Method, system and device for the analysis of vertebral body morphology in rare spine diseases

Abstract

The present disclosure provides a vertebral rare disease vertebral body morphology analysis processing method, system and device, the method comprising: acquiring a plurality of sample vertebral body images under a target scene; performing segmentation processing on each sample vertebral body image to obtain a plurality of sample segmentation results of vertebral bodies; acquiring a plurality of dimension sample morphology representation parameters corresponding to each vertebral body; performing clustering processing based on different sample morphology representation parameters to obtain different clustering ranges to acquire target classification association information for distinguishing different vertebral body morphologies of vertebral rare diseases. In the present disclosure, a multi-segmentation model fusion segmentation processing is adopted for any one unlabeled sample vertebral body image to obtain the optimal segmentation area of any one vertebral body; clustering processing is performed based on the sample morphology representation parameters of each vertebral body to quickly and accurately obtain relevant information for distinguishing different vertebral body morphologies of vertebral rare diseases, thereby ensuring the accuracy and efficiency of the analysis and processing of vertebral body morphology.

Description

Technical Field

[0001] This disclosure relates to the field of medical image processing technology, and in particular to a method, system, and device for analyzing and processing the vertebral body morphology of a rare spinal disease. Background Technology

[0002] Rare bone diseases are a group of diseases with low incidence, diverse clinical manifestations, and difficult diagnosis. They are often accompanied by abnormal changes in vertebral body morphology, such as irregularity, flattening, and deformity of the vertebrae. Currently, abnormal vertebral morphology has significant phenotypic value in many rare genetic diseases (such as osteodystrophy and congenital spinal deformities). Assessing vertebral body shape based on X-ray images is one of the key steps in early detection and auxiliary diagnosis.

[0003] Traditional methods rely on manual annotation and experience-based judgment, which are subject to problems such as high subjectivity, low efficiency, and high cost. Furthermore, due to the low incidence of rare diseases, data collection is difficult, making conventional machine learning methods based on medical X-ray images unsuitable. In other words, existing methods for analyzing vertebral morphology in rare spinal diseases cannot meet the needs of practical applications. Summary of the Invention

[0004] The technical problem to be solved by this disclosure is to overcome the problems of low quality and high cost caused by manual implementation in the existing technology, and the shortcomings of machine learning-based solutions, such as insufficient training data and inability to guarantee the accuracy of vertebral morphology analysis in rare spinal diseases. The purpose is to provide a method, system and device for analyzing and processing vertebral morphology in rare spinal diseases.

[0005] This disclosure solves the above-mentioned technical problems through the following technical solution: In a first aspect, this disclosure provides a method for analyzing and processing the vertebral body morphology of rare spinal diseases, the method comprising: Acquire several sample cone images in the target scene; Based on a preset segmentation processing strategy, each of the sample vertebral images is segmented to obtain sample segmentation results for several vertebral bodies. Obtain sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result; Clustering is performed based on different sample morphology characterization parameters to obtain different clustering ranges; Based on the different clustering ranges, target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases is obtained.

[0006] Optionally, the step of obtaining target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases based on different clustering ranges includes: The average sample morphological parameters of several dimensions corresponding to different cones within each cluster range are calculated. Based on the average sample morphological parameters corresponding to different clustering ranges, a preset classification rule is used to classify the samples to obtain different morphological classification types for distinguishing different vertebral body morphologies in rare spinal diseases.

[0007] Optionally, the analysis and processing method further includes: Based on the average sample morphological parameters corresponding to different clustering ranges, the maximum average sample morphological parameter is obtained, and a first preliminary division result is obtained based on the maximum average sample morphological parameter. Obtain the sample morphological characterization parameters corresponding to all cones in the clustering range, calculate the quantiles corresponding to the sample morphological characterization parameters in different dimensions, and obtain a second preliminary division result based on the quantiles; Based on the first preliminary classification result and the second preliminary classification result, the target classification results of all vertebrae under different morphological classification types are obtained; And / or, The step of obtaining target classification association information for distinguishing different vertebral body morphologies of rare spinal diseases based on different clustering ranges includes: calculating the average sample morphological parameters of several dimensions corresponding to different vertebrae within each clustering range. Based on the average sample morphology parameters of each dimension corresponding to different clustering ranges, the minimum average sample morphology parameter under each dimension is obtained by filtering. Based on the minimum average sample morphological parameters, vertebral morphology determination rules are generated to distinguish different vertebral morphologies in rare spinal diseases.

[0008] Optionally, the sample morphology characterization parameters include irregularity or flatness; The different morphological classification types include irregularity category, flatness category, or normal; And / or, The analysis and processing method further includes: In a number of sample vertebral images, all vertebrae belonging to the same target segmentation result are labeled as the corresponding morphological classification type.

[0009] Optionally, the analysis and processing method further includes: Based on the preset segmentation processing strategy, the target vertebral image in the target scene is segmented to obtain several target segmentation results of vertebral bodies. Obtain several dimensions of target morphological representation parameters corresponding to each vertebra in the target segmentation result; Based on the target morphology representation parameters, the vertebral morphology determination rules are used to process the data to obtain the target morphology classification type of each vertebral morphology in the target vertebral image.

[0010] Optionally, the minimum average sample morphological parameters under different dimensions include minimum average flatness and minimum irregularity; The step of processing the target morphology based on the target morphology representation parameters using the vertebral morphology determination rules to obtain the target morphology classification type of each vertebra in the target vertebral image includes: In response to the current vertebra being greater than the minimum average flatness, the flatness category corresponding to the current vertebra is determined; In response to the current vertebra being less than or equal to the minimum average flatness and greater than the minimum irregularity, the irregularity category corresponding to the current vertebra is determined; In response to the current vertebral body being less than or equal to the minimum average flatness and less than or equal to the minimum irregularity, the current vertebral body is determined to be normal.

[0011] Optionally, the step of segmenting each sample vertebral image based on a preset segmentation strategy to obtain sample segmentation results for several vertebral bodies includes: For any of the sample vertebral images, different preset segmentation network models are used for segmentation processing to obtain corresponding vertebral segmentation prediction sets; wherein, the vertebral segmentation prediction set includes vertebral association parameters corresponding to each vertebra; Based on the vertebral segmentation prediction set, the geometric intersection-union ratio of the polygons corresponding to any two vertebrae is obtained, as well as the confidence level corresponding to each vertebra. The polygon non-maximum algorithm is used to process different vertebral segmentation prediction sets, as well as the corresponding geometric intersection-union ratio and confidence level, to obtain the optimal segmentation region corresponding to the same vertebra, which is then used as the sample segmentation result for the corresponding vertebra.

[0012] Optionally, the sample morphology characterization parameters include sample irregularity; The step of obtaining the sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result includes: Obtain the first sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; Based on the preset irregularity scoring formula, and different first sample association parameters and corresponding first weights, the sample irregularity corresponding to the vertebra is calculated; And / or, The sample morphology characterization parameters include sample flatness; The step of obtaining the sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result includes: Obtain the second sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; Based on the preset flatness scoring formula, and different second sample association parameters and corresponding second weights, the flatness of the sample corresponding to the vertebra is calculated.

[0013] Optionally, the first sample association parameters include at least one of compactness, concavity, directional deviation, and area integrity; the compactness is determined based on the ratio of area to the square of perimeter; the concavity is determined based on the ratio of contour area to convex hull area; the directional deviation is determined based on the aspect ratio of the minimum bounding rectangle of the contour; and the area integrity is determined based on the ratio of contour area to the area of ​​the minimum bounding rectangle. And / or, The second sample association parameters include at least one of the following: aspect ratio of the minimum bounding rectangle, diameter ratio, and non-circularity; the aspect ratio of the minimum bounding rectangle is determined based on the lengths of the two sides of the minimum bounding rectangle of the cone profile; the diameter ratio is determined based on the maximum length of the projection of the profile in any direction; and the non-circularity is determined based on the area and the square of the perimeter of the cone profile. And / or, The step of performing clustering based on different sample morphology representation parameters to obtain different clustering ranges includes: The KMeans algorithm (an unsupervised clustering algorithm based on iterative optimization) is used to cluster the morphological characteristics of each cone sample to obtain different clustering ranges. And / or, The different target scenarios include different types of people, different regions, or different data batches.

[0014] A second aspect of this disclosure provides an analysis and processing system for vertebral body morphology in rare spinal diseases, the analysis and processing system comprising: The image acquisition module is used to acquire several sample cone images in the target scene; The first segmentation processing module is used to segment each of the sample vertebral images based on a preset segmentation processing strategy to obtain sample segmentation results of several vertebral bodies. The first parameter acquisition module is used to acquire sample morphology representation parameters of several dimensions corresponding to each cone in the sample segmentation result. The clustering processing module is used to perform clustering processing based on different sample morphology representation parameters to obtain different clustering ranges; The classification association information acquisition module is used to acquire target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases based on different clustering ranges.

[0015] A third aspect of this disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement the method for analyzing and processing vertebral morphology of rare spinal diseases provided in the first aspect.

[0016] In a fourth aspect, this disclosure provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the method for analyzing and processing the vertebral morphology of rare spinal diseases provided in the first aspect.

[0017] In a fifth aspect, this disclosure provides a computer program product, comprising a computer program, characterized in that, when the computer program is executed by a processor, it implements the method for analyzing and processing vertebral morphology of rare spinal diseases as provided in the first aspect.

[0018] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.

[0019] The positive and progressive effects of this disclosure are as follows: In this scheme, any sample vertebral body image is segmented using a preset segmentation strategy to obtain the optimal segmentation region for any vertebral body, ensuring the accuracy of subsequent analysis. Clustering is performed based on the morphological characteristics of each vertebral body sample to automatically divide it into different cluster ranges. Then, based on the different cluster ranges, relevant information for distinguishing different vertebral body morphologies of rare spinal diseases is obtained, such as morphological classification type and vertebral body morphology determination rules. This achieves the goal of training a predictive model for vertebral body morphology of rare spinal diseases without relying on manually labeled images. Instead, a novel classification strategy is proposed to construct a weakly supervised classification system that does not require labeling. This ensures the accuracy and efficiency of vertebral body morphology analysis for rare spinal diseases, while also exhibiting good generalization and interpretability, and saving the high cost of manual labeling. In addition, it can match specific target scenes to obtain matching target classification association information, so as to ensure the accuracy of the determination of the vertebral morphology of the vertebra in any image under a specific target scene, thereby achieving high-precision and personalized analysis and processing effects, and meeting the usage requirements of higher-requirement application scenarios. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 1 of this disclosure; Figure 2This is a first flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 3 This is a schematic diagram of the segmented vertebral body in Embodiment 3 of this disclosure; Figure 4 This is a second flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 5 This is the third flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 6 This is the fourth flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 7 This is the fifth flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 8 This is the sixth flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure. Figure 9 This is a logical diagram corresponding to the vertebral morphology determination rule of Embodiment 2 of this disclosure; Figure 10 This is the seventh flowchart of the method for analyzing and processing vertebral body morphology in rare spinal diseases according to Embodiment 2 of this disclosure; Figure 11 This is a schematic diagram of the modules of the vertebral body morphology analysis and processing system for rare spinal diseases according to Embodiment 3 of this disclosure; Figure 12 This is a schematic diagram of the modules of the vertebral body morphology analysis and processing system for rare spinal diseases according to Embodiment 4 of this disclosure. Figure 13 This is a schematic diagram of the structure of the electronic device according to Embodiment 5 of this disclosure. Detailed Implementation

[0021] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.

[0022] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0023] In this embodiment of the disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good morals.

[0024] Example 1 like Figure 1 As shown in this embodiment, the method for analyzing and processing the vertebral body morphology of rare spinal diseases includes: S101. Obtain several sample cone images in the target scene; The vertebral images in the samples were all unannotated. The sample vertebral images include, but are not limited to, lateral X-ray images of the thoracic and lumbar vertebrae, corresponding to common image formats such as DICOM, JPG, and PNG (DICOM, JPG, and PNG each correspond to an image format).

[0025] S102. Based on the preset segmentation processing strategy, each sample vertebral image is segmented to obtain the sample segmentation results of several vertebral bodies. The preset segmentation processing strategy is determined based on at least two preset segmentation network models; the sample segmentation result is the optimal segmentation region corresponding to each vertebra.

[0026] S103. Obtain sample morphology representation parameters of several dimensions corresponding to each cone in the sample segmentation results. Among them, the sample morphology characterization parameters of different dimensions are the descriptive parameters of the vertebral body morphology constructed in this embodiment. The sample morphology characterization parameters of different dimensions correspond to the vertebral body's irregularity score, flatness score, and other index parameters.

[0027] Irregularity is used to reflect the geometric integrity, boundary smoothness, and degree of concavity of the vertebral shape; flatness is used to measure the elongation, non-uniformity, and non-circularity of the vertebral shape.

[0028] Specifically, the characteristics of irregular vertebral bodies are: complex boundaries and obvious morphological abnormalities; the characteristics of flattened vertebral bodies are: obvious directionality and significant compression deformation characteristics; the characteristics of normal vertebral bodies are: regular shape and no significant abnormalities.

[0029] S104. Clustering is performed based on different sample morphology representation parameters to obtain different clustering ranges; This involves using one or more preset clustering algorithms for clustering; for example, the KMeans algorithm can be used for clustering. In addition, the parameters of the preset clustering algorithms can be adjusted according to the actual needs of the scenario to further ensure the processing accuracy.

[0030] S105. Based on different clustering ranges, obtain target classification association information for different vertebral body morphologies used to distinguish rare spinal diseases.

[0031] Among them, the target classification association information corresponds to the morphological classification type, the vertebral morphology determination rules, etc.

[0032] For example, for an unlabeled vertebral X-ray dataset (including 500 images), the sample segmentation results of each vertebra in each image (10 vertebrae per image) are obtained by segmentation, resulting in 5000 sets of sample morphological characterization parameters (irregularity, flatness, etc.) corresponding to 5000 vertebrae. The sample morphological characterization parameters are then processed by unsupervised clustering and other methods to obtain classification association information.

[0033] In this embodiment, a preset segmentation strategy is used to segment any sample vertebral body image to obtain the optimal segmentation region for any vertebral body, ensuring the accuracy of subsequent analysis and processing. Clustering is performed based on the morphological characteristics of each vertebral body sample to automatically divide it into different clustering ranges. Then, based on the different clustering ranges, relevant information for distinguishing different vertebral body morphologies for rare spinal diseases is obtained, such as morphological classification type and vertebral body morphology determination rules. This achieves the goal of training a predictive model for vertebral body morphology of rare spinal diseases without relying on manually labeled images. Instead, a novel classification processing strategy is proposed to construct a weakly supervised classification system that does not require labeling. This ensures the accuracy and efficiency of vertebral body morphology analysis and processing for rare spinal diseases, while also having good generalization and interpretability, and saving the high cost of manual labeling.

[0034] In addition, it can match specific target scenes to obtain matching target classification association information, so as to ensure the accuracy of the determination of the vertebral morphology of the vertebra in any image under a specific target scene, thereby achieving high-precision and personalized analysis and processing effects, and meeting the usage requirements of higher-requirement application scenarios.

[0035] Example 2 The method for analyzing and processing the vertebral body morphology of rare spinal diseases in this embodiment is a further improvement on Embodiment 1, specifically: In a feasible solution, different target scenarios include different types of people, different regions, or different data batches.

[0036] This solution does not perform mixed processing on images from different scenes, but rather distinguishes between different scenes. It can match specific target scenes to obtain matching target classification association information, thereby ensuring the accuracy of determining the vertebral morphology of vertebrae in any image under a specific target scene. This achieves high-precision and personalized analysis and processing, meeting the usage requirements of more demanding application scenarios.

[0037] In a feasible solution, such as Figure 2 As shown, step S102 includes: S1021. For any sample vertebral body image, different preset segmentation network models are used for segmentation processing to obtain the corresponding vertebral body segmentation prediction sets. The vertebral segmentation prediction set includes vertebral body association parameters corresponding to each vertebra; Different pre-defined segmentation network models, including U-Net, ResNet variants, and Vertebra-segmentation tools (all existing image segmentation models), perform preliminary segmentation on the input image to obtain the contour region of each cone in the image; for example... Figure 3 As shown, this is the segmentation result obtained based on an arbitrary preset segmentation network model, where each polygonal region corresponds to a cone.

[0038] The vertebral association parameters for each vertebra include the contour coordinates of the vertebra in the image (e.g., the coordinates of each key point on the contour arranged clockwise), the confidence level of the enclosed polygon, etc.

[0039] S1022. Based on the vertebral segmentation prediction set, obtain the geometric intersection-union ratio of the polygons corresponding to any two vertebrae, and the confidence level corresponding to each vertebra. S1023. The polygonal non-maximum algorithm is used to process different vertebral segmentation prediction sets, as well as the corresponding geometric intersection-union ratio and confidence level, to obtain the optimal segmentation region corresponding to the same vertebra, and use it as the sample segmentation result of the corresponding vertebra.

[0040] In this embodiment, instead of using a single preset segmentation network model to segment the image, two or more preset segmentation network models are used to segment the image independently and then the segmentation results are fused together. This ensures that for any sample vertebral image, each vertebra can obtain the corresponding optimal segmentation region, which effectively improves the segmentation accuracy and efficiency of each vertebra in the image, thereby further ensuring the accuracy and reliability of subsequent processing results.

[0041] Different preset segmentation network models can segment the same vertebral body image and output a structured vertebral body segmentation prediction set to improve the efficiency and feasibility of subsequent computational processing.

[0042] For example, for the same sample vertebral image, P = {P1, P2, ..., Pn}, where n represents the nth preset segmentation network model, Pi represents the vertebral segmentation prediction set for each vertebra after the nth preset segmentation network model segments the sample vertebral image (1 < i ≤ n, i is an integer), and P is the polygon segmentation prediction set for all vertebrae in the sample vertebral image obtained by the n preset segmentation network models. Each element Pi in the polygon set contains the following related parameters: vertex: vertebra number (e.g., "T12, L1, L2"); points: polygon vertex sequence [(x1, y1), (x2, y2), ...]; score: polygon confidence score output by the model.

[0043] Different preset segmentation network models segment the same vertebral image and then fuse them. Finally, each vertebra retains only one final fused polygon result, which is the optimal segmentation region for that vertebra.

[0044] Specifically, the steps for obtaining the segmented region of each vertebra based on the fusion processing of multiple preset segmentation network models are as follows: (1) Grouped by vertebral body label The input polygon set P is categorized according to the vertebra label to obtain the candidate polygon set for each vertebra: ; in, Labels indicating all vertebrae; (2) Calculate the polygon-IoU (overlap). For any two polygons (Pi, Pj) within each pair Pv, their geometric intersection-union ratio (IoU) is defined as: ; in, Indicates the area of ​​the region. This represents the geometric intersection region of two polygons; This represents the union region.

[0045] (3) Sort in descending order of confidence level For all polygons in Pv, ​​sort them by their prediction confidence score: , making ; (4) Perform polygon nonmaximum suppression (polygon NMS) operation. For each vertebral body number, there is a set of polygons (i.e., prediction results from multiple models for the same vertebral body): 1. Sort by confidence score from high to low: First consider the prediction results with high scores and high model confidence; 2. Starting with the first polygon, designate it as the current candidate polygon; 3. Compare the overlap (LOU) with each subsequent polygon; Specifically, if the intersection-union ratio (LOU) of two polygons exceeds a set threshold T1ou (e.g., 0.5), it indicates that they have similar shapes and overlapping regions; therefore, the polygon with the higher score is retained, and the other is removed from the candidate list. 4. Repeat the above process until all polygons have been processed.

[0046] Each preset segmentation network model will segment different regions corresponding to the easy vertebra. The core goal of this process is to avoid multiple models predicting multiple overlapping regions of the same vertebra, so that only one optimal segmentation result is retained. For any two polygons (Pi, Pj), the specific calculation is as follows: ; Then only the polygon with the highest score is retained, and the polygonal region containing that polygon is taken as the optimal segmentation region of the corresponding cone.

[0047] Among them, T IoU The corresponding threshold can be preset to 0.4-0.7 and can be adjusted experimentally to meet the needs of actual scenarios.

[0048] (5) Merge the optimal segmentation results of all vertebrae in the same vertebral body image. After completing PolygonNMS for each vertebral body number (e.g., "T11", "T12", "L1"): 1. For each vertebra, at most one final polygon is retained; 2. Collect all NMS results for all vertebral body numbers and form the final output list; ; in, It is the set of retained results for the polygon with vertex number v (generally containing only 1 polygon); It is the final set of segmented polygons obtained after the fusion of all vertebrae; Finally, the output polygon set This represents the best segmentation result obtained by fusing multiple preset segmentation network models for each vertebra.

[0049] In a feasible scheme, the sample morphology characterization parameters include sample irregularity; like Figure 4 As shown, step S103 includes: S10311. Obtain the first sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; S10312. Based on the preset irregularity scoring formula, and different first sample association parameters and corresponding first weights, the sample irregularity of the corresponding vertebra is calculated. Among them, the parameters associated with the first sample include compactness, concavity, orientation deviation, and area integrity. Compactness is determined based on the ratio of area to the square of perimeter; Specifically, the formula for calculating compactness is as follows: ; Where compactness is the degree of compactness, P is the perimeter of the corresponding vertebral contour, and A is the area of ​​the corresponding vertebral contour. The minimum compactness is 1 under the ideal circle condition, and the less compact the shape (such as slender, polygonal, complex edge, etc.), the larger the value.

[0050] The degree of concavity is determined based on the ratio of the contour area to the convex hull area; Specifically, the formula for calculating the degree of indentation is as follows: ; Wherein, Solidity represents the ratio of the contour area of ​​the vertebral body to its convex hull area, (1-Solidity) represents the degree of concavity, A is the area of ​​the corresponding vertebral body contour, and Aconvex represents the area of ​​the corresponding vertebral body convex hull. The more concave areas there are, the smaller the degree of concavity. Therefore, (1-Solidity) can be used as an index of concavity.

[0051] A convex hull is a closed polygon that does not allow concave corners; it is the "maximum convex approximation" of a contour; it is usually larger, smoother, and "thicker" than the original contour; it completely encloses the original contour without any indentations. For example, the original contour may be a notched cone boundary; it may have indentations, twists, or distortions (such as a "W" shape or a concave "C" shape).

[0052] The directional deviation is determined based on the aspect ratio of the smallest bounding rectangle of the profile; Specifically, the aspect ratio of the smallest bounding rectangle of the vertebral body is taken: ; Where w and h are the length and width of the bounding rectangle of the contour, respectively, and Aspect Ratio represents the aspect ratio of the smallest bounding rectangle of the cone contour. The closer the Aspect Ratio value is to 1, the more square the shape; the farther away from 1, the more slender or deformed the shape. Furthermore, the offset is: ; AR_Deviation represents the degree of directional deviation.

[0053] Area completeness is determined based on the ratio of the outline area to the area of ​​the smallest circumscribed rectangle.

[0054] Specifically, the ratio of the vertebral body's outline area to the area of ​​its smallest bounding rectangle: ; Where Area Ratio represents the area integrity, and Arect is the area of ​​the smallest bounding rectangle of the cone; the smaller the area integrity, the more irregular the shape and the more severe the dispersion.

[0055] The calculation formula corresponding to the preset irregularity scoring formula is as follows: ; Wherein, the Irregularity Score represents the degree of irregularity. , , , The first weight corresponds to each dimension parameter, and the sum of these weights is 1. These weights can be adjusted based on experiments or experience.

[0056] In this scheme, based on the polygon of the optimal segmented region of the segmented vertebra, the corresponding compactness, concavity, orientation deviation, area integrity, etc. are calculated, and then substituted into the pre-constructed preset irregularity scoring formula for corresponding calculation, so as to determine and efficiently obtain the irregularity of the corresponding vertebra.

[0057] In a feasible scheme, the sample morphology characterization parameters include sample flatness; like Figure 5 As shown, step S103 includes: S10321. Obtain the second sample association parameter of the optimal segmentation region corresponding to each cone in the sample segmentation result; S10322. Based on the preset flatness scoring formula, as well as different second sample association parameters and corresponding second weights, the sample flatness of the corresponding vertebra is calculated.

[0058] Among them, the second sample association parameters include aspect ratio, diameter ratio, and non-circularity based on the minimum bounding rectangle; The aspect ratio of the minimum bounding rectangle is determined based on the lengths of the two sides of the minimum bounding rectangle of the cone contour; Specifically, the formula for calculating the aspect ratio of the minimum bounding rectangle is: ; Here, AR represents the aspect ratio of the minimum bounding rectangle, and w and h are the lengths of the two sides of the minimum bounding rectangle, respectively. The larger the AR value, the more slender and "flat" the shape is. When w=h, Aspect Ratio = 1, indicating that the shape is close to a square or circle. When one side is much longer than the other side, Aspect Ratio is significantly greater than 1, indicating that the shape of the region is "flattened" or "slender".

[0059] The diameter ratio is determined by the maximum length of the profile projected in any direction. Specifically, the Feret diameter is the maximum length of a profile projected in any direction. Also known as the "caliper diameter," it refers to the width of a shape's projection in a given direction. In simpler terms, it's the minimum distance between two parallel lines that can accommodate the profile. The specific calculation formula is as follows: ; Among them, D max D represents the maximum distance between pairs of contour points. min It represents the minimum non-zero distance between pairs of contour points.

[0060] Maximum Feret diameter (Feret max): The farthest distance between any two points in the profile; Minimum Feret diameter (Feret min): The shortest distance between pairs of profile points, excluding 0; Here, the ratio of the maximum Feret diameter to the minimum Feret diameter of the vertebral profile is used to measure the flatness of the profile. For regular shapes (such as circles and squares), this ratio is close to 1; for "elongated" or "flattened" structures, this ratio becomes very large; it is similar to the Aspect Ratio, but pays more attention to extreme cases of profile point distribution and is more sensitive to abnormal edges and fracture deformation; Non-circularity is determined based on the area and the square of the perimeter of the cone profile.

[0061] Specifically, the formula for calculating roundness is as follows: ; Where P is the perimeter of the corresponding vertebral contour, A is the area of ​​the corresponding vertebral contour, and Circularity is the roundness, with a maximum value of 1 (perfect circle). Therefore, the formula for calculating non-circularity is as follows: ; Non-circularity refers to non-circularity.

[0062] The calculation formula corresponding to the preset flatness scoring formula is as follows: ; The Flatness Score represents the degree of flatness. , , These are the second weights corresponding to each dimension parameter, and the sum of these weights is 1. They can be adjusted based on experiments or experience.

[0063] In this scheme, based on the polygon of the optimal segmented region of the segmented vertebra, the corresponding aspect ratio, diameter ratio, and non-circularity of the minimum bounding rectangle are calculated. Then, these are substituted into the pre-constructed preset flatness scoring formula for corresponding calculations to determine and efficiently obtain the flatness of the corresponding vertebra.

[0064] It is worth noting that the weights of each of the above-mentioned related parameters can be adjusted according to the actual needs of the scenario in order to obtain high-precision processing results.

[0065] In a feasible solution, such as Figure 6 As shown, step S105 includes: S10511. Calculate the average sample morphological parameters of several dimensions corresponding to different cones within each cluster range; Sample morphology parameters include sample irregularity and sample flatness; average sample morphology parameters include average sample irregularity and average sample flatness. Taking the average sample irregularity as an example, the average sample irregularity of all cones within a certain cluster range is summed and averaged to obtain the average sample irregularity within that cluster range. Taking the average flatness of samples as an example, the average flatness of samples corresponding to all cones within a certain cluster range is obtained by summing and averaging the flatness of samples within that cluster range.

[0066] At this point, we only obtain different clustering ranges. The sample irregularity and sample flatness of each vertebra are used as two-dimensional feature vectors (vertebra morphology vectors). The KMeans algorithm is used to perform unsupervised clustering on these vectors. For example, if the number of clusters k=3 is set, the KMeans algorithm can calculate the Euclidean distance between these vectors to automatically divide the data corresponding to all vertebrae and obtain different clustering ranges. However, we do not know the specific type of each clustering range at this time.

[0067] S10512. Based on the average sample morphological parameters corresponding to different clustering ranges, classification is performed using preset classification rules to obtain different morphological classification types for distinguishing different vertebral body morphologies in rare spinal diseases.

[0068] Among them, the sample morphology characterization parameters include irregularity or flatness; different morphology classification types include irregularity category, flatness category, or normal; Specifically, the average sample morphology parameters corresponding to each cluster range are: avg(irregularity) and avg(flatness). Irregularity represents the degree of irregularity, and avg(irregularity) represents the average irregularity of the samples; flatness represents the degree of flatness, and avg(flatness) represents the average flatness of the samples.

[0069] This approach enables the rapid and reliable determination of different morphological classifications of vertebral body morphology in rare spinal diseases based on different clustering ranges obtained through clustering.

[0070] In a feasible solution, such as Figure 7 As shown, the analysis and processing method also includes: S10513. Based on the average sample morphological parameters corresponding to different clustering ranges, obtain the maximum average sample morphological parameter, and obtain the first preliminary division result based on the maximum average sample morphological parameter. S10514. Obtain the sample morphological representation parameters corresponding to all cones in the clustering range, calculate the quantiles corresponding to the sample morphological representation parameters of different dimensions, and obtain the second preliminary division result based on the quantiles. S10515. Based on the first and second preliminary classification results, the target classification results of all vertebrae under different morphological classification types are obtained.

[0071] For example, assuming that clustering results in three cluster ranges, the cone of the cluster range containing the largest average irregularity of samples is identified as the irregular class; the cone of the cluster range containing the largest average flatness of samples is identified as the flat class; and the cone of the remaining cluster range is identified as the normal class. This gives the first preliminary division result corresponding to the cone. Meanwhile, based on the sample morphology representation parameters corresponding to different clustering ranges, the quantiles of the sample morphology representation parameters under each dimension were statistically obtained; based on the quantiles, all vertebrae were classified to obtain different morphological classification types for distinguishing different vertebrae morphologies in rare spinal diseases. For example, the first quantile (Q1) and the second quantile (Q2) of all vertebrae in the two dimensions of irregularity and flatness are calculated. The first quantile (Q1) and the second quantile (Q2) can be the 75th percentile. Of course, the first quantile and the second quantile can be fine-tuned according to the actual situation to meet the calculation requirements of the actual scenario.

[0072] Vertebrae with irregularity greater than the first quantile are classified as irregular, and vertebrae with flatness greater than the second quantile are classified as flat. At this point, the cones belonging to the irregular class obtained based on the maximum sample average irregularity are intersected with the cones belonging to the irregular class obtained based on the first quantile (Q1), and finally all the cones in the intersection are determined to correspond to the irregular class. Similarly, the intersection of the cones belonging to the flat class obtained based on the maximum average flatness of the samples and the cones belonging to the flat class obtained based on the second quantile (Q2) is taken, and finally all the cones in the intersection are determined to belong to the flat class. All other remaining vertebrae were classified as normal.

[0073] In this embodiment, instead of using a single category division method, a method based on the maximum average sample morphological parameters and quantiles is fused to determine the classification, which effectively ensures the accuracy and efficiency of the classification processing of all sample vertebral images.

[0074] In one feasible embodiment, step S10512 is followed by: In a number of sample vertebral images, all vertebrae belonging to the same target segmentation result are labeled as the corresponding morphological classification type.

[0075] In this scheme, based on the determined morphological classification type, all vertebrae within the same cluster range are labeled with the same corresponding morphological classification type, thereby enabling rapid, efficient, and accurate batch labeling of several sample vertebrae images that have not been manually labeled, in order to meet the needs of subsequent model training and other purposes.

[0076] In a feasible solution, such as Figure 8 As shown, step S105 includes: S10521. Calculate the average sample morphological parameters of several dimensions corresponding to different cones within each cluster range; S10522. Based on the average sample morphology parameters of each dimension corresponding to different clustering ranges, the minimum average sample morphology parameter under each dimension is selected. For example, the minimum average sample morphological parameters include the minimum average irregularity and the minimum average flatness.

[0077] S10523. Based on the minimum average sample morphological parameters, generate vertebral morphology determination rules to distinguish different vertebral morphologies for rare spinal diseases.

[0078] In this scheme, the minimum average irregularity in all cluster ranges is used as the first threshold t1 to determine whether the vertebral body morphology is irregular or normal; the minimum average flatness in all cluster ranges is used as the second threshold t2 to determine whether the vertebral body morphology is flat or normal; based on the first threshold t1 and the second threshold t2, the vertebral body morphology determination rules for distinguishing different vertebral body morphologies for rare spinal diseases are determined, so that the vertebral body morphology determination rules can be obtained quickly and with high quality, thereby ensuring that for any vertebral body image, the vertebral body morphology of each vertebra can be automatically, efficiently and with high quality identified.

[0079] For a given target scenario, there are generally stable first threshold t1 and second threshold t2.

[0080] Specifically, such as Figure 9 As shown, the constructed vertebral body morphology determination rules include: (1) The irregularity and flatness of each vertebra in any vertebral body image are obtained by fusing the segmentation using a preset segmentation processing strategy; (2) For any vertebral body, in response to the flatness being greater than the second threshold t2, the vertebral body is determined to be of the flat type; In response to a flatness less than or equal to the second threshold t2 and an irregularity greater than the first threshold t1, the vertebra is determined to be of the irregular type. The vertebra is determined to be normal if the flatness is less than or equal to the second threshold t2 and the irregularity is less than or equal to the first threshold t1.

[0081] In a feasible solution, such as Figure 10 As shown, the analysis and processing method also includes: S201. Based on the preset segmentation processing strategy, the target vertebral image in the target scene is segmented to obtain the target segmentation results of several vertebral bodies. S202. Obtain several dimensions of target morphological representation parameters corresponding to each vertebra in the target segmentation result; S203. Based on the target morphology representation parameters, the vertebral morphology determination rules are used to process the data to obtain the target morphology classification type of each vertebral morphology in the target vertebral image.

[0082] In a feasible scheme, the minimum average sample morphological parameters under different dimensions include the minimum average flatness and the minimum irregularity; Step S203 includes: In response to the current vertebral body being greater than the minimum average flatness, the flatness category corresponding to the current vertebral body is determined; In response to the current vertebral body being less than or equal to the minimum mean flatness and greater than the minimum irregularity, the irregularity category corresponding to the current vertebral body is determined; If the current vertebral body is less than or equal to the minimum mean flatness and less than or equal to the minimum irregularity, the current vertebral body is determined to be normal.

[0083] In this scheme, a preset segmentation processing strategy is used to perform fusion segmentation processing on the target vertebral images in the target scene to obtain a high-accuracy target segmentation result for each vertebral body; the target irregularity of the vertebral body is obtained by a calculation method similar to S10311-S10312, and the target flatness of the vertebral body is obtained by a calculation method similar to S10321-S10322; then, the constructed vertebral body morphology judgment rule is used to judge, so as to quickly and accurately obtain the target morphology classification type of each vertebral body morphology.

[0084] In addition, the first weight, the second weight, the parameters of the clustering algorithm, and the model parameters of different preset segmentation models in this embodiment can all be fine-tuned during the initial analysis or intermediate analysis process, so as to obtain the analysis results in a timely and highly accurate manner.

[0085] Example 3 like Figure 11 As shown, the vertebral body morphology analysis and processing system for rare spinal diseases in this embodiment includes: Image acquisition module 1 is used to acquire several sample cone images in the target scene; The first segmentation processing module 2 is used to segment each sample vertebral image based on a preset segmentation processing strategy to obtain sample segmentation results of several vertebral bodies. The first parameter acquisition module 3 is used to acquire sample morphology representation parameters of several dimensions corresponding to each cone in the sample segmentation result. Clustering module 4 is used to perform clustering based on different sample morphology representation parameters to obtain different clustering ranges; The classification association information acquisition module 5 is used to acquire target classification association information for different vertebral body morphologies used to distinguish rare spinal diseases, based on different clustering ranges.

[0086] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0087] Example 4 like Figure 12 As shown, the vertebral body morphology analysis and processing system for rare spinal diseases in this embodiment is a further improvement on Embodiment 3, specifically: In a feasible solution, different target scenarios include different types of people, different regions, or different batches of data.

[0088] In one feasible solution, the classification association information acquisition module 5 is also used for: The average sample morphological parameters of several dimensions corresponding to different vertebrae within each cluster range are calculated. Based on the average sample morphological parameters corresponding to different cluster ranges, a preset classification rule is used to classify them to obtain different morphological classification types for different vertebrae morphologies used to distinguish rare spinal diseases.

[0089] In one feasible embodiment, the analysis and processing system also includes: The first partitioning result acquisition module 6 is used to obtain the maximum average sample morphological parameter based on the average sample morphological parameters corresponding to different clustering ranges, and to obtain the first preliminary partitioning result based on the maximum average sample morphological parameter. The second partitioning result acquisition module 7 is used to acquire the sample morphological characterization parameters corresponding to all cones in the clustering range, calculate the quantiles corresponding to the sample morphological characterization parameters in different dimensions, and acquire the second preliminary partitioning result based on the quantiles. The target segmentation result acquisition module 8 is used to obtain the target segmentation results of all vertebrae under different morphological classification types based on the first preliminary segmentation result and the second preliminary segmentation result; In one feasible solution, the classification association information acquisition module 5 is also used for: The average sample morphological parameters of several dimensions corresponding to different cones within each cluster were calculated. Based on the average sample morphology parameters of each dimension corresponding to different clustering ranges, the minimum average sample morphology parameters under each dimension are selected. Based on the minimum average sample morphological parameters, vertebral morphology determination rules are generated to distinguish different vertebral morphologies in rare spinal diseases.

[0090] In a feasible scheme, the sample morphology characterization parameters include irregularity or flatness; Different morphological classification types include irregularity category, flatness category, or normal; In a feasible solution, the analysis and processing system also includes: Image labeling module 9 is used to label all vertebrae in several sample vertebrae images that belong to the same target classification result as the corresponding morphological classification type.

[0091] In a feasible solution, the analysis and processing system also includes: The target segmentation processing module 10 is used to segment the target vertebral image in the target scene based on a preset segmentation processing strategy to obtain target segmentation results of several vertebral bodies. The target parameter acquisition module 11 is used to acquire several dimensions of target morphological representation parameters corresponding to each vertebra in the target segmentation result. The target classification processing module 12 is used to process the target morphology based on the target morphology representation parameters using vertebral morphology determination rules to obtain the target morphology classification type of each vertebral morphology in the target vertebral image.

[0092] In a feasible scheme, the minimum average sample morphological parameters under different dimensions include the minimum average flatness and the minimum irregularity; The target classification processing module 12 is also used for: In response to the current vertebral body being greater than the minimum average flatness, the flatness category corresponding to the current vertebral body is determined; In response to the current vertebral body being less than or equal to the minimum mean flatness and greater than the minimum irregularity, the irregularity category corresponding to the current vertebral body is determined; If the current vertebral body is less than or equal to the minimum mean flatness and less than or equal to the minimum irregularity, the current vertebral body is determined to be normal.

[0093] In one feasible embodiment, the first segmentation processing module 2 includes: The segmentation unit is used to segment any sample vertebral body image using different preset segmentation network models to obtain corresponding vertebral body segmentation prediction sets; wherein, the vertebral body segmentation prediction set includes the vertebral body association parameters corresponding to each vertebral body; The vertebral body parameter calculation unit is used to obtain the geometric intersection-union ratio of the polygons corresponding to any two vertebrae, as well as the confidence level of each vertebra, based on the vertebral body segmentation prediction set. The segmentation result acquisition unit is used to process different vertebral segmentation prediction sets, as well as the corresponding geometric intersection-union ratio and confidence level, using a polygon non-maximum algorithm to obtain the optimal segmentation region corresponding to the same vertebra, and use it as the sample segmentation result of the corresponding vertebra.

[0094] In one feasible scheme, the sample morphology characterization parameters include sample irregularity; the first parameter acquisition module 3 is also used to acquire the first sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; based on the preset irregularity scoring formula, as well as different first sample association parameters and corresponding first weights, the sample irregularity of the corresponding vertebra is calculated.

[0095] In one feasible scheme, the sample morphology characterization parameters include sample flatness; the first parameter acquisition module 3 is also used to acquire the second sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; based on the preset flatness scoring formula, as well as different second sample association parameters and corresponding second weights, the sample flatness of the corresponding vertebra is calculated.

[0096] In one feasible scheme, the first sample association parameters include at least one of compactness, concavity, orientation deviation, and area integrity; compactness is determined based on the ratio of area to the square of perimeter; concavity is determined based on the ratio of contour area to convex hull area; orientation deviation is determined based on the aspect ratio of the minimum bounding rectangle of the contour; and area integrity is determined based on the ratio of contour area to the area of ​​the minimum bounding rectangle.

[0097] In one feasible scheme, the second sample association parameters include at least one of the following: aspect ratio of the minimum bounding rectangle, diameter ratio, and non-circularity; the aspect ratio of the minimum bounding rectangle is determined based on the lengths of the two sides of the minimum bounding rectangle of the cone profile; the diameter ratio is determined based on the maximum length of the projection of the profile in any direction; and the non-circularity is determined based on the area and the square of the perimeter of the cone profile.

[0098] In one feasible scheme, the clustering processing module 4 is also used to perform clustering processing on the sample morphological characterization parameters of each cone using the KMeans algorithm to obtain different clustering ranges.

[0099] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0100] Example 5 Figure 13This is a schematic diagram of the structure of an electronic device according to an example embodiment of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the method described in any of the above embodiments. Figure 13 The electronic device 90 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0101] like Figure 13 As shown, the electronic device 90 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including memory 92 and processor 91).

[0102] Bus 93 includes a data bus, an address bus, and a control bus.

[0103] The memory 92 may include volatile memory, such as random access memory (RAM) 921 and / or cache memory 922, and may further include read-only memory (ROM) 923.

[0104] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) program module 924, such program module 924 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0105] The processor 91 executes various functional applications and data processing, such as the methods provided in any of the above embodiments, by running computer programs stored in the memory 92.

[0106] Electronic device 90 can also communicate with one or more external devices 94 (e.g., keyboard, pointing device, etc.). This communication can be performed through input / output (I / O) interface 95. Furthermore, electronic device 90 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 96. As shown, network adapter 96 communicates with other modules of electronic device 90 via bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0107] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0108] Example 6 This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.

[0109] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0110] Example 7 This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the above embodiments.

[0111] The program code for executing the computer program product of this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0112] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.

Claims

1. A method for analyzing and processing vertebral body morphology in a rare spinal disease, characterized in that, The analysis and processing method includes: Acquire several sample cone images in the target scene; Based on a preset segmentation processing strategy, each of the sample vertebral images is segmented to obtain sample segmentation results for several vertebral bodies. Obtain sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result; Clustering is performed based on different sample morphology characterization parameters to obtain different clustering ranges; Based on the different clustering ranges, target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases is obtained.

2. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 1, characterized in that, The step of obtaining target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases based on different clustering ranges includes: The average sample morphological parameters of several dimensions corresponding to different cones within each cluster range are calculated. Based on the average sample morphological parameters corresponding to different clustering ranges, a preset classification rule is used to classify the samples to obtain different morphological classification types for distinguishing different vertebral body morphologies in rare spinal diseases.

3. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 2, characterized in that, The analysis and processing method further includes: Based on the average sample morphological parameters corresponding to different clustering ranges, the maximum average sample morphological parameter is obtained, and a first preliminary division result is obtained based on the maximum average sample morphological parameter. Obtain the sample morphological characterization parameters corresponding to all cones in the clustering range, calculate the quantiles corresponding to the sample morphological characterization parameters in different dimensions, and obtain a second preliminary division result based on the quantiles; Based on the first preliminary classification result and the second preliminary classification result, the target classification results of all vertebrae under different morphological classification types are obtained; And / or, The step of obtaining target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases based on different clustering ranges includes: The average sample morphological parameters of several dimensions corresponding to different cones within each cluster range are calculated. Based on the average sample morphology parameters of each dimension corresponding to different clustering ranges, the minimum average sample morphology parameter under each dimension is obtained by filtering. Based on the minimum average sample morphological parameters, vertebral morphology determination rules are generated to distinguish different vertebral morphologies in rare spinal diseases.

4. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 3, characterized in that, The sample morphology characterization parameters include irregularity or flatness; The different morphological classification types include irregularity category, flatness category, or normal; And / or, The analysis and processing method further includes: In a number of sample vertebral images, all vertebrae belonging to the same target segmentation result are labeled as the corresponding morphological classification type.

5. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 4, characterized in that, The analysis and processing method further includes: Based on the preset segmentation processing strategy, the target vertebral image in the target scene is segmented to obtain several target segmentation results of vertebral bodies. Obtain several dimensions of target morphological representation parameters corresponding to each vertebra in the target segmentation result; Based on the target morphology representation parameters, the vertebral morphology determination rules are used to process the data to obtain the target morphology classification type of each vertebral morphology in the target vertebral image.

6. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 5, characterized in that, The minimum average sample morphological parameters under different dimensions include minimum average flatness and minimum irregularity; The step of processing the target morphology based on the target morphology representation parameters using the vertebral morphology determination rules to obtain the target morphology classification type of each vertebra in the target vertebral image includes: In response to the current vertebra being greater than the minimum average flatness, the flatness category corresponding to the current vertebra is determined; In response to the current vertebra being less than or equal to the minimum average flatness and greater than the minimum irregularity, the irregularity category corresponding to the current vertebra is determined; In response to the current vertebral body being less than or equal to the minimum average flatness and less than or equal to the minimum irregularity, the current vertebral body is determined to be normal.

7. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in any one of claims 1 to 6, characterized in that, The step of segmenting each sample vertebral image based on a preset segmentation strategy to obtain sample segmentation results for several vertebral bodies includes: For any of the sample vertebral images, different preset segmentation network models are used for segmentation processing to obtain corresponding vertebral segmentation prediction sets; wherein, the vertebral segmentation prediction set includes vertebral association parameters corresponding to each vertebra; Based on the vertebral segmentation prediction set, the geometric intersection-union ratio of the polygons corresponding to any two vertebrae is obtained, as well as the confidence level corresponding to each vertebra. The polygon non-maximum algorithm is used to process different vertebral segmentation prediction sets, as well as the corresponding geometric intersection-union ratio and confidence level, to obtain the optimal segmentation region corresponding to the same vertebra, which is then used as the sample segmentation result for the corresponding vertebra.

8. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 7, characterized in that, The sample morphology characterization parameters include sample irregularity; The step of obtaining the sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result includes: Obtain the first sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; Based on the preset irregularity scoring formula, and different first sample association parameters and corresponding first weights, the sample irregularity corresponding to the vertebra is calculated; And / or, The sample morphology characterization parameters include sample flatness; The step of obtaining the sample morphological representation parameters of several dimensions corresponding to each vertebra in the sample segmentation result includes: Obtain the second sample association parameter of the optimal segmentation region corresponding to each vertebra in the sample segmentation result; Based on the preset flatness scoring formula, and different second sample association parameters and corresponding second weights, the flatness of the sample corresponding to the vertebra is calculated.

9. The method for analyzing and processing vertebral body morphology in rare spinal diseases as described in claim 8, characterized in that, The first sample association parameters include at least one of compactness, concavity, orientation deviation, and area integrity; the compactness is determined based on the ratio of area to the square of perimeter; the concavity is determined based on the ratio of contour area to convex hull area; the orientation deviation is determined based on the aspect ratio of the minimum bounding rectangle of the contour; and the area integrity is determined based on the ratio of contour area to the area of ​​the minimum bounding rectangle. And / or, The second sample association parameters include at least one of the following: aspect ratio of the minimum bounding rectangle, diameter ratio, and non-circularity; the aspect ratio of the minimum bounding rectangle is determined based on the lengths of the two sides of the minimum bounding rectangle of the cone profile; the diameter ratio is determined based on the maximum length of the projection of the profile in any direction; and the non-circularity is determined based on the area and the square of the perimeter of the cone profile. And / or, The step of performing clustering based on different sample morphology representation parameters to obtain different clustering ranges includes: The KMeans algorithm was used to cluster the morphological characterization parameters of each vertebra to obtain different clustering ranges. And / or, The different target scenarios include different types of people, different regions, or different data batches.

10. A system for analyzing and processing vertebral body morphology in rare spinal diseases, characterized in that, The analysis and processing system includes: The image acquisition module is used to acquire several sample cone images in the target scene; The first segmentation processing module is used to segment each of the sample vertebral images based on a preset segmentation processing strategy to obtain sample segmentation results of several vertebral bodies. The first parameter acquisition module is used to acquire sample morphology representation parameters of several dimensions corresponding to each cone in the sample segmentation result. The clustering processing module is used to perform clustering processing based on different sample morphology representation parameters to obtain different clustering ranges; The classification association information acquisition module is used to acquire target classification association information for distinguishing different vertebral body morphologies in rare spinal diseases based on different clustering ranges.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the method for analyzing and processing the vertebral morphology of rare spinal diseases as described in any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for analyzing and processing the vertebral morphology of any one of the rare spinal diseases according to claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for analyzing and processing the vertebral morphology of rare spinal diseases as described in any one of claims 1 to 9.

Images