method adopts nonlinear order-preserving upsampling mapping with tissue boundary constraints. While achieving the mapping from low-resolution parameters to high-resolution parameters, it ensures the smoothness of the temporal sequence within the tissue and avoids the blurring of the temporal features of the tissue boundary through regional differential interpolation and order-preserving constraints. It strictly maintains the physical logic of the scanning temporal sequence and solves the pain points of boundary temporal distortion, tissue feature loss, and insufficient mapping accuracy caused by conventional linear upsampling. It is suitable for clinical diagnostic pathological fluorescence image mapping scenarios with complex tissue types and fine boundary structures.

[0059] In one alternative approach, the system's preset global proportional mapping rules and interpolation algorithm are first invoked. Based on the resolution ratio between the sampled image and the original image, a global linear upsampling mapping is performed on the initial path parameters to directly generate original path parameters that match the coordinates of the original image. This method is fast, logically simple, and suitable for the rapid mapping processing needs of routine batch pathological images.

[0060] In another alternative approach, the tissue distribution features, boundary complexity features, and scanning path features of the original fluorescence image are first extracted. Simultaneously, a mapping rule library corresponding to the scene is retrieved. Based on this rule library, a region-specific interpolation strategy and order-preserving constraint rules adapted to the current image are dynamically generated. Then, upsampling mapping is performed on the initial path parameters based on the dynamically generated strategy. Simultaneously, temporal consistency verification is performed on the mapped original path parameters to ensure complete matching between the parameters and the original scanning logic. This method can dynamically generate an adapted mapping strategy based on image features, resulting in higher mapping accuracy, more complete preservation of temporal features, and adaptability to the high-precision mapping requirements of complex pathological images.

[0061] In an exemplary scheme for determining the original path parameters, the initial path parameters obtained in step S20, the resolution information of the sampled fluorescence image, and the image coordinate information, tissue segmentation results, and tissue boundary mask of the original high-resolution fluorescence image are first loaded. Then, a resolution scaling factor is calculated. Based on the width and height pixel counts of the sampled image and the original image, a base scaling factor of 4 is calculated to determine the basic rules for coordinate mapping. Next, a region-specific mapping strategy is configured. Based on the tissue segmentation results, the original image is divided into three categories: homogeneous regions within the tissue, tissue boundary regions, and blank background regions. Differentiated interpolation algorithms and constraint rules are configured for different regions: a bicubic interpolation algorithm is used for homogeneous regions within the tissue, with smoothness constraints configured. A nonlinear order-preserving interpolation algorithm is used for tissue boundary regions, with temporal monotonicity-preserving constraints and boundary feature enhancement rules configured. A nearest neighbor interpolation algorithm is used for blank background regions, with temporal consistency constraints configured. Then, a region-specific upsampling mapping is performed. For each of the three types of regions, the corresponding interpolation algorithm is used. Based on the base scaling factor of 4, the temporal matrix of the initial path parameters is mapped to the high-resolution pixel coordinate system of the original image, generating high-resolution temporal parameters for the corresponding regions. Next, boundary smoothing fusion is performed on the mapping results of different regions to eliminate abrupt changes in temporal parameters between regions, generating a high-resolution temporal parameter matrix for the entire map. Then, a global temporal consistency check is performed to verify whether the mapped temporal parameters conform to the monotonic temporal logic of the original scan path, and to check for issues such as temporal reversal, outliers, and boundary abrupt changes. Local corrections are performed on regions that do not meet the requirements. After the check passes, the final high-resolution temporal parameter matrix is ​​determined as the original path parameters for subsequent target quenching field solving.

[0062] It should be noted that in some special cases, the aspect ratios of the original fluorescence image and the sampled fluorescence image are inconsistent, making it impossible to perform proportional mapping. In such cases, size adaptation correction is first performed on the initial path parameters, and then upsampling mapping is performed to ensure that the generated original path parameters completely match the image coordinates of the original fluorescence image.

[0063] Step S40: Solve for the target quenching field of the original fluorescence image based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function.

[0064] The target quenching field, also known as the target quenching attenuation field, is a two-dimensional matrix that perfectly matches the size of the original high-resolution fluorescence image. Each pixel location stores a corresponding fluorescence quenching attenuation coefficient, and its attenuation coefficients conform to the physical attenuation law of fluorescence quenching. It is used to characterize the degree of fluorescence signal quenching at each pixel location and is the core benchmark for pixel intensity correction of the original image. Examples include pixel-level high-resolution quenching attenuation fields, region-fitted continuous attenuation fields, multi-channel fluorescence matching attenuation fields, and scan-time-correlated global quenching coefficient matrices.

[0065] In this embodiment, after the original path parameters are generated and pass validity verification, the system automatically starts the target quenching field solution process. Alternatively, it can be triggered in six ways: First, automatic triggering after the original path parameters are generated: After the original path parameters in step S30 are generated and pass timing consistency verification, the system automatically starts the target quenching field solution process. Second, manual triggering: After the operator confirms the validity of the original path parameters through the system interface, they manually initiate the target quenching field solution command, and the system starts the solution process after receiving the command. Third, batch correction pipeline triggering: When performing batch pathological image correction tasks, after completing the path parameter mapping for a single image, the target quenching field solution process is automatically started in pipeline order to ensure continuous execution of batch tasks. Fourth, image correction pre-triggered triggering: When the system is about to start the correction process for the original fluorescence image, it automatically detects whether a matching target quenching field exists; if not, the solution process is automatically started. Fifth, parameter iteration optimization is triggered. After the original path parameters or initial quenching field parameters have been iteratively updated, the system automatically restarts the target quenching field solution process, updates the corresponding target quenching field, and ensures parameter consistency. Sixth, correction quality optimization is triggered. When the system detects artifact residue, signal distortion, or other problems in the image after the previous correction, it automatically restarts the target quenching field solution process, optimizes and adjusts the solution rules, and then executes the solution again.

[0066] After the digital pathology fluorescence image correction device completes the reading and validity verification of the original path parameters and initial quenching field parameters, it begins to load the constraint rules of the joint optimization objective function and starts the solution process for the target quenching field.

[0067] For example, there are two methods for solving the target quenching field. The first is a single-step fitting solution with globally unified constraints. Before the formal solution process starts, the original path parameters are converted into a high-resolution scanning sequence field that matches the original image. At the same time, the initial quenching field parameters are upsampled and mapped to the pixel coordinate system of the original image to generate a high-resolution initial quenching attenuation matrix. Then, with the minimization of the joint optimization objective function as the core criterion, a globally unified constraint on the physical attenuation law of fluorescence quenching is set. The pixel-level scanning time sequence of the scanning sequence field is used as the independent variable, and the pixel intensity value of the original image is used as the dependent variable. A global single exponential attenuation model is used to perform a unified nonlinear least squares fitting across the entire image to obtain the unified attenuation model parameters. Based on the obtained model parameters and the scanning time sequence of each pixel, the quenching attenuation coefficient corresponding to each pixel is calculated to generate a target quenching field covering the entire image. At the same time, based on the constraint rules of the joint optimization objective function, a global smoothness and monotonicity check is performed on the generated target quenching field to ensure that the attenuation field conforms to the physical law of fluorescence quenching. This method employs a single-step fitting solution with globally unified constraints. The solution logic is simple and stable, the calculation speed is fast, there are no abrupt changes in the attenuation coefficient between regions, and the generated target quenching field has good global consistency. It is suitable for routine pathological fluorescence image solving scenarios with uniform tissue distribution, single marker staining, and consistent fluorescence quenching characteristics.

[0068] The second approach involves a region-specific, differentiated constraint-based collaborative solution guided by homogeneous tissue regions. Before the formal solution process begins, the original path parameters are converted into a high-resolution scan sequence field, and the initial quenching field parameters are upsampled and mapped to the original image coordinate system, generating a high-resolution initial quenching attenuation matrix. Then, unsupervised segmentation using multi-feature fusion is performed on the original fluorescence image. Combining the image's fluorescence intensity features, texture features, and temporal gradient features of the scan sequence field, multiple image regions with homogeneous tissue types and stable fluorescence statistical characteristics are obtained. Simultaneously, the tissue type and diagnostic importance level of each region are labeled, generating multiple independent constraint units that are computationally independent. Using the global minimization of the joint optimization objective function as the core criterion, differentiated optimization constraint rules are set for each independent constraint unit: For tumor tissue regions with high diagnostic importance, data fidelity constraints are strengthened to improve the accuracy of attenuation coefficient calculation and ensure signal fidelity in core diagnostic regions. For normal tissue regions, balanced constraint rules are adopted to balance solution accuracy and smoothness. For blank background regions, model regularization constraints are strengthened to ensure the smoothness and physical compliance of the attenuation field. Subsequently, parallel iterative solutions are initiated simultaneously for all constraint units. Within each constraint unit, the attenuation model parameters adapted to that region are fitted by combining the scanning time sequence characteristics, fluorescence intensity characteristics, and differentiated constraint rules of the corresponding region, and the pixel-level quenching attenuation coefficient within the region is calculated. After all constraint units are solved, attenuation coefficient smoothing processing is performed on the boundary regions of adjacent units to eliminate boundary abrupt changes. At the same time, global verification is performed based on the joint optimization objective function to ensure that the attenuation field of the entire image conforms to the physical laws of fluorescence quenching, ultimately generating a high-resolution target quenching field that covers the entire image and is continuous across the entire region. This approach adopts a regional differentiated constraint collaborative solution guided by tissue homogeneity regions. It can set appropriate solution rules for the fluorescence characteristics and diagnostic value of different tissue types, reducing the computational load of high-resolution global solutions while improving the solution accuracy of core diagnostic regions. It solves the industry pain points of conventional global unified solution methods, such as insufficient fidelity of key diagnostic information, low efficiency of large-size high-resolution image solution, and inability to take into account the differentiated needs of different tissue regions. It is suitable for high-resolution pathological fluorescence image solution scenarios for clinical diagnosis with multiple marker staining and complex tissue types.

[0069] In one alternative approach, the system's preset global unified attenuation model and constraint rules are first retrieved. Based on the scanning sequence field generated from the original path parameters, a global fitting solution is performed on the original fluorescence image to directly generate the target quenching field. This method is fast, logically simple, and suitable for the rapid solution requirements of routine batch pathological images.

[0070] In another alternative approach, tissue type features, staining type features, and scanning time sequence features of the original fluorescence image are first extracted. Simultaneously, the attenuation model library and constraint rule library for the corresponding scene are retrieved. Based on the rule library, a region-specific solution strategy, differentiated constraint rules, and a suitable attenuation model adapted to the current image are dynamically generated. Then, region-specific collaborative solving is performed based on the dynamically generated strategy. Global physical compliance and signal fidelity verification are performed on the obtained target quenching field to ensure that the solution results meet the requirements of the joint optimization objective function. This method can dynamically generate suitable solution strategies based on image features, resulting in higher solution accuracy, stronger adaptability, and suitability for the high-precision solution requirements of complex pathological images.

[0071] In an exemplary scheme for determining the target quenching field, the original path parameters generated in step S30, the initial quenching field parameters obtained in step S20, the original high-resolution fluorescence image, and the basic constraint rules for jointly optimizing the objective function are first loaded. Then, high-resolution benchmark generation is performed, converting the original path parameters into a high-resolution scanning sequence field that perfectly matches the size of the original image, obtaining a full-image pixel-level scanning relative time matrix; simultaneously, the initial quenching field parameters are subjected to proportional upsampling mapping to generate a high-resolution initial quenching attenuation matrix matching the original image. Next, tissue homogeneous region segmentation is performed. An unsupervised clustering algorithm is used, combining the fluorescence intensity features and texture features of the original image, as well as the temporal gradient features of the scanning sequence field, to divide the tissue foreground region of the original image into 6 homogeneous tissue sub-regions. Each sub-region is labeled with its tissue type and diagnostic importance level, generating independent constraint units, while retaining blank background regions as independent constraint units. Subsequently, differentiated constraint configuration is implemented. Based on the diagnostic importance level of each constraint unit, differentiated weights are set for the data fidelity term and model regularization term of the joint optimization objective function. Simultaneously, an appropriate attenuation model is matched for each constraint unit: a double exponential attenuation model is matched for the core region of tumor tissue to strengthen the data fidelity constraint; a single exponential attenuation model is matched for the normal tissue region with balanced weights; and a linear attenuation model is matched for the blank background region to strengthen the model regularization constraint. Multi-unit parallel solution is then initiated, with parallel iterative solutions started simultaneously for all constraint units. Within each constraint unit, using the temporal data of the scan sequence field as the independent variable, the pixel intensity of the original image as the dependent variable, and minimizing the joint optimization objective function as the criterion, the attenuation model parameters for the corresponding region are fitted and solved, calculating the quenching attenuation coefficient for each pixel within the region. After all constraint units are solved, attenuation coefficient smoothing fusion is performed on the boundary regions of adjacent constraint units, using a linear transition algorithm to eliminate boundary abrupt changes, generating a continuous quenching attenuation coefficient matrix across the entire image. Subsequently, a global verification is performed. Based on the constraints of the joint optimization objective function, the monotonicity, smoothness, and physical compliance of the full-image attenuation coefficients are verified. Simultaneously, the signal fidelity of the core diagnostic region is verified, and local corrections are performed on regions that do not meet the requirements. After successful verification, the final full-image quenching attenuation coefficient matrix is ​​determined as the target quenching field for subsequent original image correction.

[0072] It should be noted that in some special cases, the fitting solution for certain tissue regions may not reach the convergence condition. In such cases, the constraint rules and attenuation model for that region will be automatically adjusted, and the local solution will be re-executed. If convergence still fails after multiple adjustments, the attenuation coefficient obtained from the global fitting will be used to fill the region, and abnormal situations will be recorded for subsequent rule optimization.

[0073] Step S50: Correct the original fluorescence image according to the target quenching field to obtain the target fluorescence image.

[0074] The target fluorescence image is a standardized digital pathological fluorescence image that eliminates scanning time-related fluorescence quenching brightness unevenness artifacts, fully restores the true fluorescence signal distribution of pathological tissue, and preserves tissue boundaries and cellular details. It is the final output of the calibration process and can be directly used by pathologists for slide reading and diagnosis, intelligent AI analysis, quantitative statistics, and archiving. Examples include single-marker immunofluorescence calibrated images, multicolor immunofluorescence multichannel calibrated images, full-field scanning pathological slide calibrated images, and clinically diagnostic standardized fluorescence pathological images.

[0075] In this embodiment, after the target quenching field is solved and passes validity verification, the system automatically initiates the correction process for the original fluorescence image. Alternatively, it can be triggered in six ways: First, automatic triggering upon completion of target quenching field generation: After the target quenching field solution in step S40 is completed and passes global verification, the system automatically initiates the correction process for the original fluorescence image, generating the target fluorescence image. Second, manual triggering: After the pathologist or operator confirms the validity of the target quenching field through the system's interactive interface, they manually initiate an image correction command, and the system initiates the correction process upon receiving the command. Third, batch archiving task triggering: When executing a batch archiving task for pathological images, after solving the target quenching field for a single image, the correction process automatically initiates, generating a standardized target fluorescence image for archiving. Fourth, AI diagnosis pre-processing triggering: When the system receives an AI intelligent diagnosis task for pathological images, after solving the target quenching field, the correction process automatically initiates, generating a target fluorescence image that meets the AI ​​analysis quality requirements. Fifth, batch calibration triggering: When processing multiple pathological images from the same batch with the same scanning parameters, after solving the target quenching field of a single image, the calibration process for the corresponding images is automatically initiated in batches to generate target fluorescence images. Sixth, calibration quality optimization triggering: When the system detects quality problems in the target fluorescence image after the initial calibration, the calibration process is automatically restarted, and calibration is performed again based on the optimized target quenching field to improve image quality.

[0076] Once the digital pathology fluorescence image correction device completes the reading and matching verification of the target quenching field and the original fluorescence image, it initiates the pixel-level correction process for the original fluorescence image.

[0077] For example, there are two methods for correcting the original fluorescence image based on the target quenching field and generating the target fluorescence image. The first method is a pixel-by-pixel linear inverse compensation serial correction. Before the correction process officially starts, the full-image pixel intensity information of the original high-resolution fluorescence image is read, and the full-image quenching attenuation coefficient data of the target quenching field is read simultaneously. According to the row priority order of the original image, the quenching attenuation coefficient at the same spatial position is matched for each pixel, completing the one-to-one binding of the full-image pixels and attenuation coefficients. Then, starting from the first row of the image, the serial correction calculation is performed row by row and pixel by pixel. Each single pixel is treated as an independent operation unit. Based on the principle of fluorescence quenching inverse compensation, the corrected pixel intensity is calculated using the linear inverse correction formula: Corrected pixel intensity = Original pixel intensity value / Corresponding quenching attenuation coefficient. After the correction calculation of a single pixel is completed, a reasonable range constraint is applied to the corrected pixel intensity. Intensity values ​​that exceed the maximum range of gray values ​​in the original image are truncated, and abnormal intensity values ​​less than 0 are corrected to 0 to obtain the target intensity value of the pixel. After the entire process of a single pixel is completed, the correction calculation for the next pixel begins. After all pixels in the entire image have been corrected, the complete global image is reconstructed by summing all target intensity values ​​according to the pixel coordinate system of the original image, and finally, the target fluorescence image is generated. This method uses pixel-by-pixel linear inverse compensation and row-first serial correction. The correction process is simple and stable, without neighborhood interference or parameter deviation, and can accurately restore the true fluorescence intensity of each pixel. It is suitable for the correction of routine pathological images with uniform tissue distribution and single biomarker immunofluorescence staining.

[0078] The second method is nonlinear weighted parallel correction with tissue boundary protection. Before the formal correction process begins, the original high-resolution fluorescence image and the target quenching field are read. The tissue type homogeneous region segmentation results and tissue boundary masks generated in step S40 are simultaneously loaded, dividing the original image into three feature regions: internal tissue regions, tissue boundary regions, and blank background regions. Differentiated correction constraint rules are set for different regions. Simultaneously, the image is divided into multiple computationally independent parallel correction units according to each region. Parallel correction calculations are then initiated simultaneously for all parallel correction units. Within each correction unit, for pixels in the internal tissue region, each pixel is treated as an independent computational unit, and a linear inverse compensation formula is used to complete the correction calculation, ensuring the uniformity of the fluorescence signal within the tissue. For pixels in the tissue boundary region, a nonlinear weighted compensation model based on 3×3 neighborhood tissue features is constructed. Combining the quenching attenuation coefficient of the central pixel and the average attenuation coefficient of the same type of tissue in the neighborhood, nonlinear weighted inverse correction calculations are performed to avoid boundary blurring, contrast reduction, and loss of diagnostic details caused by linear correction. For pixels in the blank background region, a neighborhood smoothing constraint is superimposed during inverse correction to suppress excessive amplification of background noise. After completing the inverse correction calculation for all pixels, differentiated and reasonable range constraints are applied to the correction results for different regions: For internal tissue regions, the full range of original image grayscale values ​​is constrained to ensure the dynamic range of fluorescence signals. For tissue boundary regions, the fluctuation range of the neighborhood mean is superimposed to avoid abrupt changes in boundary intensity and artifacts. For blank background regions, a low intensity upper limit constraint is applied to suppress background noise amplification. After processing, the target intensity values ​​of all pixels in the entire image are obtained. After all parallel correction units have completed their processing, all target intensity values ​​are summarized, and a smooth transition processing is performed on the region boundaries to reconstruct the entire image, ultimately generating the target fluorescence image. This method uses nonlinear weighted and regional parallel correction with tissue boundary protection, which can eliminate fluorescence quenching brightness unevenness artifacts while fully preserving the key tissue boundaries and cellular substructure details for pathological diagnosis. It solves the industry pain points of blurred boundaries, amplified background noise, and loss of key diagnostic information in conventional linear correction methods. At the same time, it significantly improves the correction efficiency of large-size, high-resolution pathological images through parallel computing, making it suitable for clinical diagnostic pathological image correction scenarios involving multi-marker immunofluorescence staining and complex tissue types and fine cellular structures.

[0079] In one alternative approach, the system's preset linear inverse correction rules and range constraint thresholds are first retrieved. Based on the attenuation coefficient of the target quenching field, pixel-by-pixel linear correction is performed on the original fluorescence image to directly generate the target fluorescence image. This method is fast, logically simple, and suitable for the rapid correction needs of routine batch pathological images.

[0080] In another alternative approach, tissue type features, diagnostic region distribution, and staining type features of the original fluorescence image are first extracted. Simultaneously, a correction rule library corresponding to the scene is retrieved. Based on this rule library, a region-specific correction strategy, a nonlinear weighted model, and range constraint rules adapted to the current image are dynamically generated. Then, parallel correction is performed on the original fluorescence image based on the dynamically generated strategy, while a quality check is performed on the corrected target fluorescence image to ensure that the image quality meets clinical diagnostic requirements. This method can dynamically generate a suitable correction strategy based on image features, resulting in higher correction accuracy, more complete detail preservation, and adaptability to the high-precision correction needs of clinical diagnostic-grade pathological images.

[0081] In an exemplary scheme for determining the target fluorescence image, the original high-resolution fluorescence image to be corrected, the target quenching field obtained in step S40, and the corresponding tissue homogeneous region segmentation results, tissue boundary masks, and core diagnostic region marking results are first loaded. Then, pixel-to-attenuation coefficient matching is performed. According to the pixel coordinate system of the original image, a quenching attenuation coefficient at the same spatial location is matched to each pixel, completing the one-to-one binding of the entire image's pixel intensity information with the corresponding attenuation coefficient. Simultaneously, invalid pixels in blank background areas are removed, reducing computational load. Next, correction unit division and strategy configuration are performed. Based on the tissue segmentation results, the original image is divided into four parallel correction units: tumor tissue core unit, normal tissue unit, tissue boundary unit, and blank background unit. Differentiated correction algorithms and constraint rules are configured for each unit: the tumor tissue core unit uses a linear inverse correction algorithm with full dynamic range constraints; the normal tissue unit uses a linear inverse correction algorithm with neighborhood smoothing constraints; the tissue boundary unit uses a nonlinear weighted correction algorithm with boundary feature protection constraints; and the blank background unit uses a linear correction combined with a neighborhood smoothing algorithm with low intensity upper limit constraints. Subsequently, multi-unit parallel correction calculations are initiated, with parallel processing started simultaneously for four correction units. Within each unit, according to the configured algorithm and constraint rules, inverse correction calculations and range constraints are completed for all pixels in the corresponding region to obtain the target intensity value for each pixel. After all correction units have been processed, intensity smoothing transition processing is performed on the boundary regions of adjacent units to eliminate abrupt intensity changes between regions. Based on the pixel coordinate system of the original image, the target intensity values ​​of all pixels are summarized to reconstruct a complete global corrected image. Image quality verification is then performed, calculating three quality indicators for the corrected image: local statistical uniformity, biosignal restoration, and quenching artifact residue. This confirms that the image quality meets the preset requirements, with no artifact residue, signal distortion, or loss of detail. After successful verification, the final global corrected image is designated as the target fluorescence image, completing the quenching correction process for this digital pathology fluorescence image.

[0082] It should be noted that in some special cases, if the quality of the corrected target fluorescence image does not meet the preset requirements, the system will automatically return to step S40, optimize and adjust the solution rules for the target quenching field, re-solve the target quenching field, and perform the correction again. If the quality requirements still cannot be met after multiple iterations, the abnormal situation will be automatically recorded, prompting the operator to intervene manually.

[0083] Further, please refer to Figure 2 , Figure 2 This is a flowchart of the correction process for this application. The specific implementation process for correcting digital pathological fluorescence images is as follows: First, input the original fluorescence image I_raw to be corrected. Perform downsampling processing on the original fluorescence image to generate a sampled fluorescence image. At the same time, initialize the scan path model parameter φ0. Then, perform joint optimization of the scan path and quenching attenuation field on the sampled fluorescence image to obtain the preliminary low-resolution optimal scan path parameter φ_low and low-resolution quenching attenuation field Θ_low. Then, map the low-resolution optimal scan path parameter φ_low to the high-resolution pixel coordinate system of the original fluorescence image as the initial value for high-resolution optimization. Subsequently, perform block-based fine optimization on the high-resolution image to obtain the final high-resolution target quenching field D_high. Then, apply the correction formula I_corrected=I_raw / D_high to perform inverse compensation correction of pixel intensity on the original fluorescence image. Finally, output the corrected image I_corrected.

[0084] Second Embodiment This embodiment provides an exemplary scheme for constructing a scanning sequence field of digital pathological fluorescence images. In this example, a scanning path parameterization model is first initialized based on the pixel coordinate system and image size of the sampled fluorescence image to obtain a scanning path framework adapted to the sampled fluorescence image. Then, according to the scanning path framework, the corresponding scanning relative time is matched for the pixels in the two-dimensional pixel coordinate system of the sampled fluorescence image to construct a mapping relationship between the scanning path and the scanning relative time. Finally, the scanning time sequence index corresponding to the pixels in the sampled fluorescence image is matched through the mapping relationship to obtain the scanning sequence field. Before step S20, steps A11 to A13 are also included: Step A11: Initialize the scanning path parameterization model based on the pixel coordinate system and image size of the sampled fluorescence image to obtain a scanning path framework adapted to the sampled fluorescence image.

[0085] Step A12: Based on the scanning path framework, match the corresponding scanning relative time for each pixel in the two-dimensional pixel coordinate system of the sampled fluorescence image, and construct the mapping relationship between the scanning path and the scanning relative time.

[0086] Step A13: Match the scanning time sequence index corresponding to the pixel in the sampled fluorescence image with the mapping relationship to obtain the scanning sequence field.

[0087] A scan path parameterization model is a standardized model used to mathematically and parameterize the scanning path of a digital pathology scanner for acquiring fluorescence images. It is the core framework for realizing the mapping between the scan path and the scan timing sequence. Examples include raster scan path parameterization models, serpentine scan path parameterization models, block scan path parameterization models, custom bidirectional scan path parameterization models, and continuous line scan timing parameterization models.

[0088] The scanning path framework is a spatial arrangement rule that perfectly matches the spatial size of the sampled fluorescence image, obtained after initialization through a scanning path parameterization model based on the pixel coordinate system and image size of the sampled fluorescence image. It includes core path parameters such as the scanning start point, main scanning direction, row switching rules, and scanning block logic. Examples include raster scanning path frameworks adapted to the sampled image size, serpentine scanning path frameworks, multi-block parallel scanning path frameworks, and path frameworks prioritizing regions of interest in pathological tissues.

[0089] Relative scanning time is the time difference between the time when the scanner's laser excitation end reaches the corresponding pixel and the time when the scan begins, within the scanning path framework. It is used to quantify the order of fluorescence excitation at each pixel and is a core timing parameter for constructing the correlation between scanning timing and fluorescence quenching. Examples include pixel-level relative timestamps, average relative time of scan lines, relative time at the start of block scans, and relative time difference in the scanning reversal area.

[0090] The mapping relationship between the scan path and the relative scan time is a one-to-one correspondence rule between the spatial coordinates of pixels within the scan path framework and their corresponding relative scan times. It is the core mathematical relationship for achieving precise binding between spatial position and scan timing. Examples include the linear timing mapping relationship for raster scanning, the commutation correction mapping relationship for serpentine scanning, the parallel timing mapping relationship for block scanning, and the monotonically increasing timing mapping relationship for continuous scanning.

[0091] The scan timing index is a unique index identifier assigned to each pixel in the sampled fluorescence image based on the relative time sequence of scans. It is the core index parameter for constructing the scan sequence field. Examples include pixel-level monotonically increasing timing indexes, scan row-level timing indexes, block-based scan block indexes, and scan reversal zone boundary timing indexes.

[0092] In this example, when initializing the scan path parameterization model based on the pixel coordinate system and image size of the sampled fluorescence image to obtain the scan path framework, the original scan path parameters of the scanner can be referenced. Alternatively, a custom path template adaptation method can be used. First, the tissue distribution characteristics and image size of the sampled fluorescence image are read, and a pre-stored multi-type scan path template library is matched. The path template with the highest fit to the image size and tissue distribution is selected, and the scan path parameterization model is initialized based on this template to obtain the scan path framework adapted to the sampled fluorescence image, thereby completing the construction of the basic scan path framework.

[0093] After constructing the scanning path framework, the scanning relative time matching process is initiated. For all pixels within the scanning path framework, the scanning relative time corresponding to each pixel is calculated sequentially according to the arrangement rules of the scanning path. A unique corresponding scanning relative time is matched for each pixel. Then, based on the one-to-one correspondence between pixel spatial coordinates and corresponding scanning relative times, a mapping relationship between the scanning path and scanning relative times is constructed. Next, based on this mapping relationship, a unique scanning time sequence index is assigned to each pixel in the sampled fluorescence image according to the order of scanning relative times. Finally, the time sequence indices of all pixels are summarized to generate a two-dimensional matrix that perfectly matches the size of the sampled fluorescence image, thus obtaining the scanning sequence field. In this way, through hierarchical parametric modeling and time sequence mapping, the construction accuracy of the scanning sequence field is improved, avoiding the deviation in subsequent quenching field solution caused by mismatch between spatial coordinates and time sequence matching.

[0094] For example, there are two ways to achieve the scanning sequence field through scanning path framework construction and temporal mapping. The first is a temporal-locked serial construction adapted to the scanner's native path. First, the original scanning parameters of the digital pathology scanner for this acquisition task are read, including scanning objective magnification, scanning line frequency, line switching time, scanning main direction, and scanning mode. Based on these original scanning parameters, a scanning path parameterization model that is completely consistent with the scanner's native acquisition path is initialized. Then, combined with the pixel coordinate system and image size of the sampled fluorescence image, the native scanning path parameterization model is scaled and adapted proportionally to obtain a scanning path framework that perfectly matches the size of the sampled fluorescence image. Subsequently, starting from the starting point of the scanning path framework, the relative scanning time of each pixel is calculated sequentially, row by row and pixel by pixel, according to the row priority order of the scanner's native scanning. A relative scanning time that is completely consistent with the native acquisition process is matched for each pixel, and a one-to-one mapping relationship between pixel spatial coordinates and relative scanning time is constructed synchronously. After completing the relative time matching of all pixels in the image, a unique scan timing index is assigned to each pixel according to the monotonically increasing order of the relative scan times. The timing indices of all pixels are then aggregated to generate a two-dimensional matrix, resulting in a scan sequence field that perfectly matches the scanner's native acquisition timing. This method employs a computational logic that combines scanner native path adaptation with timing locking in a serial manner. Through complete adaptation with the scanner's original acquisition parameters, it accurately restores the true scan timing during image acquisition, avoiding the quenching characteristic matching deviation problem caused by path and timing misalignment. It perfectly fits the raster and serpentine scanning modes of conventional clinical pathology scans and is suitable for single-scan mode and conventional-sized pathological fluorescence image scan sequence field construction scenarios.

[0095] The second approach involves tissue feature-guided multi-path parallel construction. First, the sampled fluorescence image is segmented into foreground and background regions to locate the effective foreground region containing pathological tissue and the background region lacking relevant information. Simultaneously, based on tissue distribution characteristics, the effective foreground region is divided into multiple independent homogeneous tissue blocks, and a corresponding scan priority and suitable scan path mode are matched to each block. Then, based on the pixel coordinate system and image size of the sampled fluorescence image, a multi-block parallel scan path parameterization model is initialized. An independent scan path sub-frame is set for each tissue block, and a fast scan path sub-frame is set for the background region, combining to form a global scan path framework adapted to the tissue distribution of the sampled fluorescence image. Subsequently, parallel temporal calculations are simultaneously initiated for all independent path sub-frames. Within each path sub-frame, according to the corresponding scan path mode, the relative scan time of all pixels within the sub-frame is calculated in parallel, synchronously constructing the mapping relationship between pixel spatial coordinates and relative scan time within each sub-frame. Finally, global temporal alignment is performed on the mapping relationships of all sub-frames to eliminate temporal conflicts between blocks, generating a unified mapping relationship between scan paths and relative scan times across the entire image. After completing the full-image mapping, a unique scan temporal index is assigned to each pixel according to a globally unified relative scan time sequence. The temporal indices of all pixels are aggregated to generate a two-dimensional matrix, resulting in a scan sequence field adapted to the tissue distribution characteristics. This method employs a computational logic of multi-path parallel construction guided by tissue features. Through pre-segmentation of tissue regions and adaptation of partitioned path, it specifically optimizes the temporal accuracy of effective areas of pathological tissues, significantly reduces the amount of invalid computation in blank background areas, and improves the construction efficiency of large-size images through multi-block parallel computation. It solves the industry pain points of insufficient temporal accuracy of effective tissue regions and low construction efficiency of large-size full-slice images in conventional globally unified path construction methods, and is suitable for the construction of complex pathological fluorescence image scan sequence fields with large field-of-view full slices and multiple tissue block distributions.

[0096] In an exemplary scheme for determining the scanning sequence field, the sampled fluorescence image generated in the first embodiment is first loaded. The pixel coordinate system, image size, and corresponding original scanner acquisition parameters of the sampled fluorescence image are read synchronously, including scanning mode, line frequency, line switching time, and objective magnification. Then, the scanning path parameterization model is initialized. First, based on the original scanner acquisition parameters, the native raster scanning path parameterization model is initialized. Then, combined with the image size of the sampled fluorescence image, the model is scaled and adapted by a factor of 4. The number of pixels per scan line, the total number of lines, and the line switching timing parameters are adjusted to perfectly match the pixel coordinate system of the sampled fluorescence image, resulting in a raster scanning path framework adapted to the sampled fluorescence image. Next, relative scanning time matching is performed. Using the scanning start point as time 0 as the reference, the relative scanning time of each pixel is calculated row by row according to the row priority order of the raster scan. Within the same scan line, the relative time of each pixel is calculated linearly according to the number of pixel columns and the scan line frequency. Between adjacent scan lines, the line switching time is superimposed to complete the continuous connection of the line timing, matching a unique relative scanning time for each pixel in the entire image. Subsequently, a mapping relationship is constructed. Based on the spatial coordinates of all pixels in the image and their corresponding relative scan times, a one-to-one mapping lookup table is built between pixel coordinates and relative scan times, forming a mapping relationship between the scan path and the relative scan time. Next, scan time-series indices are assigned. Following a monotonically increasing order of relative scan time, each pixel in the image is assigned a consecutive and unique time-series index starting from 1, ensuring that the time-series index is completely positively correlated with the relative scan time. Finally, the time-series indices of all pixels in the image are summarized to generate a two-dimensional time-series index matrix with the same width and height as the sampled fluorescence image, completing the construction of the scan sequence field. At the same time, a time-series consistency check is performed on the generated scan sequence field to ensure that the indices are monotonically continuous, without misalignment, and without repetition. After passing the check, it is used as the time-series benchmark for the subsequent joint optimization objective function construction.

[0097] It should be noted that in some special cases, the original scanner scanning parameters corresponding to the sampled fluorescence image are missing, making it impossible to complete the native path adaptation. In such cases, the system's preset standard raster scan path parameterization model is directly used, and the scan path framework is initialized in conjunction with the size of the sampled fluorescence image to complete the construction of the scan sequence field. If the sampled fluorescence image is a multi-channel fluorescence image, an independent scan sequence field is constructed for each fluorescence channel to ensure that the timing of each channel is completely matched with the acquisition process.

[0098] Third Embodiment This embodiment provides an exemplary scheme for constructing a joint optimization objective function for digital pathological fluorescence images. In this example, the fluorescence intensity of each pixel in the sampled fluorescence image is first correlated with the scanning time sequence in the scanning sequence field to obtain a correlation dataset of fluorescence intensity and scanning time sequence. Then, based on the correlation dataset and the spatially stationary statistical distribution of fluorescence signals from similar biological tissues, a data fidelity term is constructed to measure the local statistical uniformity of the corrected image. Next, based on the physical attenuation law of fluorescence quenching and the correlation dataset, constraints are applied to the spatial gradient of the quenching attenuation field to construct a model regularization term to constrain the smoothness and monotonicity of the quenching attenuation field. Finally, the data fidelity term and the model regularization term are weighted and fused to obtain the joint optimization objective function used to simultaneously optimize the scanning path and the quenching attenuation field. After step A13, steps B11-B14 are also included: Step B11: Associate and match the fluorescence intensity of each pixel in the sampled fluorescence image with the scanning time sequence in the scanning sequence field to obtain an associated dataset of fluorescence intensity and scanning time sequence.

[0099] Step B12: Based on the associated dataset and the spatially stationary statistical distribution of fluorescence signals from similar biological tissues, construct a data fidelity term to measure the local statistical uniformity of the corrected image.

[0100] Step B13: Based on the physical decay law of fluorescence quenching and the associated dataset, impose constraints on the spatial gradient of the quenching decay field and construct a model regularization term to constrain the smoothness and monotonicity of the quenching decay field.

[0101] Step B14: The data fidelity term and the model regularization term are weighted and fused to obtain the joint optimization objective function used to simultaneously optimize the scan path and the quenching attenuation field.

[0102] A correlated dataset is a pixel-level structured data set formed by binding the two-dimensional spatial coordinates, original fluorescence intensity value, and scan time index and relative scan time of each pixel in the sampled fluorescence image to the corresponding pixel in the scan sequence field. It is the core foundational data for constructing a joint optimization objective function. Examples include pixel-level fluorescence intensity and time-series correlated tuples, fluorescence time-series statistical data sets of homogeneous tissue regions, scan row-level intensity and time-series correlated datasets, and partitioned block fluorescence time-series feature datasets.

[0103] The spatially stationary statistical distribution of fluorescence signals in similar biological tissues refers to the statistical characteristics of a stationary random process in the spatial distribution of fluorescence emission intensity of homogeneous biological tissues of the same type and degree of differentiation in pathological images under ideal conditions without fluorescence quenching. That is, the mean and variance of fluorescence intensity within a homogeneous region do not change significantly with spatial location. This is a core a priori criterion for measuring the uniformity of fluorescence signals in corrected images. Examples include the stationary distribution characteristics of fluorescence intensity in similar tumor tissues, the statistical distribution patterns of fluorescence signals in normal epithelial tissues, the uniform distribution characteristics of fluorescence intensity in negative control tissues, and the spatial statistical characteristics of fluorescence signals in tissues stained with the same biomarker.

[0104] The data fidelity term is the core loss term in the joint optimization objective function. It quantifies the degree to which the fluorescence signal of similar biological tissues in the image corrected based on the current scan path and quenching attenuation field conforms to the spatially stable statistical distribution. The smaller the value, the better the uniformity of the fluorescence signal in the corrected image and the more thoroughly quenching artifacts are eliminated. Examples include the variance loss term for fluorescence intensity in homogeneous tissue regions, the loss term for fitting the stability of the corrected signal, the penalty term for the dispersion of signals in similar tissues, and the weighted loss term for fidelity in the core diagnostic region.

[0105] The physical decay law of fluorescence quenching refers to the physical characteristic that the fluorescence emission intensity of a fluorescent dye monotonically decreases with increasing excitation time under continuous excitation light irradiation. This is the core physical constraint followed by fluorescence quenching correction and is commonly expressed mathematically using single-exponential decay models and double-exponential decay models. Examples include the monotonically decreasing fluorescence intensity with excitation time, the single-exponential fluorescence quenching decay model, the double-exponential fluorescence quenching decay model, and the continuous fluorescence decay characteristics in the scanning switching region.

[0106] The model regularization term is a core constraint term in the joint optimization objective function. It is used to impose hard constraints on the spatial smoothness and temporal monotonicity of the solved quenching attenuation field, avoiding overfitting, abrupt changes in attenuation coefficients, and outliers that do not conform to physical laws during the solution process. The smaller the value, the better the physical compliance of the quenching attenuation field. Examples include the spatial gradient smoothness constraint term of the quenching attenuation field, the temporal monotonicity penalty term of the attenuation coefficient, the attenuation continuity constraint term of the scanning reversal region, and the physical compliance regularization term of the global attenuation field.

[0107] In this example, when associating the fluorescence intensity of each pixel in the sampled fluorescence image with the scan time sequence in the scan sequence field to obtain the associated dataset, pixel-level temporal matching can be used. Alternatively, a homogeneous region aggregation method can be employed. First, unsupervised homogeneous tissue segmentation is performed on the sampled fluorescence image to obtain multiple similar biological tissue regions. Then, using homogeneous tissue regions as units, the fluorescence intensity of all pixels within the matching region is aggregated with the corresponding scan time sequence to generate a region-level and pixel-level combined associated dataset, thereby completing the association matching of fluorescence intensity and scan time sequence.

[0108] After constructing the associated dataset, the data fidelity term construction process is initiated. Based on the associated dataset, fluorescence intensity and corresponding scanning time series data of similar biological tissue regions are extracted. Combined with the prior spatial stationary statistical distribution of fluorescence signals of similar biological tissues, a data fidelity term is constructed to measure the local statistical uniformity of the corrected image. Next, the model regularization term construction process is initiated. Based on the physical decay law of fluorescence quenching and combined with the correspondence between time series and fluorescence intensity in the associated dataset, constraints are imposed on the spatial gradient and temporal monotonicity of the quenching decay field to construct a model regularization term to constrain the smoothness and monotonicity of the quenching decay field. Subsequently, the constructed data fidelity term and model regularization term are weighted and fused according to preset weight coefficients to finally obtain a joint optimization objective function for simultaneously optimizing the scanning path and the quenching decay field. In this way, the solution accuracy and physical compliance of the joint optimization objective function are improved through the hierarchical construction and weighted fusion of dual constraint terms, avoiding overfitting and correction distortion problems caused by lack of constraints.

[0109] For example, there are two ways to achieve the joint optimization objective function by constructing an associated dataset and fusing it with dual constraints. The first is to construct a standardized dual-constraint objective function with global homogeneous region locking. First, global unsupervised homogeneous tissue segmentation is performed on the sampled fluorescence image to remove blank background areas and divide it into multiple homogeneous regions of the same biological tissue. Then, on a pixel-by-pixel basis, the fluorescence intensity of each pixel in the homogeneous region is bound to the scan time index and scan relative time of the corresponding pixel in the scan sequence field to generate an associated dataset of fluorescence intensity and scan time at the pixel level of the whole image. Subsequently, based on the associated dataset, for each homogeneous tissue region, the statistical variance of the fluorescence intensity of that region in the corrected image is calculated. The data fidelity term is constructed with the sum of the statistical variances of all homogeneous regions as the core to ensure that the fluorescence signal of the same tissue after correction conforms to the spatially stationary statistical distribution. Then, based on the single exponential decay model of the physical decay law of fluorescence quenching, the L2 norm smoothing constraint is applied to the spatial gradient of the quenching decay field, and the monotonically decreasing constraint is applied to the change of the decay coefficient with the scan time to construct a model regularization term to ensure that the solved quenching decay field conforms to the physical law. Finally, fixed standardized weight coefficients are set for the data fidelity term and the model regularization term, and weighted summation and fusion are performed to obtain a globally unified standardized joint optimization objective function. This method adopts a standardized double-constraint construction logic with global homogeneous region locking. Irrelevant background interference is eliminated through pre-processing homogeneous tissue segmentation. The weights of the double-constraint terms are fixed and stable, the function construction logic is simple and controllable, the solution process has good convergence, and it can provide a stable optimization benchmark for subsequent parameter solving. It is suitable for the correction scenario of conventional pathological fluorescence images with uniform tissue distribution and single marker staining.

[0110] The second approach involves constructing a multi-constraint dynamic weight objective function guided by diagnostic priority. First, pathological feature recognition is performed on the sampled fluorescence images to locate the core diagnostic region containing tumor tissue and positive biomarkers, the normal tissue region, and the blank background region. Differentiated diagnostic priority weights are assigned to different regions. Simultaneously, combined with the scanning sequence field generated in the second embodiment, scanning reversal zones and high-quenching-risk regions in long-term scanning are located, and enhanced constraint weights are assigned to high-risk regions. Then, on a pixel-by-pixel basis, corresponding diagnostic priority weights and risk region constraint weights are matched to each pixel in the entire image. The fluorescence intensity, scanning time sequence, and weight coefficients of each pixel are bound one-to-one, generating a multi-dimensional fluorescence intensity and scanning time sequence association dataset with weighted labels. Subsequently, based on the association dataset, differentiated loss calculation weights are set for regions of different priorities, amplifying the loss weight of the core diagnostic region to construct a region-weighted data fidelity term, prioritizing the correction accuracy and signal fidelity of the core diagnostic region. Based on a double-exponential decay model using the physical decay law of fluorescence quenching, this method strengthens monotonicity constraints for high-quenching-risk areas, spatial smoothness constraints for tissue boundary areas, and simplifies constraints for blank background areas, constructing multi-regional differentiated model regularization terms to balance physical compliance and solution efficiency. Finally, based on the tissue complexity and artifact severity of the image, the global weighting coefficients of the data fidelity term and the model regularization term are dynamically calculated, and weighted fusion is performed to obtain a dynamic weight joint optimization objective function adapted to the image features. This method adopts a diagnostic priority-oriented multi-constraint dynamic weight construction logic. Through pathological feature recognition and regional differentiated weight configuration, it prioritizes the correction effect of core clinical diagnostic areas, while adapting to the correction needs of different areas through dynamic weights. This solves the industry pain points of conventional standardized objective functions failing to balance core diagnostic accuracy and global correction effect, and having poor adaptability to complex artifact scenarios. It is suitable for clinical-grade pathological fluorescence image correction scenarios with multiple marker staining and complex tissue types and key diagnostic information.

[0111] In an exemplary scheme that determines the joint optimization objective function, the generated sampled fluorescence image, scan sequence field, and prior model of the spatially stationary statistical distribution of fluorescence signals from similar biological tissues are first loaded. Then, an associated dataset is constructed. Using the pixel coordinate system of the sampled fluorescence image as a reference, the corresponding original fluorescence intensity value is matched to each pixel. Simultaneously, the scan time sequence index and scan relative time corresponding to that pixel are extracted from the scan sequence field. The pixel spatial coordinates, fluorescence intensity value, scan time sequence index, and scan relative time are bound one-to-one to generate pixel-level associated tuples. The associated tuples of all pixels are then summarized to obtain the associated dataset of fluorescence intensity and scan time for the entire image. Next, the data fidelity term was constructed. First, an unsupervised clustering algorithm was used on the sampled fluorescence images to divide them into five homogeneous tissue regions based on fluorescence intensity and texture features. Blank background regions were removed. Then, based on the associated dataset, a variance loss function for the corrected fluorescence intensity was constructed for each homogeneous tissue region. The objective was to minimize the variance of fluorescence intensity of all pixels within the corrected region. Combined with a priori knowledge of the spatially stationary statistical distribution of similar tissues, the core calculation formula for the data fidelity term was generated. Simultaneously, a weighting coefficient of 1.5 was set for the core region containing positive markers to strengthen the fidelity constraint of the core region. Subsequently, the model regularization term was constructed. Based on the fluorescence quenching single-exponential physical decay model, two core constraints were set: first, a spatial smoothness constraint, applying an L2 norm penalty to the spatial gradient of the quenching decay field to avoid abrupt spatial changes in the decay coefficient; second, a temporal monotonicity constraint, applying a non-increasing penalty to the change of the decay coefficient with the scanning time sequence to ensure that the decay coefficient monotonically decreases with scanning time. These two constraints were fused to generate the core calculation formula for the model regularization term. Finally, a weighted fusion was performed, setting a global weight of 0.6 for the data fidelity term and a global weight of 0.4 for the model regularization term. The two terms were then summed to obtain the final joint optimization objective function. At the same time, the input-output range and convergence characteristics of the function were verified to confirm that the function could achieve bivariate co-optimization of the scan path and the quenching decay field. After the verification was passed, it was used as the core optimization criterion for subsequent parameter iteration.

[0112] It should be noted that in some special cases, when the sampled fluorescence image is a multi-channel immunofluorescence image, an independent data fidelity term and model regularization term are constructed for each fluorescence channel. These are then fused to generate a joint optimization objective function for simultaneous optimization of multiple channels, ensuring that the calibration process of each channel does not interfere with each other. If the effective tissue area of ​​the sampled fluorescence image is extremely low, the weight coefficient of the model regularization term is automatically increased to strengthen the constraints of physical laws and avoid calibration distortion caused by overfitting.

[0113] Fourth embodiment This embodiment provides an exemplary scheme for constructing a data fidelity term for digital pathological fluorescence image correction. In this example, based on the spatially stationary statistical distribution of fluorescence signals from similar biological tissues, the sampled fluorescence image is first uniformly divided into several local image blocks. Then, the average value of the corrected fluorescence intensity of pixels within each block is calculated to obtain the local intensity mean. Next, the fluorescence intensity variance of each local image block is calculated based on the difference between the corrected fluorescence intensity of each pixel and the corresponding local intensity mean. Finally, the average value of the fluorescence intensity variances of all local image blocks is calculated to construct a data fidelity term that measures the local statistical uniformity of the corrected image. Please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the fourth embodiment of the digital pathological fluorescence image correction method of this application. Step B12 includes steps C11 to C14: Step C11: Based on the spatially stationary statistical distribution of fluorescence signals from the same type of biological tissue, the sampled fluorescence image is uniformly divided into blocks to obtain local image blocks.

[0114] Step C12: For each local image block, calculate the average value of the corrected fluorescence intensity of all pixels within the local image block to obtain the local intensity mean.

[0115] Step C13: Calculate the fluorescence intensity variance of the local image block based on the difference between the average value of each fluorescence intensity and the corresponding local intensity mean.

[0116] Step C14: Calculate the average value of the fluorescence intensity variance of all the local image patches to obtain the data fidelity term.

[0117] A local image patch is a set of square pixels with independent spatial boundaries, obtained by uniformly dividing a sampled fluorescence image into fixed-size blocks based on the spatially stationary statistical distribution of fluorescence signals from similar biological tissues. It is the smallest processing unit for calculating the statistical characteristics of local fluorescence intensity. Examples include 8×8 pixel local image patches, 16×16 pixel local image patches, 32×32 pixel local image patches, and 64×64 pixel local image patches.

[0118] The local intensity mean is the arithmetic mean of the corrected fluorescence intensity values ​​of all pixels within a single local image patch. It is used to characterize the true fluorescence signal baseline level of similar biological tissues within that local image patch and is the core reference value for calculating the variance of local fluorescence intensity. Examples include the local intensity mean of foreground tissue patches, the local intensity mean of background patches, the local intensity mean of tumor patches, and the local intensity mean of normal tissue patches.

[0119] Fluorescence intensity variance is the average of the sum of squares of the differences between the corrected fluorescence intensity values ​​of all pixels and the corresponding local intensity mean within a single local image patch. It is used to quantify the dispersion of fluorescence intensity within that local image patch and is a core quantitative indicator for measuring local statistical uniformity. Examples include single-patch fluorescence intensity variance, row-level patch average variance, region-level patch mean variance, and total variance of the entire image patch.

[0120] In this example, when uniformly dividing the sampled fluorescence image into local image blocks, the method of synchronously reading pixels across the entire image can be used. Alternatively, a streaming block incremental processing method can be employed, reading the pixel data of the sampled fluorescence image row by row, completing the image block division row by row according to the preset block size, and synchronously caching the pixel data of each local image block while dividing the image, accumulating all local image blocks of the entire image. This reduces memory usage while completing the uniform block division of the image.

[0121] After uniformly dividing the sampled fluorescence image into blocks and obtaining all local image blocks, the calculation process for the local intensity mean is initiated. Each local image block is traversed, and the corrected fluorescence intensity values ​​of all pixels within the block are extracted to calculate the corresponding local intensity mean. Subsequently, for each local image block, the difference between the corrected fluorescence intensity of each pixel and the corresponding local intensity mean is calculated. Based on this difference, the fluorescence intensity variance of the local image block is calculated. After the variances of all local image blocks are calculated, the arithmetic mean of all fluorescence intensity variances is taken. This is used as the core to construct a data fidelity term that measures the local statistical uniformity of the corrected image. Through block-level hierarchical statistical calculation, the local uniformity of the corrected image is accurately quantified, avoiding local correction bias caused by global statistics and improving the data fidelity term's constraint on local quenching artifacts.

[0122] For example, there are two ways to construct a data fidelity term using the fluorescence intensity variance of local image patches. The first is a fixed-size, uniformly divided, serial statistical construction across the entire image. First, based on the overall size of the sampled fluorescence image and the microscopic scale characteristics of similar biological tissues, a fixed local image patch size is preset. The sampled fluorescence image is then uniformly divided into non-overlapping blocks according to the preset patch size, resulting in a set of local image patches of identical size. Starting from the beginning of this set of local image patches, individual local image patches are selected sequentially in row-major order. The arithmetic mean of the corrected fluorescence intensity of all pixels within the patch is calculated to obtain the local intensity mean of the patch. Then, based on the difference between the intensity values ​​of each pixel within the patch and the local intensity mean, the fluorescence intensity variance of the patch is calculated. After completing the statistical calculation for a single patch, the variance result for that patch is cached. The calculation process then proceeds to the next local image patch. After all local image patches in the entire image have completed their statistical calculations, the fluorescence intensity variance results of all patches are summarized, and the arithmetic mean of all variances is calculated. This average value is used as the core to construct the data fidelity term. This method employs a fixed-size uniform block division and row-first serial statistical calculation logic. The block division rules are unified, the statistical process is stable and controllable, and there is no inter-block calculation interference. It can accurately quantify the local statistical uniformity of each region in the whole image and is suitable for the correction of conventional pathological fluorescence images with uniform tissue distribution and no significant large blank areas.

[0123] The second method involves adaptive block-based parallel statistical construction constrained by tissue masks. First, the sampled fluorescence image is segmented into tissue foreground and blank background to generate tissue region masks. Based on the distribution characteristics of tissue region masks and similar biological tissues, adaptive uniform block segmentation is performed on the tissue foreground region, while large-size merging block segmentation is performed on the blank background region, resulting in a set of local image blocks adapted to the tissue distribution characteristics. Simultaneously, the tissue type attribute of each image block is labeled, generating multiple independent computational units that are computationally independent of each other. Then, parallel statistical calculations are initiated synchronously for all local image blocks. Within each local image block, corresponding statistical weights are first set based on the block's tissue type attribute, and then the weighted average of the fluorescence intensity after pixel correction within the block is calculated to obtain the local intensity mean. Next, the weighted fluorescence intensity variance within the block is calculated. After completing the parallel statistics for all blocks, a weighted average calculation is performed on the fluorescence intensity variance of all blocks based on the tissue type weights of each block. The data fidelity term is constructed using the weighted average as the core. This method employs adaptive block segmentation with tissue mask constraints and parallel statistical calculations across all blocks. By performing pre-processing tissue region segmentation and adaptive block segmentation, it avoids interference from blank background blocks on the statistical results of tissue regions. At the same time, it strengthens the constraint weights of core diagnostic tissue regions through weighted statistics, improving the adaptability of data fidelity items to key areas of pathological diagnosis. This method addresses the pain points of background region interference with statistical results and insufficient constraints on key tissue regions in conventional uniform block segmentation methods, and is suitable for clinical diagnostic pathological fluorescence image correction scenarios that include large blank backgrounds and complex tissue distributions.

[0124] Furthermore, the scan path parameterization model is established: The scan path is then created. With image two-dimensional coordinates mapping relationship ,in This indicates the relative time or sequential index at which the pixel was scanned. Model parameters. This can include scanning direction (e.g., left-to-right, serpentine), scan line width, acceleration model, etc. This is the core of introducing physical processes into software correction. Content-adaptive path optimization: A core assumption is proposed: at the microscopic scale, the statistical distribution (e.g., mean, variance) of the true fluorescence signal intensity of similar biological tissues (e.g., a continuous sheet of tumor epithelial cells) should be relatively stationary in space. Observed spatial variations in intensity can be partially attributed to the quenching field $D(T)$. Path parameters are iteratively adjusted. This makes it possible to follow the estimated path The extracted image patch signal, after being processed based on After initial correction, the spatiotemporal consistency of its statistical properties (such as local mean and variance) reaches optimal. Quenching attenuation field modeling and estimation: Quenching field Modeling as about scan order Smoothly monotonically decreasing functions (such as exponential decay) ). In the path With a fixed value, the attenuation parameter is estimated by fitting the relationship between the intensity values ​​of all pixels (or sampling points) in the image and their corresponding T values, and by combining this with the above content consistency constraint. Joint optimization framework: constructing an objective function ,in For quenching field parameters. This function contains two items: a data fidelity item: based on the current... and The generated corrected image should make the distribution of local image statistics (such as gradient magnitude and texture features) more uniform. Model regularization term: for quenching field parameters. Apply smoothness constraints to prevent overfitting. This is achieved by minimizing... To simultaneously solve for the optimal scan path and quenching field Furthermore, the specific mathematical definitions of the data fidelity term and the regularization term are given. The specific mathematical definition of the joint optimization framework is as follows: To solve for the optimal scan path parameters... and quenching field parameters Construct the following objective function: ,in This is the regularization coefficient, used to balance the weights of the two terms. (Data fidelity term) Based on the core assumption that "the statistical characteristics of fluorescence signal intensity in similar biological tissues are spatially stationary at the microscale," the corrected image... Divided into Local image patch (Sliding windows or superpixel segmentation can be used), each block contains Each pixel. The data fidelity term is defined as the average of the intensity variances within all blocks: ,in For the first The average pixel intensity within each block: This formula promotes uniform intensity within each local block, thereby eliminating the spatial brightness gradient caused by quenching, while preserving the intensity differences between different microstructures (because the mean of different blocks is lower). (These can be different). Furthermore, local statistics can be extended to gradient magnitudes or texture features to enhance adaptability to different tissue structures. Model regularization term. To avoid quenching field To prevent overfitting and ensure the smoothness of its spatial variation, the spatial gradient of the quenching field is subjected to... Norm constraint. Due to Finally mapped to each pixel location Above, define the spatial quenching field. Then the regular expression term is: ,in and These are the gradient operators for the horizontal and vertical directions of the image, respectively. These are the weighting coefficients. In practical calculations, the gradient can be approximated using finite difference: ; ; The spatial abrupt change in this penalty quenching field ensures that it monotonically decreases and changes gradually along the scanning direction, consistent with the physical characteristics of fluorescence quenching. Image correction: Obtaining optimal... and Then, the original fluorescence image was analyzed. Perform correction: ; Specific implementation steps: Input a digital pathological fluorescence image with uneven brightness. The image is downsampled to generate a low-resolution version for rapid estimation of the global trend. The scan path model parameters are initialized. (Settings can be based on the default settings of common scanners). On the sampled fluorescence image, a joint optimization algorithm is executed to solve the preliminary... and The path estimated at low resolution Upsampling is mapped to the original high-resolution image coordinates, serving as the initial values ​​for high-resolution optimization. Further finer joint optimization is then performed on the original fluorescence image using a block-based processing approach to obtain the final high-resolution quenching field. The correction formula is applied to generate the corrected image. Output the corrected image.

[0125] Fifth embodiment This embodiment provides an exemplary scheme for iteratively solving the initial correction parameters of digital pathological fluorescence images. In this example, the scanning relative time of each pixel in the sampled fluorescence image is first matched according to the initial scan path parameters to construct a pixel-level temporal correlation set. Then, the pixel-level temporal correlation set is iteratively processed by jointly optimizing the objective function, updating the path parameters and quenching field parameters round by round and calculating the corresponding objective function values. Finally, the objective function values ​​of continuous iterations are compared. When the objective function value meets the convergence condition, the corresponding path parameters and quenching field parameters are determined as the initial path parameters and initial quenching field parameters. Step S20 includes steps D11~D13: Step D11: Based on the initial scan path parameters, match the relative scanning time of each pixel in the sampled fluorescence image to construct a pixel-level temporal correlation set.

[0126] Step D12: Iteratively process the pixel-level temporal correlation set through the joint optimization objective function, update the path parameters and quenching field parameters round by round, and calculate the corresponding objective function value.

[0127] Step D13: Compare the objective function values ​​of the continuous iterations. When the objective function value satisfies the convergence condition, determine the path parameters and quenching field parameters corresponding to the objective function value as the initial path parameters and initial quenching field parameters.

[0128] A pixel-level temporal correlation set is a set of structured temporal and fluorescence intensity correlation data at the pixel level across the entire image. It binds the two-dimensional spatial coordinates, original fluorescence intensity value, and scan relative time and scan temporal index of each pixel in the sampled fluorescence image, based on the initial scan path parameter matching. This set serves as the core input data for iteratively solving the joint optimization objective function. Examples include single-pixel temporal and fluorescence intensity correlation tuple sets, homogeneous tissue region temporal feature aggregation datasets, scan row-level temporal and fluorescence intensity correlation matrices, and partitioned block-weighted temporal correlation datasets.

[0129] Convergence criteria are preset rules used to determine whether the iterative solution process of the joint optimization objective function has reached a stable optimal state. They are the core criteria for terminating the iteration loop and outputting the optimal parameters. Examples include thresholds for the change in the objective function value over consecutive iterations, limits on the maximum number of iterations, thresholds for the global minimum value of the objective function, and convergence thresholds for the change in parameter iterations.

[0130] The initial quenching field parameters are coarse-scale quenching attenuation field parameters obtained by iterative convergence with the initial path parameters in the dimension of the sampled fluorescence image. These parameters conform to the physical attenuation law of fluorescence quenching and serve as the initial benchmark for subsequent high-resolution target quenching field solutions. Examples include the low-resolution global quenching attenuation coefficient matrix, the fluorescence attenuation model parameters in the sampled image dimension, the iteratively converged optimal quenching field distribution parameters, and the scan time-series correlated attenuation coefficient set.

[0131] In this example, when constructing a pixel-level temporal association set by matching the relative scan times of each pixel in the sampled fluorescence image based on the initial scan path parameters, the pixel-level scan temporal matching method can be used. Alternatively, a homogeneous region aggregation method can be employed. First, unsupervised homogeneous tissue segmentation is performed on the sampled fluorescence image to obtain multiple similar biological tissue regions. Then, using homogeneous tissue regions as units, the corresponding relative scan times are matched for all pixels within each region. Simultaneously, weight labels are added to the pixels in the core diagnostic region, generating a pixel-level temporal association set with region priority labels. This completes the temporal and intensity association matching.

[0132] After constructing the pixel-level temporal correlation set, the joint optimization iterative solution process is initiated. The pixel-level temporal correlation set is input into the joint optimization objective function, with minimizing the objective function value as the core criterion. The scanning path parameters and quenching field parameters are updated iteratively round by round, and the objective function value corresponding to each round of iteration is calculated simultaneously. After completing multiple rounds of iteration, the objective function values ​​of consecutive iterations are compared to determine whether the preset convergence condition is met. When the convergence condition is met, the iteration loop is terminated. The path parameters and quenching field parameters corresponding to the minimum objective function value during the iteration process are determined as the initial path parameters and initial quenching field parameters, respectively. In this way, through bivariate collaborative iteration and convergence determination, the solution accuracy and global optimality of the initial correction parameters are improved, avoiding the problem of poor correction effect caused by iteration getting stuck in local optima.

[0133] For example, there are two ways to obtain the initial path parameters and initial quenching field parameters through joint optimization of the objective function iteratively. The first is a two-parameter alternating fixed-step serial iterative solution. First, the initial scan path parameters are used as the initial values ​​for the scan path iteration, and the identity matrix is ​​used as the initial values ​​for the quenching field parameters. A fixed iteration step size, a maximum number of iterations, and a convergence threshold are set. Then, the alternating serial iterative process is started. In the first iteration, the quenching field parameters are kept unchanged, and the optimization objective is to minimize the objective function value. The scan path parameters are updated, and the updated objective function value is calculated. In the second iteration, the updated scan path parameters are kept unchanged, and the optimization objective is to minimize the objective function value. The quenching field parameters are updated, and the updated objective function value is calculated. This process continues, with one set of parameters fixed and the other set optimized in each round, while the parameter combinations and objective function values ​​for each iteration are recorded synchronously. After each iteration, the objective function value of the current iteration is compared with that of the previous iteration to determine whether the preset convergence condition is met. The iteration terminates when the difference between the objective function values ​​of two consecutive iterations is less than the preset convergence threshold, or when the maximum number of iterations is reached. The scanning path parameters and quenching field parameters corresponding to the minimum objective function value during the iteration process are determined as the initial path parameters and initial quenching field parameters, respectively. This method employs a serial iteration logic with alternating dual parameters and fixed step sizes. The iteration process is simple and stable, the convergence process is controllable, and it is less prone to iteration oscillations. It can quickly lock in a stable optimal parameter combination and is suitable for solving parameters in conventional pathological fluorescence images with uniform tissue distribution and simple artifact characteristics.

[0134] The second approach employs a bivariate collaborative adaptive step-size global optimization iterative solution. First, pathological feature recognition is performed on the sampled fluorescence images to locate the core diagnostic region, normal tissue region, and blank background region. Regionally differentiated weights are set for the joint optimization objective function, with amplified weights for the data fidelity term in the core diagnostic region to prioritize optimization performance in this area. Then, the initial scan path parameters are used as the initial iteration values, and a matrix conforming to the monotonically decreasing fluorescence quenching rule is used as the initial quenching field parameter value. Adaptive iteration step-size rules, the maximum number of iterations, and a convergence threshold are set. Simultaneously, a global optimization mechanism based on simulated annealing is constructed to avoid getting trapped in local optima. Subsequently, a bivariate collaborative parallel iterative process is initiated. In each iteration, both the scan path parameters and the quenching field parameters are updated synchronously. Based on the rate of decrease in the objective function value of the current iteration, the iteration step-size is adaptively adjusted: when the objective function value decreases rapidly, the iteration step-size is increased to accelerate convergence; when the decrease slows, the iteration step-size is decreased to improve solution accuracy. Simultaneously, through the simulated annealing mechanism, parameter combinations with a preset probability of slightly increasing objective function values ​​are accepted to escape local optima. The method synchronously records the parameter combinations and objective function values ​​for each iteration, marking the optimal parameter combination corresponding to the global minimum objective function value. After each iteration, the objective function values ​​from multiple consecutive iterations are compared to determine if the preset convergence condition is met. When the objective function value from multiple consecutive iterations is consistently lower than the preset convergence threshold, or when the number of iterations reaches the maximum number of iterations, the iteration is terminated. The scanning path parameters and quenching field parameters corresponding to the global minimum objective function value are then determined as the initial path parameters and initial quenching field parameters, respectively. This method employs a bivariate collaborative adaptive step-size global optimization iterative logic, combining the convergence efficiency advantage of adaptive step-size with the global optimization capability of simulated annealing. Furthermore, it enhances the optimization effect of the core diagnostic region through regional differentiated weights. This addresses the industry pain points of conventional alternating iterative methods, such as being prone to getting trapped in local optima, failing to balance convergence speed and solution accuracy, and insufficient optimization of the core diagnostic region. It is suitable for solving pathological fluorescence image parameter problems involving multiple marker staining, complex artifact features, and high clinical diagnostic requirements.

[0135] In an exemplary scheme for determining initial path parameters and initial quenching field parameters, the sampled fluorescence image generated in the first embodiment, the scanning sequence field generated in the second embodiment, the joint optimization objective function constructed in the third embodiment, and the initialized scanning path parameters corresponding to the scanner are first loaded. Then, pixel-level temporal correlation set construction is performed. Based on the initialized scanning path parameters and combined with the temporal index of the scanning sequence field, a corresponding relative scanning time is matched for each pixel of the sampled fluorescence image. The pixel spatial coordinates, original fluorescence intensity values, relative scanning time, and scanning temporal index are bound one by one to generate pixel-level correlation tuples. All correlation tuples are then summarized to obtain a complete pixel-level temporal correlation set. Next, iteration rules are set, using the initialized scanning path parameters as the initial value for scanning path iteration, and the identity matrix conforming to the monotonically decreasing law of fluorescence quenching as the initial value for the quenching field parameters. The initial iteration step size is set to 0.1, the adaptive step size adjustment rule is set, the maximum number of iterations is 100, and the convergence threshold is 1e-6. Simultaneously, a simulated annealing global optimization mechanism is enabled. Subsequently, a bivariate collaborative iterative solution process is initiated. In each iteration, the scan path parameters and quenching field parameters are updated synchronously. The updated parameters and pixel-level temporal correlation sets are input into the joint optimization objective function to calculate the objective function value for the current iteration. This value is compared to the objective function value from the previous iteration, and the iteration step size is adaptively adjusted for the next iteration. Simultaneously, a simulated annealing mechanism is used to determine whether to accept the current parameter combination, avoiding getting trapped in local optima. The parameter combinations and objective function values ​​for each iteration are recorded synchronously, and the parameter combination corresponding to the global minimum objective function value is marked. After each iteration, a convergence check is performed, comparing the objective function values ​​from five consecutive iterations. If the differences are all less than a preset convergence threshold, or the number of iterations reaches 100, the iteration loop is terminated. Then, parameter validity is verified to confirm that the initial path parameters obtained match the scan temporal logic, and that the initial quenching field parameters conform to the physical law of monotonically decreasing fluorescence quenching, with no outliers or overfitting issues. After verification, the scanning path parameters corresponding to the global minimum objective function value during the iteration process are determined as the initial path parameters, and the corresponding quenching field parameters are determined as the initial quenching field parameters, which are used for subsequent upsampling mapping and high-resolution solution processes.

[0136] It should be noted that in some special cases, the objective function value may oscillate continuously during the iterative solution process, failing to meet the preset convergence condition. In such cases, the initial iteration values ​​will be automatically reset, the iteration step size and the weight coefficients of the joint optimization objective function will be adjusted, and the solution process will be restarted. If convergence still fails after multiple restarts, the initial scan path parameters will be used as the initial path parameters, and the initial quenching field matrix conforming to physical laws will be used as the initial quenching field parameters. Simultaneously, abnormal situations will be recorded for subsequent iterations of optimization rules. If the sampled fluorescence image is a multi-channel immunofluorescence image, an independent iterative solution process will be executed for each fluorescence channel to obtain the corresponding initial path parameters and initial quenching field parameters, ensuring that the calibration process of each channel does not interfere with each other.

[0137] Sixth Embodiment This embodiment provides an exemplary scheme for solving the target quenching field of a high-resolution digital pathological fluorescence image. In this example, the original path parameters and the initial quenching field parameters are first transformed by coordinate mapping and matched to the pixel coordinate system of the original fluorescence image to generate an original scanning sequence field. Then, the original fluorescence image is segmented based on the original scanning sequence field to obtain image regions of homogeneous tissue types. Next, the image regions of homogeneous tissue types are used as constraint units, and the original scanning sequence field is optimized by region segmentation according to the constraint units to obtain the target scanning sequence field. Finally, based on the target scanning sequence field and the physical attenuation law of fluorescence quenching, the target quenching field is obtained by solving the joint optimization objective function. Step S40 includes steps E11~E14: Step E11: The original path parameters and the initial quenching field parameters are transformed by coordinate mapping and matched to the pixel coordinate system of the original fluorescence image to generate the original scanning sequence field.

[0138] Step E12: Segment the original fluorescence image based on the original scanning sequence field to obtain image regions of homogeneous tissue type.

[0139] Step E13: Using the image regions of homogeneous tissue types as constraint units, optimize the original scanning sequence field by region according to the constraint units to obtain the target scanning sequence field.

[0140] Step E14: Based on the target scanning sequence field and the physical attenuation law of fluorescence quenching, the target quenching field is obtained by solving the joint optimization objective function.

[0141] The original scan sequence field is a two-dimensional matrix that perfectly matches the size of the original high-resolution fluorescence image after the original path parameters have been transformed by coordinate mapping. Each pixel position stores a corresponding scan time sequence index. It fully represents the scanning sequence and temporal distribution characteristics of all pixels in the original image and is the core high-resolution temporal reference for subsequent image segmentation, temporal field optimization, and quenching field solution. Examples include the full-field high-resolution scan time sequence index matrix, the original scan path matching temporal field, the pixel-level global scan sequence field, and the temporal reference matrix with region feature markers.

[0142] Constraint units are the smallest processing units that treat homogeneous image regions of each tissue type as independent optimization constraints. Each unit possesses independent tissue features, fluorescence statistical properties, diagnostic priorities, and optimization constraint rules, with no computational dependency among them. They are the core carriers for achieving regional parallel differentiated optimization. Examples include core diagnostic region priority constraint units, tissue boundary feature protection constraint units, normal tissue balance constraint units, background region simplification constraint units, and high quenching risk region enhancement constraint units.

[0143] The target scanning sequence field is a globally optimal high-resolution scanning timing matrix that perfectly matches the actual acquisition timing and tissue distribution characteristics of the original fluorescence image after regional differentiation optimization of the original scanning sequence field. The timing distribution conforms to the scanning physical logic and is globally continuous without abrupt changes, serving as the final timing benchmark for solving the target quenching field. Examples include regionally optimized globally optimal timing fields, tissue feature-adapted scanning sequence fields, high-resolution physically compliant timing index matrices, and precise timing fields prioritizing clinical diagnosis.

[0144] In this example, when transforming the original path parameters and initial quenching field parameters through coordinate mapping to generate the original scan sequence field, the sampling mapping method can be referenced. Alternatively, a tissue feature-guided differential mapping method can be used. First, the tissue segmentation results and core diagnostic region markers of the original fluorescence image are read. Differential mapping interpolation rules are set for the core diagnostic region, tissue boundary region, and blank background region to complete the coordinate mapping transformation of the parameters and generate the original scan sequence field with region priority markers, thereby completing the construction of a high-resolution temporal reference.

[0145] After constructing the original scanning sequence field, a homogeneous region segmentation process for the original fluorescence image is initiated. Combining the temporal gradient features of the original scanning sequence field, the fluorescence intensity features and texture features of the original fluorescence image, unsupervised segmentation with multi-feature fusion is performed to obtain image regions of multiple tissue types. Next, each homogeneous image region of tissue type is treated as an independent constraint unit, and suitable optimization constraint rules are set for each constraint unit. Using the minimization of the joint optimization objective function as the criterion, iterative optimization is performed on the original scanning sequence field in different regions. After all constraint units are optimized, global temporal smoothing fusion is performed to obtain a globally continuous target scanning sequence field. Finally, based on the full-image pixel-level scanning temporal sequence of the target scanning sequence field, combined with the physical attenuation law of fluorescence quenching, and using the minimization of the joint optimization objective function as the criterion, the attenuation model parameters of the entire image are fitted and solved to obtain the target quenching field covering the entire image. This approach, through regional differentiated constraints and parallel optimization, simultaneously improves the solution accuracy and computational efficiency of the high-resolution quenching field, solving the pain point in conventional globally unified solution methods where core diagnostic accuracy and global computational efficiency cannot be simultaneously achieved.

[0146] For example, there are two ways to obtain the target quenching field through regional optimization and joint solution. The first is a global mapping and coherent solution with global homogeneous constraints. First, the resolution ratio coefficient between the sampled fluorescence image and the original high-resolution fluorescence image is calculated. Based on this coefficient, a global equal-proportional linear coordinate mapping transformation is performed on the original path parameters and the initial quenching field parameters, matching them to the pixel coordinate system of the original fluorescence image. This generates an original scanning sequence field that perfectly matches the width and height of the original image and has a continuous temporal distribution, and a high-resolution initial quenching attenuation matrix is ​​generated simultaneously. Then, based on the temporal gradient characteristics of the original scanning sequence field, combined with the fluorescence intensity and texture characteristics of the original fluorescence image, global homogeneous tissue segmentation is performed to remove the blank background regions without signal, resulting in multiple image regions with homogeneous tissue types. Then, all homogeneous image regions are used as unified global constraint units. With minimizing the constructed joint optimization objective function as the core criterion, global iterative optimization is performed on the original scanning sequence field to correct scanning temporal deviations, resulting in a globally continuous target scanning sequence field that conforms to the scanning physical logic. Finally, based on the full-image pixel-level scanning time sequence of the target scanning sequence field, and with the fluorescence quenching single-exponential physical decay model as the core constraint, a global nonlinear least squares fitting solution is performed through joint optimization of the objective function to obtain the target quenching field covering the entire image. This method adopts a global homogeneous constraint-based global mapping and coherent solution logic, with a simple and stable process, good global consistency of the time sequence field, no temporal abrupt changes between regions, strong convergence of the solution process, and can quickly generate a high-resolution quenching field that conforms to physical laws. It is suitable for solving conventional pathological fluorescence images with uniform tissue distribution, single marker staining, and no complex pathological features.

[0147] The second approach is a diagnosis priority-oriented, region-specific, differentially constrained parallel optimization solution. First, the pathological feature recognition results of the original fluorescence image are read to locate the core diagnostic region containing tumor tissue and positive markers, the normal tissue region, the tissue boundary region, and the blank background region. Differentiated diagnostic priority weights are assigned to different regions. Simultaneously, the resolution ratio coefficient between the sampled fluorescence image and the original high-resolution fluorescence image is calculated, and differentiated coordinate mapping rules are configured for different regions: bicubic order-preserving interpolation is used for the core diagnostic region and the tissue boundary region to maximize the preservation of temporal details and boundary features; bilinear interpolation is used for the normal tissue region to balance accuracy and efficiency; and nearest-neighbor interpolation is used for the blank background region to reduce unnecessary computation. Based on the differential rules, a region-specific coordinate mapping transformation is performed on the original path parameters and the initial quenching field parameters, matching them to the pixel coordinate system of the original fluorescence image to generate an original scan sequence field with region priority labels. Subsequently, combining the temporal gradient features of the original scan sequence field, the intensity and texture features of the original fluorescence image, and the region priority markers, unsupervised segmentation with multi-feature fusion is performed to obtain multiple image regions with homogeneous tissue types and clear diagnostic priorities. Each region is treated as an independent constraint unit, and differentiated optimization rules and joint optimization objective function weights are set for different constraint units: the weights of data fidelity terms are amplified in core diagnostic regions to enhance temporal optimization accuracy; spatial smoothness constraints are strengthened in tissue boundary regions to protect pathological boundary features; monotonicity constraints are strengthened in high-quenching-risk regions to ensure physical compliance. Blank background regions simplify constraint rules and improve computational efficiency. Parallel iterative optimization is then initiated simultaneously for all constraint units, using the minimization of the joint optimization objective function as the criterion. Differential optimization is performed on the original scan sequence field within each constraint unit. After all constraint units are optimized, temporal smoothing transition processing is performed on the boundary regions of adjacent units to eliminate boundary temporal abrupt changes, resulting in a globally continuous target scan sequence field with adapted region features. Finally, based on the full-image pixel-level scanning time sequence of the target scanning sequence field, combined with the fluorescence quenching double-exponential physical attenuation model, and using the model regularization term of the joint optimization objective function as a hard constraint, the attenuation model parameters are solved by fitting the region, and the resulting target quenching field covers the entire image, conforms to physical laws, and prioritizes the accuracy of the core region. This method adopts a diagnostic priority-oriented regional differential constraint parallel optimization logic. Through pre-identification of pathological features and regional differential configuration, it prioritizes the solution accuracy of the core clinical diagnostic region. At the same time, it significantly reduces the solution time of large-size whole-slice images through parallel computing, solving the industry pain points of insufficient fidelity of core diagnostic information, low solution efficiency of large-size high-resolution images, and loss of tissue boundary details in conventional global unified solution methods. It is suitable for the solution scenario of clinical diagnostic-grade whole-slice pathological fluorescence images with multiple marker staining and complex pathological features.

[0148] In an exemplary scheme for determining the target quenching field, the initial path parameters and initial quenching field parameters obtained from the solution, the constructed joint optimization objective function, and the pixel coordinate system information and tissue feature recognition results of the original high-resolution fluorescence image are first loaded. Then, coordinate mapping transformation is performed, and a resolution scaling factor of 4 times that of the original image is calculated. Based on the division of the core diagnostic region, tissue boundary region, normal tissue region, and blank background region, differentiated interpolation rules are configured for different regions. Regional coordinate mapping transformation is performed on the original path parameters and initial quenching field parameters, matching them to the pixel coordinate system of the original fluorescence image, generating an original scanning sequence field that perfectly matches the size of the original image and has region priority labels. Next, homogeneous region segmentation is performed. An unsupervised clustering algorithm with multi-feature fusion is used, combining the temporal gradient features of the original scanning sequence field, the fluorescence intensity and texture features of the original fluorescence image, to divide the tissue foreground region of the original image into 8 image regions of homogeneous tissue types. The diagnostic priority of each region is simultaneously labeled, generating 8 independent constraint units, while the blank background region is retained as an independent constraint unit. Subsequently, parallel optimization is performed across regions. Differentiated joint optimization objective function weights and optimization rules are configured for each constraint unit. Parallel iterative optimization is initiated simultaneously for all constraint units, using the minimization of the joint optimization objective function as the criterion. Iterative optimization is performed on the original scan sequence field within each constraint unit. After all constraint units are optimized, temporal smoothing transitions are performed on the boundaries of adjacent units to eliminate boundary abrupt changes, generating a globally continuous target scan sequence field. Finally, quenching field solving is performed. Based on the full-image pixel-level scan temporal sequence of the target scan sequence field, combined with a fluorescence quenching double-exponential physical attenuation model, and using the model regularization term of the joint optimization objective function as a hard constraint, attenuation model parameters are solved by region-by-region fitting. The quenching attenuation coefficient corresponding to each pixel is calculated, and the resulting target quenching field covers the entire image. Simultaneously, a global physical compliance check is performed on the generated target quenching field to ensure that the attenuation coefficient conforms to the monotonically decreasing physical law. Once the check passes, it is used as the core benchmark for subsequent original image correction.

[0149] It should be noted that in some special cases, if the scanning sequence field optimization of certain tissue regions fails to meet the convergence condition, the constraint rules and iteration step size of that region will be automatically adjusted, and local optimization will be re-executed. If convergence still fails after multiple adjustments, the temporal parameters obtained from global optimization will be used to fill the region, and abnormal situations will be recorded. If the original fluorescence image is a multi-channel immunofluorescence image, an independent scanning sequence field optimization and quenching field solution process will be performed for each fluorescence channel to obtain the target quenching field corresponding to each channel, ensuring that the correction process of each channel does not interfere with each other. If the effective tissue region of the original fluorescence image accounts for a very low proportion, the constraint weight of the model regularization term will be automatically increased to strengthen the constraints of physical laws and avoid abnormal attenuation coefficients caused by overfitting.

[0150] Seventh Embodiment This embodiment provides an exemplary scheme for inverse correction of pixel intensity in digital pathological fluorescence images and generation of the corrected image. In this example, firstly, based on the pixel intensity information of the original fluorescence image, the quenching attenuation coefficient corresponding to the same spatial position in the target quenching field is matched. Then, after treating a single pixel as an independent computing unit, the pixel intensity information of the original fluorescence image is inversely corrected according to the principle of fluorescence quenching inverse compensation, combined with the quenching attenuation coefficient of the independent computing unit, to obtain the corrected pixel intensity. Next, the corrected pixel intensity is subject to reasonable range constraints to avoid the distortion of biofluorescence signals and loss of local details caused by overcorrection, thus obtaining the target pixel intensity. Finally, all the target pixel intensities are summarized to reconstruct the global image and generate the target fluorescence image. Step S50 includes steps F11~F14: Step F11: The original path parameters and the initial quenching field parameters are transformed by coordinate mapping and matched to the pixel coordinate system of the original fluorescence image to generate the original scanning sequence field.

[0151] Step F12: Segment the original fluorescence image based on the original scanning sequence field to obtain image regions of homogeneous tissue type.

[0152] Step F13: Using the image regions of homogeneous tissue types as constraint units, optimize the original scanning sequence field by region according to the constraint units to obtain the target scanning sequence field.

[0153] Step F14: Based on the target scanning sequence field and the physical attenuation law of fluorescence quenching, the target quenching field is obtained by solving the joint optimization objective function.

[0154] Pixel intensity information refers to the gray-scale quantification of the fluorescence signal at each pixel location in the original acquired digital pathological fluorescence image. It characterizes the fluorescence emission intensity of the pathological tissue at that location and is the core foundational data for inverse correction of fluorescence signals. Examples include 8-bit gray-scale pixel intensity values, 16-bit gray-scale pixel intensity values, single-channel fluorescence pixel intensity values, and multi-channel fluorescence sub-channel pixel intensity values.

[0155] An independent computational unit (ICU) is a single pixel in the original fluorescence image. It is formed by binding the pixel's spatial coordinates, original pixel intensity information, and corresponding quenching attenuation coefficient. This ICU enables precise pixel-level inverse correction, avoiding the blurring of tissue details and loss of spatial resolution caused by block-level correction. Examples include single-pixel single-channel computational units, single-pixel multi-channel parallel computational units, and pixel computational units with neighboring tissue feature markers.

[0156] The principle of fluorescence quenching reverse compensation is based on the physical attenuation law of fluorescence quenching. It eliminates the attenuation of fluorescence signal caused by continuous excitation light irradiation during scanning through inverse mathematical operations, restoring the true fluorescence emission intensity of pathological tissue. The core logic is that the corrected true fluorescence intensity of the tissue equals the original acquired intensity value divided by the quenching attenuation coefficient at the corresponding location. Examples include the linear attenuation reverse compensation principle, the single exponential attenuation reverse compensation principle, the double exponential attenuation reverse compensation principle, and the tissue feature-weighted reverse compensation principle.

[0157] Reverse correction calculation is based on the principle of fluorescence quenching reverse compensation. It uses a single pixel as an independent computing unit and performs mathematical operations between the original pixel intensity value and the corresponding quenching attenuation coefficient to eliminate the quenching attenuation effect caused by the scanning time sequence and restore the true fluorescence intensity of the tissue. Examples include linear reverse division correction, nonlinear neighborhood weighted reverse correction, tissue boundary constraint reverse correction, and multi-channel synchronous reverse correction.

[0158] Reasonable range constraints are rules that set effective value ranges and reasonable limits for pixel intensity values ​​after inverse correction. They are used to avoid problems such as overcorrection, intensity value overflow, and biological signal distortion caused by abnormal attenuation coefficients and local noise interference, thus ensuring the authenticity of pathological information and visual rationality of the corrected image. Examples include grayscale value truncation constraints, local neighborhood mean fluctuation constraints, tissue type-specific range constraints, and global contrast normalization constraints.

[0159] The target pixel intensity, obtained after inverse correction calculation and reasonable range constraints, represents the pixel intensity value that restores the true fluorescence emission intensity of pathological tissue, avoids overcorrection distortion, and preserves tissue pathological details. It is the smallest basic unit of the reconstructed and corrected digital pathological fluorescence image. Examples include single-channel target grayscale value, multi-channel sub-channel target intensity value, tissue boundary smooth transition intensity value, and high-fidelity intensity value of key diagnostic areas.

[0160] A global image is a complete two-dimensional digital image reconstructed by summarizing the target pixel intensities of all pixels in the entire image and strictly following the pixel coordinate system, row and column arrangement rules, and spatial resolution parameters of the original fluorescence image. It serves as the complete data carrier for the corrected digital pathological fluorescence image. Examples include single-channel global corrected images, multi-channel fused global corrected images, full-field high-resolution global images, and standardized digital pathological slide global images.

[0161] In this example, when matching the corresponding quenching attenuation coefficient based on the pixel intensity information of the original fluorescence image, the method of synchronous reading of all pixels in the image can be used. Alternatively, a streaming block incremental matching method can be adopted, reading the pixel intensity information of the original high-resolution image by scan line, and synchronously reading the attenuation coefficient data of the corresponding line in the target quenching field. The one-to-one matching and parameter binding of the pixel spatial position is completed row by row. The intensity and attenuation coefficient association data of the corresponding pixel are cached while matching, and the matching processing of all pixels in the image is completed. In this way, the memory occupation of large-size full-field pathological images is reduced while achieving accurate matching of attenuation coefficients at the pixel level, avoiding correction deviations caused by spatial misalignment.

[0162] After matching and binding the pixel intensity information of the entire image with the corresponding quenching attenuation coefficient, the pixel intensity inverse correction process is initiated. Each single pixel is treated as an independent computing unit. Based on the principle of fluorescence quenching inverse compensation, combined with the quenching attenuation coefficient bound to the unit, the original pixel intensity information is reverse corrected to obtain the corrected intensity value of each pixel. Then, reasonable range constraints are applied to the corrected pixel intensity value to eliminate outliers and avoid signal distortion caused by overcorrection, thus obtaining the target pixel intensity of each pixel. After the target pixel intensity of all pixels in the entire image is calculated, all target pixel intensities are summarized and reconstructed according to the pixel coordinate system and row and column arrangement rules of the original fluorescence image to obtain the global image. Finally, the target fluorescence image is generated. In this way, through pixel-level independent correction and targeted constraints, scanning quenching artifacts are eliminated while the diagnostic details of pathological tissue and the true fluorescence signal distribution are completely preserved.

[0163] For example, there are two ways to generate corrected digital pathological fluorescence images through pixel-level inverse correction calculation. The first is pixel-by-pixel linear inverse compensation serial correction. First, the full-image pixel intensity information of the original high-resolution fluorescence image is read, and the full-image attenuation coefficient data of the target quenching field is read simultaneously. According to the row priority order of the original fluorescence image, the quenching attenuation coefficient at the same spatial location is matched for each pixel, completing the one-to-one binding of the full-image pixels with the attenuation coefficient. Then, starting from the first row of the image, serial correction calculation is performed row by row and pixel by pixel. Each single pixel is treated as an independent operation unit. Based on the principle of fluorescence quenching linear inverse compensation, the corrected pixel intensity is calculated using the linear inverse correction formula: Corrected pixel intensity = Original pixel intensity value / Corresponding quenching attenuation coefficient. After completing the correction calculation for a single pixel, a reasonable range constraint is applied to the corrected pixel intensity. Intensity values ​​that exceed the maximum range of gray values ​​in the original fluorescence image are truncated, and abnormal intensity values ​​less than 0 are corrected to 0 to obtain the target pixel intensity. After completing the full-process processing of a single pixel, the correction calculation for the next pixel is then performed. After all pixels in the entire image have been corrected, the intensity of all target pixels is reconstructed according to the pixel coordinate system of the original fluorescence image to obtain a complete global image, and finally, the target fluorescence image is generated. This method adopts a calculation logic of pixel-by-pixel linear inverse compensation and row-first serial correction. The correction process is simple and stable, without neighborhood interference and parameter deviation, and can accurately restore the true fluorescence intensity of each pixel. It is suitable for the correction of conventional pathological images with uniform tissue distribution and single biomarker immunofluorescence staining.

[0164] The second method is a nonlinear weighted parallel correction with tissue boundary protection. First, the original high-resolution fluorescence image and the target quenching field are read. Simultaneously, the tissue-type homogeneous region segmentation results and tissue boundary masks are loaded, dividing the original fluorescence image into three feature regions: internal tissue regions, tissue boundary regions, and blank background regions. Differentiated correction constraints are set for different regions. Simultaneously, the image is divided into multiple computationally independent parallel correction units according to each region. Then, parallel correction calculations are initiated simultaneously for all parallel correction units. Within each correction unit, for pixels in the internal tissue region, each pixel is treated as an independent computational unit, and a linear inverse compensation formula is used to complete the correction calculation, ensuring the uniformity of the fluorescence signal within the tissue. For pixels in the tissue boundary region, a nonlinear weighted compensation model based on 3×3 neighborhood tissue features is constructed. Combining the quenching attenuation coefficient of the central pixel and the average attenuation coefficient of the same type of tissue in the neighborhood, nonlinear weighted inverse correction calculations are performed to avoid boundary blurring, contrast reduction, and loss of diagnostic details caused by linear correction. For pixels in the blank background region, a neighborhood smoothing constraint is superimposed during inverse correction to suppress excessive amplification of background noise. After completing the inverse correction calculation for all pixels, differentiated reasonable range constraints are applied to the correction results for different regions: For internal tissue regions, the full range of grayscale values ​​of the original fluorescence image is constrained to ensure the dynamic range of the fluorescence signal. For tissue boundary regions, the fluctuation range of the neighborhood mean is superimposed to avoid abrupt changes in boundary intensity and artifacts. For blank background regions, a low intensity upper limit constraint is applied to suppress background noise amplification. After processing, the target pixel intensity of all pixels in the entire image is obtained. After all parallel correction units have completed their processing, all target pixel intensities are summarized, and a smooth transition processing is performed on the boundaries of adjacent regions to reconstruct the entire image, ultimately generating the target fluorescence image. This method employs a nonlinear weighted calculation logic with tissue boundary protection and regional parallel correction. It can eliminate fluorescence quenching brightness unevenness artifacts while fully preserving the key tissue boundaries and cellular substructure details for pathological diagnosis. It solves the industry pain points of blurred boundaries, amplified background noise, and loss of key diagnostic information in conventional linear correction methods. At the same time, it significantly improves the correction efficiency of large-size, high-resolution pathological images through parallel computing, making it suitable for pathological fluorescence image correction scenarios involving multi-marker immunofluorescence staining and complex tissue types and fine cellular structures in clinical diagnosis.

[0165] Eighth embodiment This embodiment provides an exemplary scheme for self-iterative optimization of parameters in the digital pathology fluorescence image correction process. In this example, based on the spatially stable statistical distribution of fluorescence signals from similar biological tissues and the target quenching field, three quality evaluation indicators are calculated for the target fluorescence image: local statistical uniformity, biosignal restoration degree, and residual quenching artifacts. Then, based on these three quality evaluation indicators, the problem region of the target fluorescence image is located. Next, the weight correlation between the problem region and the data fidelity term and model regularization term in the joint optimization objective function is matched to determine the optimization direction and adjustment boundary of the weight parameters. Then, based on the optimization direction and adjustment boundary of the weight parameters, combined with the tissue type characteristics and corresponding scanning time sequence characteristics of the problem region, the weighting coefficients of the data fidelity term and model regularization term in the joint optimization objective function are iteratively adjusted to generate an optimized set of weight parameters. Finally, the optimized set of weight parameters is updated into the joint optimization objective function of the digital pathology fluorescence image correction process, completing the self-iterative optimization of the correction process parameters. Please refer to... Figure 4 , Figure 4 This is a flowchart illustrating the eighth embodiment of the digital pathological fluorescence image correction method of this application. Following step S50, steps G11-G15 are also included: Step G11: Based on the spatially stable statistical distribution of fluorescence signals from similar biological tissues and the target quenching field, calculate the local statistical uniformity, biological signal restoration degree, and quenching artifact residue of the target fluorescence image.

[0166] Local statistical homogeneity is a core quality evaluation index based on the spatially stationary statistical distribution of fluorescence signals from similar biological tissues. It is used to quantify the consistency of fluorescence intensity distribution within homogeneous tissue regions of the same type in a corrected image and is a key basis for measuring the effectiveness of local quenching artifact elimination. Examples include the mean variance of fluorescence intensity within a homogeneous tissue block, the coefficient of variation of local brightness, the stability index of intensity distribution within the same tissue region, and the difference in homogeneity between blocks.

[0167] Biological signal fidelity is a core quality assessment indicator used to quantify the degree of restoration of the true fluorescence signal of pathological tissue in a corrected image and the degree of matching between the corrected signal and the inherent fluorescence characteristics of the tissue. It is a core basis for measuring the fidelity of pathological information in a corrected image. Examples include the retention rate of the dynamic range of fluorescence signals in the tumor region, the fidelity of the signal intensity of positive markers, the retention rate of tissue boundary contrast, and the fidelity of cell substructure details.

[0168] Quenching artifact residue is a core quality evaluation indicator used to quantify the degree of residual fluorescence quenching brightness unevenness artifacts caused by differences in scanning timing that have not been completely eliminated in the corrected image. It is a key basis for measuring the effectiveness of the correction process. Examples include the brightness gradient residue value in the entire scanning direction, the brightness change amplitude in the scanning reversal area, the intensity attenuation residue rate in long-term scanning areas, and the proportion of artifact areas in the entire image.

[0169] In this embodiment, the above-mentioned self-iterative optimization process of correction parameters can be triggered in six ways. First, automatic triggering upon completion of single image correction: After the correction of a single digital pathology fluorescence image is completed and the corrected image is generated, the system automatically initiates the parameter self-iterative optimization process, adaptively optimizing the weight parameters based on the correction results for subsequent corrections of similar images. Second, triggering based on correction quality thresholds: When any one of the three indicators calculated by the system—local statistical uniformity, biosignal restoration, and residual quenching artifacts—fails to reach a preset qualified quality threshold, the parameter self-iterative optimization process is automatically initiated to specifically optimize the weight parameters to improve correction quality. Third, user-initiated triggering: When a pathologist or operator discovers problems such as residual artifacts or signal distortion in the corrected image through the digital pathology image reading system interface, and initiates a parameter optimization request, the system initiates the parameter self-iterative optimization process upon receiving the request. Fourth, batch archiving preprocessing trigger: When the system performs batch archiving tasks for pathological slide images, after completing the correction processing of each group of pathological slides of the same type, it automatically initiates a batch parameter self-iterative optimization process. Based on the batch correction results, it optimizes the weight parameters to improve the consistency of correction for subsequent batch images. Fifth, AI analysis pre-verification trigger: When the system receives a pathological image AI intelligent analysis task, it first performs quality verification on the corrected image. If the image quality indicators do not meet the input requirements of AI analysis, it automatically initiates a parameter self-iterative optimization process to optimize the correction parameters to generate high-quality images that meet the requirements of AI analysis. Sixth, batch scanning of samples in the same batch trigger: After the scanner completes the acquisition of multiple pathological slide images of the same batch, the same staining type, and the same scanning parameters, after the system completes the correction processing of the first few samples, it automatically initiates a parameter self-iterative optimization process. Based on the correction results of the previous samples, it optimizes the weight parameters to provide the optimal parameter benchmark for the correction of subsequent samples in the same batch.

[0170] Once the digital pathology fluorescence image correction device receives the parameter self-iterative optimization trigger command, it begins to acquire the target fluorescence image, the corresponding target quenching field, and the prior information of the spatially stationary statistical distribution of fluorescence signals of similar biological tissues, and initiates the calculation process of three types of quality evaluation indicators.

[0171] For example, the calculation methods for the local statistical uniformity, biosignal restoration, and quenching artifact residue of the corrected image include two types. The first type is a precise calculation of all indicators based on homogeneous region locking. Before the indicator calculation process officially starts, unsupervised homogeneous tissue segmentation is performed on the corrected image to divide it into several biological tissue regions of the same type and degree of differentiation and blank background regions, eliminating the interference of blank background regions on indicator calculation. Then, for each homogeneous tissue region, based on the spatially stable statistical distribution of fluorescence signals of the same type of biological tissue, the variance and coefficient of variation of fluorescence intensity in the region are calculated, and the calculation results of all homogeneous regions are summarized to obtain the local statistical uniformity of the whole image. Then, combined with the attenuation law of the original fluorescence image and the target quenching field, the matching degree between the corrected signal and the inherent fluorescence characteristics of the tissue and the tissue boundary contrast retention rate are calculated to obtain the biosignal restoration. Then, along the scanning path direction, the brightness gradient distribution of the corrected image is calculated, compared with the original quenching attenuation trend, and the brightness attenuation amplitude and abrupt change regions that were not eliminated are statistically analyzed to obtain the quenching artifact residue. This method eliminates interference from irrelevant backgrounds and different tissue types by locking in homogeneous tissue regions in the early stage. It has high accuracy in index calculation and can accurately reflect the true quality of the corrected image, providing a reliable quantitative basis for subsequent problem area localization.

[0172] The second method prioritizes weighted index calculation for key diagnostic regions. Before the formal index calculation process begins, the diagnostic region annotation information of the pathological image is retrieved, or a pre-trained pathological tissue recognition model is used to locate key regions in the image containing core diagnostic information such as tumor tissue and positive markers. Differential calculation weights are assigned to different regions: the highest calculation weight is set for key diagnostic regions to enhance their index contribution; a medium calculation weight is set for normal tissue regions; and the lowest calculation weight is set for blank background regions. Based on the set weights, the local statistical uniformity, biosignal restoration degree, and quenching artifact residue of the entire image are calculated respectively. The index calculation results for key diagnostic regions are amplified according to their weights, while the results for non-key regions are reduced according to their weights, ultimately resulting in three weighted quality evaluation indicators. This method aligns with the core needs of clinical pathological diagnosis, prioritizing the correction quality and evaluation accuracy of key diagnostic regions, avoiding interference from non-key regions, accurately locating correction issues affecting diagnostic results, and adapting to the parameter optimization needs of pathological images used for clinical diagnosis.

[0173] In one alternative approach, the system first retrieves pre-defined standardized index calculation rules and quality thresholds. Following pre-defined homogeneous region segmentation rules, the corrected image is uniformly divided and tissue type determined. Then, according to the standardized calculation rules, the local statistical homogeneity, biosignal restoration degree, and quenching artifact residue of each image block are calculated sequentially. The results for all three indexes of the entire image are then summarized. Simultaneously, the calculation results are compared with the pre-defined quality thresholds, and image blocks that do not meet the threshold requirements are marked. This method relies on the system's pre-defined standardized rules, has simple and stable calculation logic, and executes quickly, making it suitable for routine quality evaluation and parameter optimization needs of conventional pathological fluorescence images.

[0174] In another alternative approach, the tissue type features, scanning time sequence features, and staining type features of the corrected image are first extracted. Simultaneously, a priori distribution library of fluorescence signals from pathological images corresponding to the staining type and scanning mode is retrieved. Based on this priori distribution library, indicator calculation rules and weight allocation schemes adapted to the current image are dynamically generated. Then, based on the dynamically generated rules, the local statistical uniformity, biosignal restoration degree, and quenching artifact residue of the corrected image are calculated. Furthermore, a dynamic quality judgment threshold is generated based on the priori distribution library to complete the verification and anomaly marking of the indicator results. This method can dynamically generate suitable calculation rules based on the image's own features and corresponding prior information, resulting in higher adaptability and accuracy in indicator calculation. It can fully explore the correction quality issues of images with different staining types and scanning modes, adapting to the high-precision parameter optimization needs of complex staining and scanning scenarios.

[0175] In an exemplary scheme for determining three types of quality evaluation indicators, the target fluorescence image, the corresponding target quenching field obtained by solution, and the scanning path parameters used in this correction are first loaded. Simultaneously, a priori model of the spatially stationary statistical distribution of fluorescence signals of similar biological tissues, preset by the system, is retrieved. Image preprocessing is then performed: Gaussian filtering is applied to the corrected image to remove random noise; Otsu's thresholding method is used to segment the tissue foreground and blank background, generating a tissue foreground mask to eliminate interference from blank background regions on indicator calculations. Next, homogeneous tissue region segmentation is performed. A clustering algorithm is used, based on the fluorescence intensity and texture features of the tissue foreground region, to divide the tissue foreground into eight homogeneous tissue sub-regions. Small regions with areas smaller than a preset threshold are merged to obtain several independent homogeneous tissue calculation units. Subsequently, local statistical uniformity calculation is initiated. Each homogeneous tissue calculation unit is traversed, and the average, variance, and coefficient of variation of the fluorescence intensity of all pixels within the unit are calculated. A weighted average of the coefficients of variation for all units is calculated to obtain the local statistical uniformity of the corrected image. Next, the biosignal restoration calculation is initiated. Based on the attenuation coefficient distribution of the target quenching field, the dynamic range of the fluorescence signal in the corrected image is calculated and compared with the dynamic range of the original fluorescence image to obtain the signal dynamic range retention rate. The Cannibal edge detection operator is used to extract the tissue boundaries between the corrected image and the original fluorescence image. The number of boundaries and contrast are compared to obtain the tissue boundary contrast retention rate. The biosignal restoration is then calculated by weighting these two indicators. Next, the quenching artifact residue calculation is initiated. Based on the scanning path parameters, the main scanning direction of the image is determined. The average brightness change trend of the corrected image is calculated line by line along the main scanning direction, and a brightness change curve is fitted. This curve is compared with the theoretical attenuation curve of the target quenching field, and the sum of squared residuals of the two curves is calculated to obtain the overall brightness gradient residue value. Simultaneously, the scanning reversal area and long-term scanning area are located, and the brightness variation coefficient of these areas is calculated. The quenching artifact residue is then calculated by weighting the overall brightness gradient residue value. Finally, the calculated local statistical uniformity, biosignal restoration, and quenching artifact residue are summarized to generate a standardized set of quality evaluation indicators for subsequent problem area localization.

[0176] It should be noted that in some special cases, if all three quality evaluation indicators of the corrected image reach the preset optimal quality threshold and there is no correction problem that needs to be optimized, then the subsequent problem area localization and weight parameter optimization process is skipped, the current self-iterative optimization process is terminated directly, and the original weight parameters are retained unchanged.

[0177] Step G12: Based on the local statistical uniformity, the biosignal restoration degree, and the quenching artifact residue, locate the problem area of ​​the target fluorescence image.

[0178] In this embodiment, after the calculation of the three types of quality evaluation indicators is completed, the problem area localization process is initiated. Based on the calculation results of the three types of indicators and combined with the spatial distribution characteristics of the image, the problem area is accurately located.

[0179] For example, there are two methods for locating the problem area. The first is block-based location using multi-index threshold linkage. First, the corrected image is divided into several non-overlapping image blocks according to a preset size. For each image block, three sub-indicators are calculated: local statistical uniformity, biosignal restoration degree, and quenching artifact residue. Then, a corresponding pass / fail threshold is set for each sub-indicator. A multi-index linkage judgment rule is used: when any two sub-indicators of a single image block fail to reach the pass / fail threshold, or when a single sub-indicator significantly exceeds the threshold range, the image block is marked as a problem block. After judging all image blocks in the entire image, spatially adjacent problem blocks are merged, while isolated small problem blocks with an area smaller than the preset threshold are removed, finally obtaining the complete problem area. This method uses block-based multi-index linkage judgment, has a simple and stable location logic, can accurately pinpoint the spatial location and range of the problem area, has no missed detections, and is suitable for the location requirements of conventional correction problems.

[0180] The second method is feature matching localization guided by diagnostic priority. First, based on the global calculation results of three quality evaluation indicators, the core problem types in this correction are determined, such as artifact residue, overcorrection, and signal distortion. Then, combined with the distribution of key diagnostic regions in the image, differentiated judgment thresholds are set for different regions: for key diagnostic regions, the qualified thresholds are tightened to improve the sensitivity of problem detection and avoid missing minor problems affecting diagnosis. For non-critical normal tissue regions, standard qualified thresholds are used. For blank background regions, the qualified thresholds are relaxed to detect only severe artifact problems. Then, combined with the core problem types, image features and indicator features of the corresponding regions are extracted. Through feature matching, the problem regions corresponding to the core problem types are accurately located, and the problem type, severity, and diagnostic priority of each problem region are marked. This method, guided by diagnostic priority, can accurately locate the core problem regions affecting pathological diagnosis, avoid interference from irrelevant background regions, and has stronger targeting of the localization results, adapting to the problem localization needs of pathological images used for clinical diagnosis.

[0181] In one alternative approach, the system's preset problem judgment thresholds and segmentation rules are first retrieved. The corrected image is then uniformly segmented according to a fixed size. Each image block's three quality indicators are then compared against the preset thresholds, and all image blocks with substandard indicators are marked. Adjacent problem blocks are merged to obtain the final problem area. This method relies on fixed thresholds and segmentation rules, offering fast execution speed, intuitive logic, and suitability for the rapid problem localization needs of routine batch pathological images.

[0182] In another alternative approach, based on the scanning parameters, staining type, and tissue type used in this correction, a problem feature library of similar samples from historical correction tasks is retrieved. A classification model trained using this feature library is then input into the corrected image and three quality indicators. The model directly locates problem regions in the image through inference, simultaneously outputting the problem type and severity for each region. This method achieves intelligent problem region localization using a machine learning model based on historical data. It can adapt to complex, multi-type overlapping correction problems, offering higher localization accuracy and intelligence, and is suitable for problem localization needs in complex scanning scenarios and multi-marker stained images.

[0183] In an exemplary scheme for identifying a problem area, the three sets of quality evaluation indicators generated in the previous step, the corrected image, the homogeneous tissue region segmentation results, and the key diagnostic region identification results are first loaded. Then, multi-indicator linkage judgment rules are set, with primary and secondary warning thresholds for local statistical homogeneity, biosignal restoration, and quenching artifact residue, respectively. The primary threshold is the minimum requirement for indicator compliance, and the secondary warning threshold is the judgment boundary for potential problems. Next, full-image block processing is performed, dividing the corrected image into several non-overlapping image blocks of a fixed size of 16×16 pixels. Each image block is matched with corresponding homogeneous tissue region attributes and diagnostic priority attributes. Subsequently, for each image patch, three corresponding sub-indicators are extracted, and multi-indicator linkage judgment is performed: if the image patch is located within a critical diagnostic region and any one sub-indicator is below the secondary warning threshold, it is marked as a candidate problem patch; if the image patch is located within a normal tissue region and any two sub-indicators are below the primary acceptable threshold, or a single sub-indicator is significantly below the primary acceptable threshold, it is marked as a candidate problem patch; if the image patch is located within a blank background region and all three sub-indicators are significantly below the primary acceptable threshold, it is marked as a candidate problem patch. After judging all image patches in the entire image, connected component merging is performed on spatially adjacent candidate problem patches to generate several candidate problem regions. Then, area verification is performed on each candidate problem region to remove isolated small regions with an area smaller than 3×3 pixels. At the same time, for each retained problem region, its problem type, severity, tissue type attribute, and diagnostic priority attribute are marked, and finally a complete set of problem regions is generated for subsequent weight association matching and parameter optimization.

[0184] It should be noted that in some special cases, if the quality indicators of all regions in the entire map meet the requirements and no problem areas are located, the current self-iterative optimization process will be terminated directly, and the original weight parameters will remain unchanged.

[0185] Step G13: Match the weight correlation between the problem region and the data fidelity term and model regularization term in the joint optimization objective function to determine the optimization direction and adjustment boundary of the weight parameters.

[0186] Weight correlation is the mapping relationship between the type and cause of the correction problem in the problem region and the weight parameters of the data fidelity term and the model regularization term in the joint optimization objective function. It is used to clarify the direction of weight parameter adjustment for different correction problems and is the core mapping basis for parameter optimization. For example, there is a positive correlation between the artifact residue region and the weight of the data fidelity term, a positive correlation between the overcorrection region and the weight of the model regularization term, and a cross-correlation relationship between the boundary distortion region and the weights of the two terms.

[0187] In this embodiment, after the problem area is located and its attributes are marked, the weight association matching process is initiated. Based on the problem type and cause of the problem area, the corresponding weight association is matched, thereby determining the optimization direction and adjustment boundary of the weight parameters.

[0188] For example, there are two methods for matching weight association relationships and determining the optimization direction. The first method is matching problem types with a weight mapping rule base. A pre-built rule base is constructed, storing the corresponding mapping relationships between different correction problem types and the weights of data fidelity items and model regularization items. For example, for problems like quenching artifact residue and insufficient local uniformity, the weight of the data fidelity item is increased, and the weight of the model regularization item is moderately decreased. For problems like over-correction and signal distortion, the weight of the model regularization item is increased, and the weight of the data fidelity item is moderately decreased. For problems with blurred organizational boundaries, the proportion of regional difference constraints for both weights is increased simultaneously. Then, based on the problem type of the located problem area, the corresponding mapping relationship in the rule base is directly matched to clarify the optimization direction of the weight parameters. Simultaneously, based on the severity of the problem, the adjustment range and upper and lower limits of the weight parameters are set. This method relies on a pre-built rule base, has simple and stable matching logic, and a clear adjustment direction. It can quickly determine the optimization direction and adjustment boundaries, adapting to the parameter optimization needs of common single-type correction problems.

[0189] The second approach is dynamic correlation matching that traces the root cause of the problem backward. First, for the identified problem area, the entire calibration process is traced backward, including the scan path optimization, quenching attenuation field solution, and original weight parameter settings. By combining the tissue type characteristics and scan sequence characteristics of the problem area, the root cause of the problem is analyzed. This clarifies whether the artifacts are due to insufficient data fidelity constraints, over-calibration due to excessive model regularization constraints, or local calibration failure due to an imbalance in the weight ratio. Based on the root cause, the correlation between the problem area and the two weights is dynamically constructed, clarifying the optimization direction of the weight parameters. Simultaneously, considering the diagnostic priority of the problem area and the overall calibration quality of the entire image, adjustment boundaries for the weight parameters are set to prevent local weight adjustments from deteriorating the overall calibration effect. This method, through backward tracing of the problem's root cause, dynamically constructs precise weight correlations, adapting to calibration problems with multiple overlapping types and complex causes. It offers higher accuracy in optimization direction, avoids fluctuations in calibration effects caused by blind adjustments, and meets the precise parameter optimization needs of complex calibration problems.

[0190] In one alternative approach, the system first retrieves a pre-defined standardized problem and weight mapping rule base. Based on the problem type of the problem region, it directly matches the corresponding weight optimization direction and adjustment boundary in the rule base, without requiring additional causal analysis, and directly outputs the optimization direction and adjustment boundary. This method is fast, has simple and controllable logic, and is suitable for the rapid parameter optimization needs of routine batch pathological images.

[0191] In another alternative approach, based on the features of the problem region, overall image quality indicators, and the parameters of the entire correction process, a pre-trained weight parameter optimization decision model is retrieved. The aforementioned data is then input into the decision model, and through model inference, the optimization direction and adjustment boundary of the data fidelity term and model regularization term weights are directly output, along with the recommended adjustment step size. This method, based on a machine learning model, achieves intelligent association matching and optimization direction determination. It can adapt to complex multi-problem scenarios, offering greater accuracy and adaptability in optimization decisions, and meeting the high-precision, high-requirement needs of clinical pathology image correction parameter optimization.

[0192] In an exemplary scheme for determining the direction of weight optimization and adjustment boundaries, the problem region set generated in the previous step, the original weight parameters of the joint optimization objective function used in this correction, and the preset problem-weight mapping rule library are first loaded. Then, for each problem region, problem type and cause analysis is performed. Based on the quality indicator characteristics, organizational type characteristics, and scan time sequence characteristics of the problem region, the core problem type is determined to be one or more of the following: quenching artifact residue, overcorrection, signal distortion, and boundary ambiguity. Simultaneously, the optimization process of this region in this correction process is traced back to clarify the core cause of the problem. Next, weight correlation matching is performed. For the quenching artifact residue problem caused by insufficient data fidelity constraints, a correlation is found to increase the weight of the data fidelity term and appropriately decrease the weight of the model regularization term. For the overcorrection and signal distortion problems caused by excessive model regularization constraints, a correlation is found to increase the weight of the model regularization term and appropriately decrease the weight of the data fidelity term. For the boundary ambiguity problem caused by an imbalance in the weight ratio of the two components, a correlation is found to simultaneously adjust the two weights and strengthen the regional differentiation constraints. Subsequently, the optimization direction was determined. Based on the correlations obtained from matching, and considering the severity of the problem and the diagnostic priority, the specific adjustment direction and initial adjustment range of the weighting coefficients for data fidelity and model regularization were clarified. Next, adjustment boundaries were set. Based on the system's preset safe weight value range, the minimum lower limit and maximum upper limit of the weights for data fidelity and model regularization were set. A fixed constraint boundary was also set so that the sum of the two weights is 1, preventing weight adjustments from exceeding a reasonable range. For problems in key diagnostic areas, the upper limit of the adjustment range was appropriately relaxed to ensure the optimization effect in core areas. For problems in non-critical areas, the upper limit of the adjustment range was tightened to avoid local adjustments affecting the overall correction stability of the image. Finally, the weight optimization directions and unified adjustment boundaries for all problem areas were summarized and generated for subsequent iterative adjustments of the weight coefficients.

[0193] It should be noted that in some special cases, the identified problem area is an isolated, atypical, or abnormal area without a corresponding standardized weight correlation. In such cases, the current weight optimization process is skipped, the original weight parameters are retained, and the abnormal area and problem features are entered into the rule base for subsequent rule iteration optimization.

[0194] Step G14: Based on the optimization direction and adjustment boundary of the weight parameters, and combined with the organizational type characteristics and corresponding scanning time sequence characteristics of the problem region, the weighting coefficients of the data fidelity term and the model regularization term in the joint optimization objective function are iteratively adjusted to generate the optimized weight parameter set.

[0195] In this embodiment, after determining the optimization direction and adjustment boundary of the weight parameters, the iterative adjustment process of the weighting coefficients is initiated. Combining the organizational type characteristics and scanning timing characteristics of the problem region, the weight coefficients are iteratively optimized within the adjustment boundary, and finally the optimized weight parameter set is generated.

[0196] For example, there are two methods for iterative adjustment of weighted coefficients and generation of weight parameter sets. The first is global weight closed-loop iterative optimization. The original weight parameters are used as the initial values ​​for iteration. The iteration step size and maximum number of iterations are set. Within the adjustment boundary, the global weighted coefficients of the data fidelity term and the model regularization term are adjusted successively according to the determined optimization direction. After each weight adjustment, based on the adjusted weight parameters, a simulated correction is performed on the problem region. The quality index of the problem region after simulated correction is calculated, and it is determined whether the quality index has improved. If the quality index continues to improve, iterative adjustment continues. If the quality index reaches the qualified threshold, or the number of iterations reaches the maximum value, the iteration terminates, and the current weight parameters are used as the globally optimal weight parameters. Combined with the basic adaptation coefficients for different tissue types and scan sequences, a complete optimized weight parameter set is generated. This method uses closed-loop iterative optimization of global weights. The iteration process is stable and controllable, and it can quickly find the optimal global weight parameters that fit the entire image. This addresses the parameter optimization needs of similar correction problems that are common across the entire image.

[0197] The second approach is regionally differentiated weighted collaborative iterative optimization. First, based on the tissue type and scanning timing characteristics of the problem area, the problem area is divided into different optimization units, such as tumor tissue optimization units, normal tissue optimization units, scanning reversal area optimization units, and long-term scanning area optimization units. Then, for each optimization unit, based on the corresponding optimization direction and adjustment boundary, independent initial iteration values, iteration step sizes, and adjustment sub-boundaries are set, and parallel iterative adjustments of all optimization units are initiated simultaneously. After each iteration, simulated correction is performed on each optimization unit, the quality index of the corresponding region is calculated, the optimization effect is judged, and the global compatibility of the weight parameters of different units is verified to avoid conflicts in the correction effects of adjacent regions caused by local weight adjustments. Once the quality index of all optimization units reaches the qualified threshold and the global compatibility verification is passed, the iteration is terminated. The differentiated weight parameters of each optimization unit are then merged with the global basic weight parameters to generate a complete optimized weight parameter set containing global weights, tissue type differentiated weights, and scanning timing graded weights. This approach employs collaborative iterative optimization with differentiated weights for different regions, enabling precise weight adjustments based on the characteristics of different problem areas while maintaining consistency in global correction results. It addresses the pain point that a globally uniform weight cannot meet the correction needs of different regions, thus adapting to the precise parameter optimization requirements of multi-region and multi-type correction problems.

[0198] In one alternative approach, based on the determined optimization direction and adjustment boundaries, a single weight adjustment is performed using the original weight parameters as a foundation and following a preset fixed step size. This directly generates the optimized weight parameters without requiring multiple iterations. The adjusted weight parameters are then validated to ensure they remain within the adjustment boundaries before generating the final weight parameter set. This method offers a simple adjustment process, fast execution speed, and suitability for general parameter optimization needs with relatively minor issues and clearly defined optimization directions.

[0199] In another alternative approach, an objective function for optimizing the weight parameters is first constructed, with the goal of maximizing the quality index of the problem region. The adjustment boundary of the weight parameters serves as a constraint. An adaptive moment estimation algorithm is used to iteratively solve for the optimal combination of weight parameters within the constraints. Simultaneously, considering the tissue type and scanning time sequence characteristics of the problem region, corresponding differentiated weight sub-coefficients are generated. Finally, these are fused to generate a complete optimized set of weight parameters. This method uses a numerical optimization algorithm to solve for the optimal weight parameters, achieving high optimization accuracy and fast convergence. It is adaptable to complex weight optimization scenarios with multiple variables and constraints, and meets the high-precision clinical pathology image correction parameter optimization requirements.

[0200] In an exemplary scheme for generating an optimized set of weight parameters, the weight optimization direction, adjustment boundary, problem region set determined in the previous step, as well as the original weight parameters and joint optimization objective function model for this correction, are first loaded. Then, optimization unit partitioning is performed. Based on the tissue type characteristics and scanning time sequence characteristics of the problem regions, all problem regions are divided into four independent optimization units: tumor tissue optimization units, normal tissue optimization units, scanning reversal zone optimization units, and long-term scanning zone optimization units. Simultaneously, a corresponding optimization direction, adjustment sub-boundary, and iteration step size are matched to each optimization unit. Next, iterative optimization rules are set, with a maximum of 30 iterations. The convergence condition is that the quality indicators of all optimization units reach the qualified threshold, or that the quality indicators show no significant improvement after three consecutive iterations. Multi-unit collaborative iterative optimization is initiated. Using the original weight parameters as initial values, within the adjustment sub-boundary of each optimization unit, the differentiated weight coefficients and global basic weight coefficients of the corresponding units are simultaneously adjusted according to the corresponding optimization direction. After each iteration, based on the current weight parameters, simulated correction is performed on all optimization units. The corrected quality index for each unit is calculated to determine if the index has improved. Simultaneously, the compatibility of weight parameters between adjacent optimization units is checked to avoid abrupt changes in boundary correction. If the quality index continues to improve after this iteration, the next iteration continues. If the quality index of all optimization units reaches the acceptable threshold or the maximum number of iterations is reached, the iteration cycle terminates. Subsequently, the validity of the weight parameters is verified to confirm that the weight parameters obtained through iteration are within the preset adjustment boundaries and that the overall quality index after full-map simulated correction does not decrease. After successful verification, the global basic weight parameters obtained through iteration, the organization type-differentiated weight parameters of each optimization unit, and the scan time sequence-level weight parameters are summarized and integrated to generate a standardized set of optimized weight parameters for subsequent correction process updates.

[0201] It should be noted that in some special cases, if the quality indicators of the problem area do not improve after iterative adjustments, or even if the quality of the whole map correction decreases, then the current iteration optimization will be terminated, the original weight parameters will remain unchanged, and the relevant data of this optimization will be entered into the system for subsequent optimization rule iterations.

[0202] Step G15: Update the optimized set of weight parameters into the joint optimization objective function of the digital pathology fluorescence image correction process to complete the parameter self-iterative optimization of the correction process.

[0203] In this embodiment, after the optimized weight parameter set is generated and its validity is verified, the parameter update process of the correction process is started to update the optimized weight parameter set into the joint optimization objective function, thus completing this parameter self-iterative optimization.

[0204] For example, the implementation of parameter updates and self-iterative optimization includes two methods. The first is a full-process global parameter coverage update, where the optimized global basic weight parameters from the weight parameter set are directly updated to the joint optimization objective function of the digital pathology fluorescence image correction process, replacing the original weight parameters. Simultaneously, the optimized tissue type differentiation weights and scan time sequence grading weights are updated to the regional optimization module of the correction process, serving as the default parameters for subsequent corrections of similar images. After the parameter update is completed, the changes in quality indicators and parameter adjustments before and after this iteration are recorded, generating an iteration optimization log, thus completing the self-iterative parameter optimization of this correction process. This method uses a global parameter coverage update, which is simple and stable, and can quickly apply the optimized parameters to subsequent correction processes, adapting to the continuous correction needs of conventional pathological images of the same type.

[0205] The second approach involves scenario-based parameter classification and storage with adaptive retrieval and updates. First, the optimized weight parameter set is tagged with scenario-specific features such as staining type, scanning mode, tissue type, and scanner model. Then, the tagged weight parameter set is stored in the system's weight parameter library, without directly overwriting the global default parameters. When a new pathological image correction task is executed, the scenario features of the image to be corrected are extracted and matched with the weight parameter set tags in the parameter library. The optimized weight parameter set with the highest fit is automatically retrieved and loaded into the joint optimization objective function, achieving adaptive parameter optimization of the correction process. Simultaneously, based on the results of each correction, the weight parameter set in the parameter library is continuously iteratively optimized, completing the cumulative self-iterative optimization of the correction process. This approach, employing scenario-based classification and storage with adaptive retrieval, can match the optimal weight parameters for different correction scenarios, avoiding a decrease in adaptability to some scenarios caused by global parameter updates. It can simultaneously adapt to the pathological image correction needs of multiple staining types and scanning modes, improving the scenario adaptability of the correction process.

[0206] In one alternative approach, the optimized set of weight parameters is directly applied to the joint optimization objective function of the calibration process, replacing the original weight parameters. After the parameter update is complete, the current iterative optimization process is terminated. This method offers a simple update process, fast execution speed, and suitability for continuous calibration needs of single scenarios and similar samples.

[0207] In another alternative approach, the optimized set of weight parameters is first set as candidate parameters for the calibration process. In subsequent calibration tasks, dual-path calibration is performed simultaneously using both the original parameters and the optimized candidate parameters. The quality metrics of the two calibration results are compared. If the candidate parameters consistently outperform the original parameters, they are officially updated to the default parameters. If there is no advantage, the original parameters are retained. This method employs a progressive update through dual-path comparison, avoiding fluctuations in calibration results caused by parameter updates, ensuring the stability of the calibration process, and meeting the high-stability calibration requirements of pathological images used in clinical diagnosis.

[0208] In an exemplary scheme for completing the self-iterative optimization of calibration process parameters, the optimized weight parameter set generated in the previous step, the original fluorescence image features, scanning parameters, staining type information, and the system's weight parameter library and calibration process core configuration file are first loaded. Then, parameter scenario labeling is performed, adding corresponding scenario feature labels such as staining type, scanning mode, tissue type, scanner model, and slide type to the optimized weight parameter set. Simultaneously, the changes in quality indicators before and after optimization, the optimization effect on problem areas, and a parameter optimization description document are recorded. Next, a secondary validation of parameter validity is performed. Three test pathological images of the same scenario and type are selected, and calibration is performed using both the original weight parameters and the optimized weight parameter set. The quality indicators of the two sets of calibration results are compared to confirm that the optimized weight parameter set can stably improve calibration quality without any reverse deterioration. After successful validation, the labeled optimized weight parameter set is stored in the system's weight parameter library. Simultaneously, the default weight parameters of the joint optimization objective function of the calibration process are updated, and the optimized global basic weight parameters are set as the default parameters for image calibration in the same scenario. Subsequently, the regional optimization rules of the calibration process are updated, and the optimized tissue type differentiation weights and scan time sequence classification weights are updated to the rule base of the regional optimization module. Finally, a complete log of this parameter self-iterative optimization is generated, recording the parameter adjustment content, optimization effect, and scene adaptation information, completing the parameter self-iterative optimization of this calibration process and providing a better parameter benchmark for subsequent pathological fluorescence image calibration.

[0209] It should be noted that in some special cases, the optimized weight parameter set is only effective for this single image and does not have a stable optimization effect on other test images of the same type. In such cases, the global default parameters of the calibration process will not be updated. Instead, the weight parameter set will be stored in the parameter library and marked as single-sample dedicated parameters for subsequent recalibration of the same slice, so as to avoid affecting the overall stability of the calibration process.

[0210] This application provides a digital pathological fluorescence image correction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the digital pathological fluorescence image correction method in Embodiment 1 above.

[0211] The following is for reference. Figure 5 The diagram illustrates a structural schematic of a correction device suitable for implementing digital pathological fluorescence images in the embodiments of this application. The correction device for digital pathological fluorescence images in the embodiments of this application may include, but is not limited to, mobile terminals such as dedicated fluorescence image correction standard devices, fluorescence calibration slides, and metrological-grade universal calibration standards, as well as fixed terminals such as digital pathological slide scanning systems and pathological slide scanners. Figure 5 The digital pathological fluorescence image correction device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0212] like Figure 5 As shown, the digital pathology fluorescence image correction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the digital pathology fluorescence image correction device. The processing unit 1001, the read-only memory 1002, and the RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the digital pathology fluorescence image correction device to communicate wirelessly or wiredly with other devices to exchange data. Although a digital pathology fluorescence image correction device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0213] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0214] The digital pathological fluorescence image correction device provided in this application, employing the digital pathological fluorescence image correction method described in the above embodiments, can solve the technical problem of distortion in pathological quantitative analysis results. Compared with the prior art, the beneficial effects of the digital pathological fluorescence image correction device provided in this application are the same as those of the digital pathological fluorescence image correction method provided in the above embodiments, and other technical features of the digital pathological fluorescence image correction device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0215] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0216] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0217] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the correction method for digital pathological fluorescence images in the above embodiments.

[0218] The aforementioned computer-readable storage medium may be included in a digital pathological fluorescence image correction device; or it may exist independently and not incorporated into a digital pathological fluorescence image correction device.

[0219] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a digital pathological fluorescence image correction device, cause the digital pathological fluorescence image correction device to: downsample the original fluorescence image to obtain a sampled fluorescence image, wherein the resolution of the sampled fluorescence image is lower than that of the original fluorescence image; solve for the initial path parameters and initial quenching field parameters of the sampled fluorescence image based on initial scan path parameters and a joint optimization objective function, wherein the joint optimization objective function aims to minimize the objective function value; upsample and map the initial path parameters to the image coordinates of the original fluorescence image to obtain the original path parameters; solve for the target quenching field of the original fluorescence image based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function; and correct the original fluorescence image according to the target quenching field to obtain the target fluorescence image.

[0220] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0221] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0222] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described correction method for digital pathological fluorescence images, thereby solving the technical problem of distortion in pathological quantitative analysis results. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the correction method for digital pathological fluorescence images provided in the above embodiments, and will not be repeated here.

[0223] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for correcting digital pathological fluorescence images, characterized in that, The method includes: A sampled fluorescence image is obtained by downsampling the original fluorescence image, wherein the resolution of the sampled fluorescence image is lower than that of the original fluorescence image; The initial path parameters and initial quenching field parameters of the sampled fluorescence image are solved based on the initial scan path parameters and the joint optimization objective function, wherein the joint optimization objective function aims to minimize the objective function value. The initial path parameters are upsampled and mapped to the image coordinates of the original fluorescence image to obtain the original path parameters; The target quenching field of the original fluorescence image is solved based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function. The original fluorescence image is corrected according to the target quenching field to obtain the target fluorescence image.

2. The correction method for digital pathological fluorescence images as described in claim 1, characterized in that, Before the step of solving for the initial path parameters and initial quenching field parameters of the sampled fluorescence image based on the initial scan path parameters and the joint optimization objective function, and before the step of minimizing the objective function value as the optimization objective function, the correction method for digital pathological fluorescence images further includes: The scanning path parameterization model is initialized based on the pixel coordinate system and image size of the sampled fluorescence image to obtain a scanning path framework adapted to the sampled fluorescence image. Based on the scanning path framework, the corresponding scanning relative time is matched for each pixel in the two-dimensional pixel coordinate system of the sampled fluorescence image, and the mapping relationship between the scanning path and the scanning relative time is constructed. The scanning sequence field is obtained by matching the scanning time index corresponding to the pixel in the sampled fluorescence image with the mapping relationship.

3. The correction method for digital pathological fluorescence images as described in claim 2, characterized in that, After the step of matching the scan time index corresponding to the pixel in the sampled fluorescence image through the mapping relationship to obtain the scan sequence field, the correction method for the digital pathological fluorescence image further includes: The fluorescence intensity of each pixel in the sampled fluorescence image is correlated with the scanning time sequence in the scanning sequence field to obtain a correlation dataset of fluorescence intensity and scanning time sequence; Based on the associated dataset and the spatially stationary statistical distribution of fluorescence signals from similar biological tissues, a data fidelity term is constructed to measure the local statistical uniformity of the corrected image. Based on the physical decay law of fluorescence quenching and the associated dataset, constraints are applied to the spatial gradient of the quenching decay field, and a model regularization term is constructed to constrain the smoothness and monotonicity of the quenching decay field. The data fidelity term and the model regularization term are weighted and fused to obtain the joint optimization objective function used to simultaneously optimize the scan path and the quenching attenuation field.

4. The correction method for digital pathological fluorescence images as described in claim 3, characterized in that, The step of constructing a data fidelity term to measure the local statistical uniformity of the corrected image based on the associated dataset and the spatially stationary statistical distribution of fluorescence signals from similar biological tissues includes: Based on the spatially stationary statistical distribution of fluorescence signals from the same type of biological tissue, the sampled fluorescence image is uniformly divided into blocks to obtain local image blocks. For each local image block, the average value of the corrected fluorescence intensity of all pixels within the local image block is calculated to obtain the local average intensity. The fluorescence intensity variance of the local image patch is calculated based on the difference between the average value of each fluorescence intensity and the corresponding local intensity mean. The average value of the fluorescence intensity variance of all the local image patches is used to obtain the data fidelity term.

5. The method for correcting digital pathological fluorescence images as described in claim 1, characterized in that, The step of solving for the initial path parameters and initial quenching field parameters of the sampled fluorescence image based on the initial scan path parameters and the joint optimization objective function, wherein the joint optimization objective function aims to minimize the objective function value, includes: Based on the initial scan path parameters, the relative scan times of each pixel in the sampled fluorescence image are matched to construct a pixel-level temporal correlation set; The pixel-level temporal correlation set is iteratively processed by the joint optimization objective function, and the path parameters and quenching field parameters are updated round by round, and the corresponding objective function values ​​are calculated. By comparing the objective function values ​​obtained through continuous iterations, once the objective function value satisfies the convergence condition, the path parameters and quenching field parameters corresponding to the objective function value are determined as the initial path parameters and initial quenching field parameters.

6. The method for correcting digital pathological fluorescence images as described in claim 1, characterized in that, The step of solving the target quenching field of the original fluorescence image based on the original path parameters, the initial quenching field parameters, and the joint optimization objective function includes: The original path parameters and the initial quenching field parameters are transformed by coordinate mapping and matched to the pixel coordinate system of the original fluorescence image to generate the original scanning sequence field; The original fluorescence image is segmented based on the original scanning sequence field to obtain image regions of homogeneous tissue type; Using image regions of homogeneous tissue types as constraint units, the original scanning sequence field is optimized by region division according to the constraint units to obtain the target scanning sequence field; Based on the target scanning sequence field and the physical attenuation law of fluorescence quenching, the target quenching field is obtained by solving the joint optimization objective function.

7. The method for correcting digital pathological fluorescence images as described in claim 1, characterized in that, The step of correcting the original fluorescence image based on the target quenching field to obtain the target fluorescence image includes: Based on the pixel intensity information of the original fluorescence image, the quenching attenuation coefficient corresponding to the same spatial location in the target quenching field is matched; Based on the principle of reverse compensation for fluorescence quenching and the quenching attenuation coefficient, the pixel intensity information of the original fluorescence image is reversely corrected to obtain the corrected pixel intensity. By subjecting the corrected pixel intensity to a reasonable range constraint, the distortion of biofluorescence signals and loss of local details caused by overcorrection are eliminated, and the target pixel intensity is obtained. The target fluorescence image is obtained by reconstructing the global image by summarizing the intensities of all the target pixels.

8. The method for correcting digital pathological fluorescence images as described in claim 1, characterized in that, After the step of correcting the original fluorescence image according to the target quenching field to obtain the target fluorescence image, the correction method for the digital pathological fluorescence image further includes: Based on the spatially stationary statistical distribution of fluorescence signals from similar biological tissues and the target quenching field, the local statistical uniformity, biological signal restoration degree, and quenching artifact residue of the target fluorescence image are calculated. Based on the local statistical uniformity, the biosignal fidelity, and the residual amount of quenching artifacts, the problem region of the target fluorescence image is located; Match the weight correlation between the problem region and the data fidelity term and model regularization term in the joint optimization objective function to determine the optimization direction and adjustment boundary of the weight parameters; Based on the optimization direction and adjustment boundary of the weight parameters, and combined with the organizational type characteristics and corresponding scanning time sequence characteristics of the problem region, the weighting coefficients of the data fidelity term and the model regularization term in the joint optimization objective function are iteratively adjusted to generate an optimized set of weight parameters. The optimized set of weight parameters is updated into the joint optimization objective function of the digital pathology fluorescence image correction process to complete the parameter self-iterative optimization of the correction process.

9. A correction device for digital pathological fluorescence images, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the correction method for digital pathological fluorescence images as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the correction method for digital pathological fluorescence images as described in any one of claims 1 to 8.

Images