A method and device for cross-slice association of multi-stain glomeruli based on tissue segmentation

By organizing block division and cost distance calculation, combined with the Hungarian algorithm and transformation techniques, the mismatch problem in glomerular cross-slice association was solved, achieving accurate matching and stable association of glomeruli in different stained slices, thus improving diagnostic efficiency and accuracy.

CN123072784APending Publication Date: 2026-11-10SHENZHEN SHENGQIANG TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611578602.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies suffer from cross-tissue mismatch and unstable association results in glomerular cross-slice association, especially in cases of global registration and local deformation of kidney tissue slices, making it difficult to accurately match the same glomerulus, leading to increased diagnostic errors and time consumption.

Method used

By employing a tissue segmentation approach, registration groups are constructed by calculating the cost distance and center representation score between tissue blocks. The Hungarian algorithm is then used to match tissue blocks, and affine transformation and thin-plate spline transformation are combined to ensure accurate association of glomeruli in different stained sections.

Benefits of technology

It improves the accuracy and stability of multichromosomal glomerular association, reduces registration error, reduces computational load, and ensures the reliability and consistency of diagnostic results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123072784A_ABST
    Figure CN123072784A_ABST
Patent Text Reader

Abstract

This application proposes a method and apparatus for cross-slice association of multi-stained glomeruli based on tissue block segmentation, comprising: acquiring multiple stained slices of a target case and constructing a tissue descriptor for each tissue block; calculating the cost distance between slices based on the tissue descriptor, and using the slice with the lowest central representation score as the reference slice; constructing registration groups for each tissue block of the reference slice, assigning the moving slice tissue block to the group, and registering the glomerular coordinates to the reference local coordinate system within the group using affine and controlled thin-plate spline gating; calculating the comprehensive cost for detection pairs of different slices within the same registration group, generating multiple hypothetical trajectories based on this cost through bundle search, and selecting the trajectory with the highest total cost and no overlap after precise solution, packaging and outputting it as the sequence of the same glomerulus in different stained slices. This scheme uses tissue blocks as the granularity and registration unit, avoiding cross-tissue mismatches in global registration, decoupling detection and association, and improving the accuracy and stability of multi-stained glomerular association.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of digital healthcare, and in particular to a method and apparatus for cross-slice association of multi-stained glomeruli based on tissue segmentation. Background Technology

[0002] Precise diagnosis and treatment of chronic kidney disease (CKD) heavily relies on renal biopsy pathological evaluation, with the glomerulus being the core focus. Clinically, kidney tissue from the same case is typically prepared into continuous thin sections and stained with various methods such as HE, PAS, Masson, and PASM to reveal pathological features of the glomerulus, including cell proliferation, mesangial matrix, basement membrane changes, and immune complex deposition, from different perspectives. Since different types of lesions are often only clearly visible under different staining conditions, physicians need to repeatedly compare multiple stained sections to comprehensively interpret the structure and lesions of the same glomerulus. This process of "manual comparison across staining sections" is not only time-consuming and labor-intensive but also highly dependent on the reader's experience, making it prone to omissions or mismatches. Therefore, how to automatically and accurately link the same glomerulus in different stained sections and assign it a unified number has become a crucial problem that urgently needs to be solved in digital pathology-assisted diagnosis.

[0003] Existing methods for associating multi-slice tissue structures have two main limitations. First, most methods employ a global registration approach for the entire slice, treating two slices as a whole and performing affine or flexible alignment, then searching for correspondences point-by-point in the registration result. However, a single kidney tissue slice often contains multiple unconnected tissues, whose relative positions on the image may rotate, translate, or even flip during slice preparation, staining, and scanning. Global registration forcibly aligns regions that originally belong to different tissues, leading to "cross-tissue mismatches" between different stainings, i.e., misidentifying a glomerulus from the left tissue as the same glomerulus from the right tissue, thus contaminating subsequent lesion controls. Second, some methods use a pairwise voting or matching approach based on test results. This approach neither considers the structural correspondence between tissue blocks nor actively controls the degree of deformation and registration quality. When there are significant staining differences, local tissue shrinkage, or missed detections in a particular slice, unstable or even erroneous association results are easily produced. In addition, existing methods typically couple glomerular detection with cross-slice association, and the association results are extremely sensitive to the version and parameters of the detection model, making parameter tuning and result reproduction difficult. Summary of the Invention

[0004] This application provides a method and apparatus for cross-slice association of multi-stained glomeruli based on tissue blocks. By using tissue blocks as the granularity and registration unit, it avoids cross-tissue mismatches in global registration, decouples detection and association, and improves the accuracy and stability of multi-stained glomeruli association.

[0005] In a first aspect, embodiments of this application provide a method for cross-slice association of multi-stained glomeruli based on tissue segmentation, the method comprising: Multiple stained sections of the target case are obtained, each tissue block in each stained section is obtained, and a tissue descriptor corresponding to each tissue block is obtained. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The cost distance between any two stained sections is calculated based on tissue descriptors of different tissue blocks. The stained section with the smallest center representative score is used as the reference section, and the stained sections other than the reference section are used as moving sections. The cost distance represents the similarity of tissue blocks between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. For each tissue block within the reference slice, a corresponding registration group is constructed. Based on the similarity between tissue blocks, each tissue block within each moving slice is assigned to a different registration group. Within each registration group, the glomerular coordinates of different moving slice tissue blocks are registered to the local coordinate system of the reference slice tissue block. Multiple detection pairs are obtained within each registration group, and the comprehensive cost between each detection pair is calculated, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. Based on the combined cost between different detection pairs within each registration group, a bundle search is performed to obtain multiple hypothetical trajectories. These hypothetical trajectories are then packaged and output, with each hypothetical trajectory representing the sequence of the same glomerulus in different stained sections.

[0006] Secondly, embodiments of this application provide a multi-slice glomerular cross-section association device based on tissue segmentation, including: The acquisition module is used to acquire multiple stained sections of the target case, acquire each tissue block in each stained section, and the tissue descriptor corresponding to each tissue block. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The reference calibration module calculates the cost distance between every two stained sections based on tissue descriptors of different tissue blocks, and uses the stained section with the smallest center representative score as the reference section, and the stained sections other than the reference section as moving sections. The cost distance represents the tissue block similarity between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. The grouping module constructs a corresponding registration group for each tissue block within the reference slice. Based on the similarity between tissue blocks, it assigns each tissue block within each moving slice to a different registration group, and within each registration group, it registers the glomerular coordinates within different moving slice tissue blocks to the local coordinate system of the reference slice tissue block. The cost calculation module is used to acquire multiple detection pairs within each registration group and calculate the comprehensive cost between each detection pair, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. The registration module performs a bundle search based on the combined cost between different detection pairs within each registration group to obtain multiple hypothetical trajectories, and then packages and outputs these hypothetical trajectories. Each hypothetical trajectory represents the sequence of the same glomerulus in different stained sections.

[0007] Thirdly, embodiments of this application provide an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform a multi-slice association method for glomeruli based on tissue segmentation.

[0008] The main contributions and innovations of this invention are as follows: This application's embodiments calculate the cost distance between slices based on tissue descriptors and use the slice with the smallest center representation score as the reference slice. The cost distance comprehensively characterizes the overall similarity between slices. The slice with the smallest center representation score is the slice that is most similar to other slices as a whole and has the most complete tissue blocks. Using it as a reference can minimize the cumulative registration error and ensure that the reference slice can represent the overall distribution of all slices in the entire case, providing the most reliable benchmark for subsequent registration. This application's embodiments use the Hungarian algorithm to solve tissue block matching. The Hungarian algorithm obtains the globally optimal one-to-one matching between two tissue sets in polynomial time, avoiding greedy matching from getting trapped in local optima, and maximizing the accuracy of tissue block pairing between different slices. This ensures that the cost distance calculation and registration group division are based on the optimal tissue correspondence. This application's embodiments construct registration groups in order of increasing cost distance from the moving slice to the reference slice. Slices with a cost distance not exceeding a threshold are included in the group; those exceeding the threshold are created as new groups. The closer the slice, the better, and the progressively larger groups are registered. This can reduce error accumulation. A maximum cost threshold is set, and exceeding the threshold is allowed. The tissue block self-registration group takes into account differences such as missing, added, deleted, and deformed tissue blocks, ensuring that each tissue block is properly placed and avoiding forced pairing of tissues with excessive differences that may lead to incorrect associations. In this embodiment, elastic transformation is only enabled when the registration is accurate and has a positive effect. Cross-validation is used to check alignment improvement, and displacement guards are used to prevent the glomerular coordinates from being excessively pulled away by the whole group. This not only addresses non-rigid deformations such as local tissue shrinkage, but also avoids errors introduced by mismatch or overfitting, achieving active control over the degree of deformation and registration quality. If the gating fails, the result is rolled back to the affine result, ensuring a reliable fallback in any situation. In this embodiment, the comprehensive cost is calculated only for detection pairs whose post-registration distance does not exceed the candidate radius within the same registration group and between different stained sections. Candidate pairs are screened based on spatial nearest neighbors after registration, significantly narrowing the matching search range, reducing the amount of computation, and preventing cross-tissue mismatches from entering trajectory construction from the source. The comprehensive cost integrates four types of information: spatial, depth features, manual contour features, and detection box area, and is weighted and summed to characterize glomerular similarity from multiple dimensions and improve the accuracy of corresponding judgments.

[0009] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to an embodiment of this application; Figure 2This is a structural block diagram of a multi-stain glomerular cross-slice association device based on tissue segmentation according to an embodiment of this application; Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0011] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0012] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0013] Example 1 This application provides a method for cross-slice association of multi-stained glomeruli based on tissue blocks. This method uses tissue blocks as the granularity and registration unit, avoiding cross-tissue mismatches in global registration, decoupling detection and association, and improving the accuracy and stability of multi-stained glomeruli association. Specifically, refer to... Figure 1 The method includes: Multiple stained sections of the target case are obtained, each tissue block in each stained section is obtained, and a tissue descriptor corresponding to each tissue block is obtained. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The cost distance between any two stained sections is calculated based on tissue descriptors of different tissue blocks. The stained section with the smallest center representative score is used as the reference section, and the stained sections other than the reference section are used as moving sections. The cost distance represents the similarity of tissue blocks between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. For each tissue block within the reference slice, a corresponding registration group is constructed. Based on the similarity between tissue blocks, each tissue block within each moving slice is assigned to a different registration group. Within each registration group, each tissue block of different moving slices is registered to the local coordinate system of the tissue block in the reference slice. Multiple detection pairs are obtained within each registration group, and the comprehensive cost between each detection pair is calculated, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. Based on the combined cost between different detection pairs within each registration group, a bundle search is performed to obtain multiple hypothetical trajectories. These hypothetical trajectories are then packaged and output, with each hypothetical trajectory representing the sequence of the same glomerulus in different stained sections.

[0014] In the current embodiment, any pre-trained target detection model is used to obtain a corresponding single analysis package for each stained slide. The single analysis package includes an analysis list, a mask for each tissue block, and a feature file for each tissue block.

[0015] Specifically, the number of tissue blocks recorded in the analysis list within the stained sections is recorded, and the detection information of each glomerulus contained within the corresponding tissue block is recorded in the feature file of the tissue block.

[0016] Specifically, after obtaining a single analysis package, ensure that the number of tissue blocks, the number of tissue block masks, and the number of feature files recorded in the analysis list are the same. If they are different, it indicates that feature misalignment has occurred. In this case, do not perform the subsequent registration step, and re-obtain the single analysis package of the stained section.

[0017] For example, the detection information of the glomerulus includes the local center coordinates, bounding box, depth feature vector, and hand-crafted contour features of each glomerulus. The local center coordinates are the center of the bounding box of the glomerulus in the local coordinate system of the tissue block. The bounding box is the detection rectangle given by the target detection model. The depth feature vector is the feature representation within the bounding box. The hand-crafted contour features are Hu moment, area, perimeter, aspect ratio of the minimum bounding rectangle, and radius of the minimum bounding circle.

[0018] Furthermore, a tissue descriptor is constructed based on the tissue mask of each tissue block, the tissue block detection frame, and the local center of each glomerulus within the tissue block and the area of ​​the glomerular detection frame.

[0019] Specifically, the tissue mask is binarized to obtain a binary mask. Based on the binary mask, seven Hu invariant moments are constructed. Hu invariant moments are insensitive to translation, rotation, and scale, so they can accurately express the similarity between tissue blocks, rather than comparing the absolute positions between tissue blocks. The local center of each glomerulus and the area of ​​the glomerular detection frame can reflect the distribution of all glomeruli within the tissue block, thereby further characterizing the structural similarity between tissue blocks. Finally, the tissue descriptor integrates the morphological features of the tissue block itself and the distribution features of the glomeruli within the tissue block, providing a complete feature basis for subsequent tissue block matching.

[0020] Specifically, in the step of constructing the 7 Hu invariant moments, the binary mask is accumulated with geometric moments according to pixel coordinates, and then converted into central moments and normalized central moments, thus obtaining the 7 Hu invariant moments.

[0021] For example, the organization descriptor generated by this scheme has 17 dimensions, and the meaning of each dimension is shown in Table 1: Table 1. Meaning of different dimensions of the organization descriptor.

[0022] Specifically, to ensure that the organization descriptors of different organization blocks can be directly compared, the organization descriptors of each organization block are standardized according to the median and interquartile range, and then clipped to [-4,4].

[0023] In the current embodiment, the cost of each pair of tissue blocks between two stained sections is calculated. The matching method with the minimum sum of costs is selected as the tissue block matching result based on the Hungarian algorithm. The sum of costs corresponding to the tissue block matching results is taken as the cost distance between the two stained sections. The cost is calculated based on the Euclidean distance between the tissue descriptors of the two tissue blocks and the number of glomeruli within the two tissue blocks. The formula for calculating the cost is expressed as follows:

[0024] in, For organization blocks With tissue blocks The cost, , Tissue blocks With tissue blocks Organization descriptor, , Tissue blocks With tissue blocks The number of glomeruli within the tissue is used to characterize the similarity between tissue blocks using the cost. According to the cost formula, the smaller the cost between two tissue blocks, the higher the similarity between the two tissue blocks; conversely, the larger the cost between two tissue blocks, the lower the similarity between the two tissue blocks.

[0025] Specifically, the Hungarian algorithm is a classic combinatorial optimization algorithm for solving task allocation problems. It can obtain the globally optimal matching result between two sets in polynomial time, avoiding the defect of greedy matching that is prone to getting trapped in local optima. It maximizes the accuracy of tissue block matching between different slices. In other words, this scheme uses the Hungarian algorithm to obtain the most suitable tissue block pairing between two stained slices.

[0026] After determining the cost distance between each stained section, the center representation score of each section is calculated. The section with the smallest center representation score is selected as the reference section for the entire association process, and the remaining sections are used as moving sections to be registered. This selection method can minimize the cumulative registration error and ensure that the reference section can represent the overall distribution characteristics of all sections in the entire case.

[0027] In other words, the obtained reference slide is a stained slide that is similar to other moving slides and has the most complete tissue block.

[0028] In the current embodiment, the cost matrix for each mobile slice is calculated sequentially from the nearest to the farthest cost distance from the reference slice. The cost matrix represents the cost of each tissue block in the current mobile slice to each tissue block in the reference slice. The Hungarian algorithm is used to calculate the one-to-one assignment between the tissue blocks in the mobile slice and the tissue blocks in the reference slice based on the cost matrix. If the cost of the mobile slice tissue block and the corresponding reference slice tissue block in the assignment result is not greater than a preset maximum cost, the mobile slice tissue block is assigned to the registration group of the corresponding reference slice tissue block. If the cost of the mobile slice tissue block and the corresponding reference slice tissue block in the assignment result is greater than the preset maximum cost, a new registration group is constructed for the mobile slice tissue block. That is, after constructing a new registration group for the mobile slice tissue block, the corresponding mobile slice tissue block is regarded as a reference slice tissue block, and subsequent registration operations are performed based on this.

[0029] Specifically, after completing the allocation of different registration groups, each block is assigned a corresponding registration group number, reference slice number, and list of blocks not included in the group to facilitate subsequent calculations.

[0030] For example, the reference slice HE has two groups: a left block and a right block. The cost of the left block of the PAS slice and the left block of HE is 0.4, and the cost of the right block of HE is 1.8. Hungary assigns the left block of PAS to group 0 (0.4 ≤ 1.5) and the right block of PAS to group 1. If the cost of a block is greater than 1.5 for both groups, it cannot be assigned to an existing group.

[0031] In the current embodiment, the local coordinate system of the reference slice tissue block is obtained as the reference coordinate system within each registration group, the local coordinate system of each moving slice tissue block is obtained as the moving coordinate system, the coordinates of the center of each glomerulus in the reference slice tissue block in the reference coordinate system are obtained as the reference point set, and the coordinates of the center of each glomerulus in each moving slice tissue block in the moving coordinate system are obtained as the moving point set corresponding to each moving slice tissue block. An affine transformation matrix is ​​constructed based on the affine transformation relationship from each moving point set to the reference point set. The affine transformation matrix is ​​used to register the glomerular coordinates in the corresponding moving slice tissue block to the local coordinate system of the reference slice tissue block.

[0032] Specifically, a coarse affine from the moving tissue block to the reference tissue block is obtained using the translation method, the principal axis method, or the soft correspondence method. The translation method uses the difference between the median of the coordinates of the moving point set and the median of the coordinates of the reference point set as the translation amount to construct the coarse affine. The principal axis method performs principal component analysis on the centered moving point set and the reference point set respectively, and then uses the four diagonal signs of the principal axes of the reference coordinate system and the moving coordinate system to generate a scaled linear part, and constructs the coarse affine based on the scaled linear part. The soft correspondence method uses the moving point set as the centroid of the Gaussian mixture and performs expectation maximization on the reference point set to construct the coarse affine.

[0033] Specifically, the symmetric residuals between the coordinate points after coarse affine transformation are calculated, and the coarse affine result with the smallest symmetric residuals is applied.

[0034] For example, the reference point set includes three coordinate points (100, 100), (200, 100), and (150, 200), and the moving point set includes three coordinate points (70, 140), (165, 145), and (118, 238). The median of the coordinates of the reference point set is (150, 100), and the median of the coordinates of the moving point set is (118, 145). Therefore, the translation is (32, 45) After transformation, the three points of the moving point set return to approximately the same area as the three points of the reference point set, with a symmetrical residual of about ten pixels; if the principal axis or soft-corresponding affine residual is smaller, the latter should be used instead.

[0035] Specifically, after obtaining the coarse affine, the coarse affine is corrected by the residual between the moving point and the reference point after applying the coarse affine transformation, and the corrected result is used as the final affine transformation matrix.

[0036] In the process of correcting the coarse affine, the coarse affine is first corrected by the point density map refinement method to obtain the first corrected affine. The point density map refinement method is to draw the reference point and the moving point after the coarse affine transformation on the same screen, and to perform affine correction using enhanced correlation metric. In addition, when the number of points is less than 5, the solution fails, or the determinant goes out of bounds, the original coarse affine is retained as the first corrected affine.

[0037] After applying the first corrected affine to the moving point set, multiple pairs of mutual anchor points are obtained. When there are no fewer than 4 pairs of mutual anchor points, a new affine is estimated using random sampling consensus. When the symmetric residual of the new affine is less than the old value, the new affine replaces the first corrected affine to obtain the second corrected affine. The second corrected affine is used as the affine transformation matrix. If the nearest reference point of moving point i is j, and the nearest moving point of j is exactly i, and the distance does not exceed the candidate radius R, then they are recorded as a pair of mutual anchor points.

[0038] In the current embodiment, the registered tissue blocks within the registration group are used as tissue pairs. Thin-plate spline gating is performed based on each pair of mutual anchor points in the tissue pair. After the thin-plate spline gating passes, the glomerular coordinates within the moving slice tissue block in the tissue pair are transformed using thin-plate spline. The result of the thin-plate spline transformation is used as the new registration result. Specifically, two glomerular coordinate points belonging to the reference slice tissue block and the moving slice tissue block, respectively, whose distance does not exceed the candidate radius, and which are the nearest neighbors to each other, are used as mutual anchor point pairs. The thin-plate spline gating must simultaneously satisfy the following conditions: the number of mutual anchor point pairs is greater than a preset lower limit; verification glomerular coordinate points are obtained from the mutual anchor point pairs, and the cross-validation of the verification glomerular coordinate points is effective; the median residual of the verification glomerular coordinate points after the thin-plate spline transformation is reduced to less than 0.9 times the original value; and the 95th percentile of the displacement of all glomerular coordinate points in the moving slice tissue block does not exceed the upper limit.

[0039] In other words, if the thin plate spline gate passes, it means that implementing the thin plate spline transformation will have a positive effect.

[0040] In some embodiments, the specific process of cross-validation in the thin-plate spline gating is as follows: From the mutual anchor point pairs in the tissue pair, every four pairs are sequentially retained, with one glomerular coordinate point belonging to the moving slice tissue block in each pair serving as the validation glomerular coordinate point. The remaining mutual anchor point pairs with glomerular coordinate points belonging to the moving slice tissue block serve as training glomerular coordinate points. The training glomerular coordinate points are used to fit the thin-plate spline, while the validation glomerular coordinate points do not participate in the fitting and are only used to verify the alignment effect of the thin-plate spline. When the number of training glomerular coordinate points is not less than a preset lower limit and the validation glomerular coordinate points are not empty, the cross-validation of the validation glomerular coordinate points is deemed valid.

[0041] Specifically, when cross-validation is effective, a thin-plate spline is fitted to the coordinate points of the training glomeruli, and the fitted thin-plate spline is applied to the coordinate points of the validation glomeruli. The median of the residuals after deformation of the validation glomeruli coordinate points is calculated and compared with the median of the residuals before deformation. When the median of the residuals after deformation of the validation glomeruli coordinate points drops to less than 0.9 times the median of the residuals before deformation, it is determined that the thin-plate spline can improve the alignment effect.

[0042] Furthermore, the specific process of displacement guarding in the thin-plate spline gating is as follows: after the coordinate points of all glomeruli of the moving tissue slice are transformed by the thin-plate spline, the displacement of each glomerular coordinate point relative to the original displacement is calculated. The displacements of all glomerular coordinate points are sorted from smallest to largest, and the displacement corresponding to the 95th percentile in the sort is taken as the 95th percentile displacement. When the 95th percentile displacement does not exceed the preset upper limit, it is determined that the thin-plate spline transformation will not excessively pull the glomerular coordinate points away.

[0043] In other words, for thin plate spline gating to pass, the following conditions must be met simultaneously: the number of mutual anchor point pairs is greater than the preset lower limit, cross-validation is valid, the median of the residuals after deformation of the glomerular coordinate points is reduced to less than 0.9 times the median of the residuals before deformation, and the 95% quantile displacement does not exceed the preset upper limit. When all of the above conditions are met, the thin plate spline gating is judged to pass.

[0044] After the thin-plate spline gate passes, the thin-plate spline is fitted and smoothed using all the mutual anchor points in the tissue pair. All glomerular coordinate points in the moving tissue block of the tissue pair are then registered to the local coordinate system of the reference tissue block using the thin-plate spline transformation. The coordinates after the thin-plate spline transformation are used as the new registration result, overriding the registration result obtained by the aforementioned affine transformation matrix. At the same time, the reason for enabling the thin-plate spline transformation is recorded as cross-validation improvement.

[0045] Specifically, when fitting a smooth thin-plate spline, a preset smoothing coefficient is used to avoid the fitted smooth thin-plate spline from overfitting the mismatched glomerular coordinate points in the anchor point pair.

[0046] For example, in cases of localized shrinkage, the median residual of the validation glomerular coordinate points in a certain registration group before deformation was 12 pixels. After smoothing thin-plate spline transformation, the median residual after deformation decreased to 8 pixels, which is less than 0.9 times the median residual before deformation. Moreover, the 95% quantile displacement did not exceed the preset upper limit, and the thin-plate spline gating was passed. After implementing the thin-plate spline transformation, more real detection pairs entered the candidate radius, and the registration effect was improved.

[0047] When the thin-plate spline gating fails, the thin-plate spline transformation is not performed. The registration result obtained by the aforementioned affine transformation matrix is ​​retained as the new registration result, and the reason for the failure of the thin-plate spline gating is recorded. The reasons include not allowing the use of thin-plate splines, the number of mutual anchor point pairs being less than the preset lower limit, invalid cross-validation, the median of the residuals after the deformation of the glomerular coordinate points not decreasing to less than 0.9 times the median of the residuals before deformation, and the 95% quantile displacement exceeding the preset upper limit.

[0048] For example, after the thin-plate spline transformation, the 95th percentile displacement of all glomerular coordinate points of the moving slice tissue block in a certain registration group is 210 pixels, which exceeds the preset upper limit. This triggers the displacement guard, and the thin-plate spline gate fails. The registration result obtained by the aforementioned affine transformation matrix is ​​retained as the new registration result to avoid pulling the entire group of glomerular coordinate points apart, thereby avoiding incorrect registration.

[0049] In other words, regardless of whether the thin plate spline gating passes, the final registration result adopted by the tissue pair is the unified coordinate of the tissue pair in the reference coordinate system. The unified coordinate may include the registration result of the thin plate spline transformation, or it may only be the registration result obtained by the affine transformation matrix.

[0050] In the current embodiment, after obtaining the final registration results of each tissue pair, the comprehensive cost is calculated only for detection pairs whose registered distance does not exceed the candidate radius, within the same registration group and between different stained sections. The detection pair is the corresponding glomerulus in any two different tissue blocks.

[0051] Specifically, the comprehensive cost includes spatial cost, depth feature cost, manual contour feature cost, and detection box area cost. The comprehensive cost is obtained by multiplying the spatial cost, depth feature cost, manual contour feature cost, and detection box area cost by their respective weights and then summing them. The spatial cost is determined based on the Euclidean distance between the two glomeruli in the unified coordinates after registration. The depth feature cost is determined based on the similarity between the depth feature vectors of the two glomeruli. The manual contour feature cost is determined based on the similarity between the manual contour features of the two glomeruli. The detection box area cost is determined based on the difference between the detection box areas of the two glomeruli.

[0052] Furthermore, when the overall cost of a detection pair exceeds the preset maximum overall cost, or when the absence of a key component of any glomerulus in the detection pair results in an overall cost that is not finite, the detection pair will not be used to generate the hypothetical trajectory; detection pairs used to generate hypothetical trajectories will not be constructed across registration groups.

[0053] In the current embodiment, the bundle search uses one end of the glomerulus in the detection pair as a seed. Starting from the bundle set containing only the seed, it searches for neighboring glomeruli located in different stained sections and not occupying the glomerular observation of the current stained section in the adjacent detection pairs. A preset number of neighboring glomeruli are added to the set in ascending order of the minimum comprehensive cost of the neighboring glomeruli to the existing members in the bundle set. The bundle set is expanded round by round. When the number of glomerular observations in the bundle set reaches the minimum section coverage, the bundle set is recorded as the hypothetical trajectory. Here, the adjacent detection pair is a detection pair within the same registration group, belonging to different stained sections and having an adjacent section relationship.

[0054] Specifically, the minimum number of glomerular observations in the bundle set means that the number of different stained sections covered by the glomeruli in the bundle set reaches a preset minimum requirement, that is, to ensure that the hypothetical trajectory can cover a sufficient number of sections and avoid invalid short trajectories that only connect a few sections.

[0055] In the current embodiment, in the step of packaging and outputting multiple hypothetical trajectories, the hypothetical trajectory with the highest total comprehensive cost for accurate solution among the multiple hypothetical trajectories and which do not overlap is taken as the reserved trajectory. The reserved trajectory is output, and the constraint that there is only one glomerulus in the same stained slice for the same hypothetical trajectory is used as the constraint for accurate solution.

[0056] Specifically, when the exact solution times out, it degenerates into a score-based greedy solution.

[0057] Specifically, the number of different stained slices that appeared for each hypothetical trajectory is counted, and hypothetical trajectories whose slice coverage is below a preset threshold are deleted.

[0058] Specifically, in the final slide viewing interface, each stained slide is compared with a unified number. The same unified number falls in the same anatomical position on the comparison map of multiple stained slides, thus completing the cross-slide association of multi-stained glomeruli.

[0059] Example 2 Based on the same concept, referencing Figure 2 This application also proposes a multi-slice glomerular cross-slice association device based on tissue segmentation, comprising: The acquisition module is used to acquire multiple stained sections of the target case, acquire each tissue block in each stained section, and the tissue descriptor corresponding to each tissue block. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The reference calibration module calculates the cost distance between every two stained sections based on tissue descriptors of different tissue blocks, and uses the stained section with the smallest center representative score as the reference section, and the stained sections other than the reference section as moving sections. The cost distance represents the tissue block similarity between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. The grouping module constructs a corresponding registration group for each tissue block within the reference slice. Based on the similarity between tissue blocks, it assigns each tissue block within each moving slice to a different registration group, and within each registration group, it registers the glomerular coordinates within different moving slice tissue blocks to the local coordinate system of the reference slice tissue block. The cost calculation module is used to acquire multiple detection pairs within each registration group and calculate the comprehensive cost between each detection pair, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. The registration module performs a bundle search based on the combined cost between different detection pairs within each registration group to obtain multiple hypothetical trajectories, and then packages and outputs these hypothetical trajectories. Each hypothetical trajectory represents the sequence of the same glomerulus in different stained sections.

[0060] Example 3 This embodiment also provides an electronic device, see reference. Figure 3 It includes a memory 402 and a processor 401, wherein the memory 402 stores a computer program and the processor 401 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0061] Specifically, the processor 401 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0062] The memory 402 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 402 may include removable or non-removable (or fixed) media. Where appropriate, the memory 402 may be internal or external to a data processing device. In a particular embodiment, the memory 402 is non-volatile memory. In a particular embodiment, the memory 402 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0063] The memory 402 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 401.

[0064] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any of the multi-slice association methods for multi-stained glomeruli based on tissue segmentation in the above embodiments.

[0065] Optionally, the electronic device may further include a transmission device 403 and an input / output device 404, wherein the transmission device 403 is connected to the processor 401 and the input / output device 404 is connected to the processor 401.

[0066] The transmission device 403 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 403 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0067] The input / output device 404 is used to input or output information. In this embodiment, the input information may be multiple stained sections, the detection results of each stained section, etc., and the output information may be the sequence of the same glomerulus in different stained sections, etc.

[0068] Optionally, in this embodiment, the processor 401 can be configured to perform the following steps via a computer program: Multiple stained sections of the target case are obtained, each tissue block in each stained section is obtained, and a tissue descriptor corresponding to each tissue block is obtained. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The cost distance between any two stained sections is calculated based on tissue descriptors of different tissue blocks. The stained section with the smallest center representative score is used as the reference section, and the stained sections other than the reference section are used as moving sections. The cost distance represents the similarity of tissue blocks between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. For each tissue block within the reference slice, a corresponding registration group is constructed. Based on the similarity between tissue blocks, each tissue block within each moving slice is assigned to a different registration group. Within each registration group, the glomerular coordinates of different moving slice tissue blocks are registered to the local coordinate system of the reference slice tissue block. Multiple detection pairs are obtained within each registration group, and the comprehensive cost between each detection pair is calculated, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. Based on the combined cost between different detection pairs within each registration group, a bundle search is performed to obtain multiple hypothetical trajectories. These hypothetical trajectories are then packaged and output, with each hypothetical trajectory representing the sequence of the same glomerulus in different stained sections.

[0069] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0070] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0071] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 3Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0072] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0073] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for cross-slice association of multi-stained glomeruli based on tissue segmentation, characterized in that, Includes the following steps: Multiple stained sections of the target case are obtained, each tissue block in each stained section is obtained, and a tissue descriptor corresponding to each tissue block is obtained. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The cost distance between any two stained sections is calculated based on tissue descriptors of different tissue blocks. The stained section with the smallest center representative score is used as the reference section, and the stained sections other than the reference section are used as moving sections. The cost distance represents the similarity of tissue blocks between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. For each tissue block within the reference slice, a corresponding registration group is constructed. Based on the similarity between tissue blocks, each tissue block within each moving slice is assigned to a different registration group. Within each registration group, the glomerular coordinates of different moving slice tissue blocks are registered to the local coordinate system of the reference slice tissue block. Multiple detection pairs are obtained within each registration group, and the comprehensive cost between each detection pair is calculated, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. Based on the combined cost between different detection pairs within each registration group, a bundle search is performed to obtain multiple hypothetical trajectories. These hypothetical trajectories are then packaged and output, with each hypothetical trajectory representing the sequence of the same glomerulus in different stained sections.

2. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, A tissue descriptor is constructed based on the tissue mask, tissue block detection frame, and the local center of each glomerulus within the tissue block and the area of ​​the glomerular detection frame.

3. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, Calculate the cost of each pair of tissue blocks between two stained sections. Based on the Hungarian algorithm, select the matching method with the minimum sum of costs as the tissue block matching result. The sum of costs corresponding to the tissue block matching results is taken as the cost distance between the two stained sections.

4. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 3, characterized in that, The cost is used to characterize the similarity between tissue blocks, and the cost is calculated based on the Euclidean distance between the tissue descriptors of the two tissue blocks and the number of glomeruli in the two tissue blocks.

5. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, The cost matrix for each mobile slice is calculated sequentially from the nearest to the farthest cost distance from the reference slice. The cost matrix represents the cost of each tissue block in the current mobile slice to each tissue block in the reference slice. The Hungarian algorithm is used to calculate the one-to-one assignment between the tissue blocks in the mobile slice and the tissue blocks in the reference slice based on the cost matrix to obtain the assignment result. If the cost of the mobile slice tissue block to the corresponding reference slice tissue block in the assignment result is not greater than the preset maximum cost, then the mobile slice tissue block is assigned to the registration group of the corresponding reference slice tissue block. If the cost of moving the tissue block and the corresponding reference tissue block in the assignment result is greater than the preset maximum cost, then a new registration group is constructed for the moving tissue block.

6. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, Within each registration group, the local coordinate system of the reference slice tissue block is obtained as the reference coordinate system, and the local coordinate system of each moving slice tissue block is obtained as the moving coordinate system. The coordinates of the center of each glomerulus in the reference slice tissue block in the reference coordinate system are obtained as the reference point set, and the coordinates of the center of each glomerulus in each moving slice tissue block in the moving coordinate system are obtained as the moving point set corresponding to each moving slice tissue block. An affine transformation matrix is ​​constructed based on the affine transformation relationship from each moving point set to the reference point set. The affine transformation matrix is ​​used to register the glomerular coordinates in the corresponding moving slice tissue block to the local coordinate system of the reference slice tissue block.

7. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, The tissue blocks that have been registered within the registration group are used as tissue pairs. Thin-plate spline gating is performed based on each pair of mutual anchor points in the tissue pair. After the thin-plate spline gating passes, the glomerular coordinates in the moving slice tissue block in the tissue pair are transformed using thin-plate spline. The result of the thin-plate spline transformation is used as the new registration result. Among them, the two glomerular coordinate points in the tissue pair that belong to the reference slice tissue block and the moving slice tissue block respectively, are within a distance of no more than the candidate radius, and are the nearest neighbors to each other are used as mutual anchor point pairs.

8. The method for cross-slice association of multi-stained glomeruli based on tissue segmentation according to claim 1, characterized in that, The bundle search uses one end of the glomerulus in the detection pair as a seed. Starting from the bundle set containing only this seed, it searches for neighboring glomeruli in adjacent detection pairs that are located on different stained sections and do not occupy the glomerular observation of the current stained section. A preset number of neighboring glomeruli are added to the set in ascending order of the minimum comprehensive cost of the neighboring glomeruli to the existing members in the bundle set. The bundle set is expanded round by round. When the number of glomerular observations in the bundle set reaches the minimum section coverage, the bundle set is recorded as the hypothetical trajectory. Here, the adjacent detection pair is a detection pair within the same registration group that belongs to different stained sections and has an adjacent section relationship.

9. A device for cross-slice association of multi-stained glomeruli based on tissue segmentation, characterized in that, include: The acquisition module is used to acquire multiple stained sections of the target case, acquire each tissue block in each stained section, and the tissue descriptor corresponding to each tissue block. The tissue descriptor contains the detection information of the corresponding tissue block and the detection information of each glomerulus in the tissue block. The reference calibration module calculates the cost distance between every two stained sections based on tissue descriptors of different tissue blocks, and uses the stained section with the smallest center representative score as the reference section, and the stained sections other than the reference section as moving sections. The cost distance represents the tissue block similarity between different stained sections, and the center representative score is the sum of the cost distances between the current stained section and every other stained section. The grouping module constructs a corresponding registration group for each tissue block within the reference slice. Based on the similarity between tissue blocks, it assigns each tissue block within each moving slice to a different registration group, and within each registration group, it registers the glomerular coordinates within different moving slice tissue blocks to the local coordinate system of the reference slice tissue block. The cost calculation module is used to acquire multiple detection pairs within each registration group and calculate the comprehensive cost between each detection pair, wherein the detection pair is the corresponding glomerulus in any two different tissue blocks, and the comprehensive cost represents the similarity between the corresponding glomerulus. The registration module performs a bundle search based on the combined cost between different detection pairs within each registration group to obtain multiple hypothetical trajectories, and then packages and outputs these hypothetical trajectories. Each hypothetical trajectory represents the sequence of the same glomerulus in different stained sections.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform a method for cross-slice association of multi-stained glomeruli based on tissue segmentation as described in any one of claims 1-8.