Basilar membrane sampling method based on glomerular capillary branch characteristics

Abstract

The invention provides a basement membrane sampling method based on glomerular capillary branch characteristics, and relates to the technical field of image recognition, and the method comprises the steps: employing the center and radius of a glomerular region, carrying out the analysis of the mesangial crossing condition of a basement membrane, obtaining a mesangial crossing result, carrying out the region division and sampling window size calculation, and carrying out the calculation of the size of a sampling window. Obtaining a plurality of sectors and corresponding window areas; random sampling is carried out in the sectors to obtain candidate center points; analyzing the candidate center point based on the window area, if the sampling is not qualified, performing failure retry and annealing strategy to adjust the window, and resampling; otherwise, obtaining a qualified window; and combining all qualified windows to obtain a basement membrane sampling result, and completing the basement membrane sampling based on the glomerular capillary branch characteristics. According to the method, the problem of how to perform balanced, low-deviation, reproducible and efficient sampling and labeling on the glomerular basement membrane under the anatomical priori constraint in the light microscope glomerular pathological image is solved.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of image recognition, and in particular to a glomerular basement membrane sampling method based on glomerular capillary branch characteristics. BACKGROUND

[0002] Early diagnosis and intervention of diabetic nephropathy are of great significance for preventing the disease from progressing to end-stage renal disease. Studies have shown that the early characteristic pathological changes of diabetic nephropathy mainly manifest as diffuse thickening of the glomerular basement membrane, which is a key morphological marker of disease development. However, the existing AI random sampling technology has obvious limitations in diabetic nephropathy basement membrane labeling. Traditional random sampling methods are mainly designed for macroscopic pathological structures, and it is difficult to effectively capture the characteristics of such subtle lesions as the basement membrane. The sampling process lacks pertinence and is prone to miss key lesion areas. At the same time, these methods fail to combine the unique vascular branch anatomical features of the kidney, especially the 4-5 cluster capillary network formed by the glomerular arteriole branches, resulting in sampling results that cannot accurately reflect the lesion distribution pattern and waste of computing resources.

[0003] The existing technology also has the following problems: the existing random labeling lacks awareness of blood vessel cluster zoning at the glomerular basement membrane (GBM) level, resulting in sampling concentrated in local areas and missing other clusters, and the training data do not match the real distribution. Artificial "picking typical areas" can easily introduce subjective bias, especially in the early DN scenario of subtle thickening of the GBM, the statistical features learned by the model will deviate. Without explicitly using the anatomical prior that the glomerular arteriole forms 4-5 capillary clusters within the glomerulus, it is impossible to achieve balanced coverage in different zones. Lack of reproducible random process and adaptive window size affects both efficiency and consistency of subsequent model comparison experiments. SUMMARY

[0004] In view of the above deficiencies in the prior art, the glomerular basement membrane sampling method based on glomerular capillary branch characteristics provided by the present application solves the problem of how to sample and label the glomerular basement membrane in a balanced, low-bias, reproducible and efficient manner under the constraint of anatomical prior in the light microscopic glomerular pathological image.

[0005] In order to achieve the above-mentioned purpose of the application, the technical scheme adopted by the present application is as follows: a glomerular basement membrane sampling method based on glomerular capillary branch characteristics, comprising: S1: analyzing the pathological image of the glomerular capillary branch characteristics to obtain the glomerular region center and radius; S2: analyzing the mesangial intersection of the basement membrane using the glomerular region center and radius to obtain the mesangial intersection result; S3: performing region division and sampling window size calculation according to the mesangial intersection result to obtain a plurality of sectors and corresponding window areas; S4: Random sampling is performed in the sector to obtain a candidate center point; S5: Based on the window area, the candidate center point is analyzed by using the containing condition and the avoidance condition, if the sampling is unqualified, the failure retry and the annealing strategy are adjusted, the window is returned to S4 for resampling; otherwise, a qualified window is obtained; S6: All qualified windows are merged to obtain the basement membrane sampling result, and the basement membrane sampling based on the glomerular capillary branch feature is completed.

[0006] The beneficial effects of the present application are: the present application provides a basement membrane sampling method based on the glomerular capillary branch feature, which combines the anatomical features of the glomerular capillary loop to design an intelligent sampling strategy, effectively solving the incomplete coverage problem of traditional random sampling in the analysis of diabetic nephropathy basement membrane. (1) It can ensure that the basement membrane of each cluster of capillary network in the glomerulus is evenly sampled, thereby providing more comprehensive and representative pathological information. (2) Uniform random sampling is performed in each cluster, and the area adaptive window and the geometric constraint of avoiding the mesangial area are used to generate a representative, low-bias and reproducible GBM training label set; without relying on an electron microscope, it supports quantitative analysis and model training at the light microscope level in the early stage of DN. (3) Through partitioning, balanced sampling and identifying the spatial distribution of each cluster of capillary network, random sampling is then independently performed in each cluster of capillary network, ensuring that the basement membrane of all regions can be evenly covered.

[0007] Further, the S1 comprises: Identifying the glomerular region and the mesangial area inside it from the pathological image of the glomerular capillary branch feature; Based on the glomerular region and the mesangial area inside it, the glomerular region center and the radius are obtained by calculation.

[0008] Further, the expression of the glomerular region radius is: ; Wherein, The maximum radius of the glomerular region is represented by R, The maximum value is represented by max, The coordinates of a point in G are represented by G, The glomerular region is represented by G, The glomerular region center is represented by C, The Euclidean distance between two points is represented by dist.

[0009] Further, the S2 comprises: When the glomerular region and the mesangial area inside it intersect, the mesangial centroid and the direction vector are recorded, and the multiple asymmetric sectors are divided with the direction vector as the reference direction; wherein, the expression of the direction vector is: ; wherein, denotes a direction vector, denotes a glomerular region center, denotes a mesangium centroid; When there is no mesangium intersection, the glomerular region is divided into multiple sectors by equal angles.

[0010] Further, the S3 comprises: According to the mesangium intersection result, the region is divided to obtain multiple sectors; The sampling window size calculation is performed on each sector to obtain the window area.

[0011] Not only the objective advantage of random sampling is retained, but also the sampling bias is avoided through the anatomy-guided partitioning strategy. This sampling method can more comprehensively reflect the overall pathological changes of the glomerular basement membrane, and provide more reliable pathological basis for early diagnosis of diabetic nephropathy.

[0012] Further, the expression of the window area is: ; ; ; wherein, denotes a window area, denotes a glomerular area, denotes a coverage ratio, denotes a sector number, denotes an area calculation function, denotes a glomerular region, denotes a square window side length.

[0013] Further, the expression of the containing condition is: ; wherein, denotes a sampling window, denotes a square sampling region with as the center and as the side length, denotes a square window center, denotes a square window side length, denotes a glomerular region.

[0014] Further, the expression of the avoidance condition is: ; wherein, denotes a sampling window, Indicates the center of the square window. This indicates the side length of the square window. This refers to the mesangial area inside the glomerulus. This represents the empty set.

[0015] Furthermore, the expression for adjusting the window is: ; in, Indicates the side length of a square window, "Indicates an update operation, This indicates the scaling factor.

[0016] First, suspicious areas are located using low-resolution scanning, and then fine sampling is performed at high resolution. This avoids wasting computational resources on global sampling and improves the sensitivity of lesion detection. At the same time, the dynamic sampling window adjustment mechanism based on blood vessel orientation can adapt to basement membrane structures with different branch angles, avoiding the omission of important features due to fixed sampling modes. Attached Figure Description

[0017] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is an exemplary flowchart of a basement membrane sampling method based on glomerular capillary branching characteristics, as shown in some embodiments of this specification. Detailed Implementation

[0018] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0019] Example Figure 1 This is an exemplary flowchart illustrating a basement membrane sampling method based on glomerular capillary branching characteristics, according to some embodiments of this specification. Figure 1 As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0020] S1: Analyze the pathological images of glomerular capillary branching characteristics to obtain the center and radius of the glomerular region.

[0021] In some embodiments, the processor can identify the glomerular region and the mesangial region inside the glomerular region from the pathological image of the glomerular capillary branch feature; based on the glomerular region and the mesangial region inside the glomerular region, the glomerular region center and the radius are obtained through calculation.

[0022] The glomerular region G is the closed boundary contour of a glomerulus in the pathological section.

[0023] The mesangial region M inside is the mesangial tissue region (which can be empty) inside the glomerulus, which needs to be avoided during sampling.

[0024] In some embodiments, the expression of the glomerular region radius can be: ; Wherein, represents the maximum radius of the glomerular region, represents taking the maximum value, represents the coordinates of a point in G, represents the glomerular region, represents the glomerular region center, represents the Euclidean distance between two points.

[0025] S2: Using the glomerular region center and the radius, the mesangial intersection of the basement membrane is analyzed to obtain the mesangial intersection result.

[0026] The basement membrane is the basement membrane containing the capillary cluster in the glomerular region.

[0027] The capillary cluster is K capillary clusters formed after the afferent arteriole enters the glomerulus, and K can take a common value of 5.

[0028] The mesangial intersection result is the result of the intersection of the mesangial region and the boundary of the glomerular region. For example, if and intersect with , record the mesangial centroid and the direction vector.

[0029] It is not limited to equal-angle sector division, and can include various implementation manners such as anatomical guide partitioning and blood vessel skeleton clustering partitioning.

[0030] In some embodiments, when the glomerular region and the mesangial region inside the glomerular region intersect, the mesangial centroid and the direction vector are recorded, and a plurality of asymmetric sectors are divided with the direction vector as the reference direction; when there is no mesangial intersection, the glomerular region is divided into a plurality of sectors according to equal angles.

[0031] In some embodiments, the expression of the direction vector can be: ; Wherein, Represents the direction vector. Indicates the center of the glomerular region. It indicates the mesangial centroid.

[0032] S3: Based on the mesangial cross results, perform region division and calculate the sampling window size to obtain multiple sectors and corresponding window areas.

[0033] A sector is a partition, and each sector contains a sampling window.

[0034] The sampling window is a square region centered at x with a side length of a.

[0035] In some embodiments, the processor can divide the region into multiple sectors based on the mesentery cross-section results; and calculate the sampling window size for each sector to obtain the window area.

[0036] In some embodiments, the expression for the window area can be: ; ; ; in, Indicates the window area. Indicates the glomerular area, Indicates the coverage ratio. Indicates the number of sectors. This represents the area calculation function. Indicates the glomerular region. This indicates the side length of the square window.

[0037] Coverage ratio is the proportion of the total area of ​​all sampling windows to the glomerular area. For example, r=0.05 means 5% coverage area.

[0038] The formula can be extended to adaptive calculation of the size of rectangular windows or other regular shapes, ensuring consistent sampling coverage of glomeruli of different sizes, which facilitates the consistency of batch training data.

[0039] S4: Random sampling is performed within the sector to obtain candidate center points.

[0040] The candidate center point is in the sector The center point is generated uniformly and randomly within the interior.

[0041] Implementation methods include rejection sampling and direct sampling within the feasible region. Sampling center points are uniformly and randomly generated within the feasible region of each partition, ensuring equal probability selection; this eliminates reliance on manual selection. This guarantees randomness and representativeness, reducing human and systematic biases.

[0042] S5: Based on the window area, analyze the candidate center point using the inclusion condition and the avoidance condition. If the sampling is not qualified, perform a failure retry and annealing strategy to adjust the window, return to S4 to resample; otherwise, get a qualified window.

[0043] The inclusion condition is a condition for judging whether the sampling window is completely within the glomerular region.

[0044] In some embodiments, the expression of the inclusion condition can be: ; wherein, represents the sampling window, represents a square sampling region with as the center and as the side length, represents the center of the square window, represents the side length of the square window, represents the glomerular region.

[0045] The avoidance condition is a condition for judging whether the sampling window avoids the mesangial dense area.

[0046] In some embodiments, the expression of the avoidance condition can be: ; wherein, represents the sampling window, represents the center of the square window, represents the side length of the square window, represents the "mesangial area" inside the glomerulus, represents an empty set.

[0047] In some embodiments, the expression of adjusting the window is: ; wherein, represents the side length of the square window, represents the update operation, represents the shrinkage ratio factor.

[0048] The qualified window is a sampling window that meets the inclusion condition and the avoidance condition.

[0049] Ensure that the sampling region is entirely the target structure (basement membrane) and reduce non-target tissue interference.

[0050] If the sampling is not qualified, repeat the random attempt, and reduce the window (ratio ) after exceeding the upper limit. If necessary, fine-tune the partition boundary or switch to an alternative partition scheme. Improve the sampling success rate in glomeruli with complex shapes and a large mesangial proportion.

[0051] S6: Merge all qualified windows to get the result of basement membrane sampling, and complete the basement membrane sampling based on the characteristics of glomerular capillary branches.

[0052] The result of basement membrane sampling is the set of all windows merged into polygons .

[0053] Combining medical prior knowledge with machine learning makes the sampling process not only rely on data-driven, but also fully utilize the anatomical rules of kidney microvessels, so that the sampling result is more biologically meaningful. This method is not only suitable for basement membrane analysis of diabetic nephropathy, but also provides a new technical idea for the study of other micro-pathological structures.

[0054] In some embodiments of the present specification, a basement membrane sampling method based on the characteristics of glomerular capillary branches is provided. By combining the anatomical characteristics of glomerular capillary loops, an intelligent sampling strategy is designed to effectively solve the incomplete coverage problem of traditional random sampling in the analysis of basement membrane of diabetic nephropathy. (1) It can ensure that the basement membrane of each cluster of capillary network in the glomerulus is evenly sampled, thereby providing more comprehensive and representative pathological information. (2) Uniform random sampling is performed within each cluster. Using area-adaptive windows and geometric constraints to avoid mesangial areas, a representative, low-bias, and reproducible GBM training label set is generated. Without relying on electron microscopy, it supports quantitative analysis and model training at the light microscope level for early DN. (3) By partitioning, evenly sampling, and identifying the spatial distribution of each cluster of capillary network, random sampling is then independently performed within each cluster of capillary network to ensure that the basement membrane of all areas can be evenly covered.

Claims

1. A basement membrane sampling method based on glomerular capillary branching characteristics, characterized in that, include: S1: Analyze the pathological images of glomerular capillary branching characteristics to obtain the center and radius of the glomerular region; S2: Using the center and radius of the glomerular region, the mesangial crossing of the basement membrane is analyzed to obtain the mesangial crossing results; S3: Based on the mesangial cross results, perform region division and calculate the sampling window size to obtain multiple sectors and corresponding window areas; S4: Random sampling is performed within the sector to obtain candidate center points; S5: Based on the window area, analyze the candidate center point using the inclusion condition and the avoidance condition. If the sampling is not qualified, perform a failure retry and annealing strategy to adjust the window, and return to S4 to resample. Otherwise, a qualified window will be obtained; S6: Merge all qualified windows to obtain the basement membrane sampling results, and complete the basement membrane sampling based on the characteristics of glomerular capillary branches.

2. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, S1 includes: Identify the glomerular region and its internal mesangial area from pathological images of glomerular capillary branching characteristics; Based on the glomerular region and its internal mesangial area, the center and radius of the glomerular region are calculated.

3. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, The expression for the radius of the glomerular region is: ； in, Indicates the maximum radius of the glomerular region. This indicates taking the maximum value. This represents the coordinates of a point in G. Indicates the glomerular region. Indicates the center of the glomerular region. This indicates the calculation of the Euclidean distance between two points.

4. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 2, characterized in that, S2 includes: When the glomerular region intersects with the internal mesangial region, the mesangial centroid and direction vector are recorded. Using the direction vector as a reference direction, multiple asymmetric sectors are divided. The expression for the direction vector is: ； in, Represents the direction vector. Indicates the center of the glomerular region. Indicates the mesangial centroid; When there is no mesangial crossing, the glomerular region is divided into multiple sectors at equal angles.

5. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, S3 includes: Based on the mesangial crossing results, the region is divided into multiple sectors; The sampling window size is calculated for each sector to obtain the window area.

6. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 5, characterized in that, The expression for the window area is: ； ； ； in, Indicates the window area. Indicates the glomerular area, Indicates the coverage ratio. Indicates the number of sectors. This represents the area calculation function. Indicates the glomerular region. This indicates the side length of the square window.

7. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, The expression containing the condition is: ； in, Indicates the sampling window. Indicates a Centered on, with side length as Square sampling area Indicates the center of the square window. This indicates the side length of the square window. This indicates the glomerular region.

8. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, The expression for the avoidance condition is: ； in, Indicates the sampling window. Indicates the center of the square window. This indicates the side length of the square window. This refers to the mesangial area inside the glomerulus. This represents the empty set.

9. The basement membrane sampling method based on glomerular capillary branching characteristics according to claim 1, characterized in that, The expression for adjusting the window is: ； in, Indicates the side length of a square window. "Indicates an update operation, This represents the scaling factor.

Images