Model training method, and centrum compression fracture condition identification method and system

Abstract

The invention provides a model training method and a centrum compression fracture condition identification method and system. The training method comprises the following steps: acquiring a plurality of sample centrum images; obtaining sample centrum feature data corresponding to each sample centrum in each sample centrum image, and sample labeling data representing whether integral compression occurs in each sample centrum image; wherein the sample centrum feature data comprises sample area associated data of the sample centrum; and performing model training based on the sample centrum feature data of each sample centrum corresponding to each sample centrum image and the corresponding sample labeling data to obtain a centrum compression fracture analysis model used for outputting whether the centrum in any target centrum image is subjected to integral compression or not. According to the scheme, the precision, efficiency and reliability of the prediction result of whether the centrum in any target centrum image is subjected to integral compression or not output by the centrum compression fracture analysis model obtained through training are effectively ensured.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technologies, and particularly to a model training method, a method and a system for identifying vertebral compression fractures. Background Art

[0002] Currently, for the diagnosis of vertebral compression fractures, X-ray films are generally used, which have low detection costs and faster detection speeds. Clinically, when identifying compression fractures, it is necessary to evaluate and determine the location and degree of the fractures. However, currently, the treatment plans for vertebral (such as thoracolumbar) compression fractures generally have problems such as complex treatment processes and low accuracy. Summary of the Invention

[0003] The technical problem to be solved by the present disclosure is to overcome the above-mentioned defects in the prior art and provide a model training method, a method and a system for identifying vertebral compression fractures.

[0004] The present disclosure solves the above technical problem through the following technical solutions:

[0005] The present disclosure provides a training method for a vertebral compression fracture analysis model, and the training method includes:

[0006] Obtain a plurality of sample vertebral images;

[0007] Obtain the sample vertebral feature data corresponding to each sample vertebral in each of the sample vertebral images, and the sample annotation data indicating whether overall compression has occurred in each of the sample vertebral images; wherein, the sample vertebral feature data includes the sample area correlation data of the sample vertebral;

[0008] Based on the sample vertebral feature data corresponding to each sample vertebral in each of the sample vertebral images and the corresponding sample annotation data, perform model training to obtain the vertebral compression fracture analysis model for outputting whether the vertebral body in any target vertebral image has undergone overall compression.

[0009] Optionally, the sample area correlation data includes at least one of sample theoretical area, sample actual area, difference between the sample theoretical area and the sample theoretical area of an adjacent vertebral body, ratio of the sample theoretical area of the sample vertebral body to the sample theoretical area of an adjacent vertebral body, ratio of the sample actual area to the sample theoretical area, convex hull area, and ratio of the convex hull area to the theoretical area;

[0010] And / or

[0011] The sample vertebral feature data further includes the sample position data and / or sample length correlation data of the sample vertebral;

[0012] Among them, the sample length-related data includes at least one of the length of the posterior edge of the vertebral body, the ratio of the length to the posterior edge of the adjacent vertebral body, and the difference in length from the posterior edge of the adjacent vertebral body.

[0013] Optionally, after the step of obtaining the sample vertebral body feature data corresponding to each vertebral body and before model training, the training method further includes:

[0014] Performing feature screening processing on the sample vertebral body feature data corresponding to each vertebral body to obtain the sample vertebral body feature data after screening processing;

[0015] And / or, the step of calculating the sample true area of the sample vertebral body includes:

[0016] Using a preset edge processing rule to identify each edge in the sample vertebral body to obtain a true edge line corresponding to each edge;

[0017] Generating a true edge graph corresponding to the vertebral body based on each true edge line;

[0018] Obtaining the area corresponding to the true edge graph and using it as the sample true area of the sample vertebral body;

[0019] And / or,

[0020] The step of calculating the sample theoretical area of the sample vertebral body includes:

[0021] Using a preset edge processing rule to identify each edge in the sample vertebral body to obtain a true edge line corresponding to each edge;

[0022] Among them, the true edge line includes a true anterior edge line, a true posterior edge line, a true upper edge line, and a true lower edge line;

[0023] Forming a closed graph based on each true edge line and using it as the true edge graph corresponding to the vertebral body;

[0024] Generating a circumscribed rectangle of the true edge graph according to the true anterior edge line, the true posterior edge line, and the true edge graph, and using the circumscribed rectangle as the theoretical shape of the sample vertebral body;

[0025] Obtaining the area corresponding to the theoretical shape and using it as the sample theoretical area of the sample vertebral body.

[0026] The present disclosure also provides a method for identifying the condition of vertebral compression fracture, and the identification method includes:

[0027] Obtaining a target vertebral body image;

[0028] Obtain the target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image;

[0029] Among them, the target vertebral body feature data includes the target area correlation data of the target vertebral body;

[0030] Input the target area correlation data into the vertebral body compression fracture analysis model to obtain the target recognition result of whether the vertebral body in the target vertebral body image has undergone overall compression;

[0031] Among them, the vertebral body compression fracture analysis model is obtained based on the training method of the vertebral body compression fracture analysis model as described above.

[0032] Optionally, in response to the current overall compression of the target vertebral body, a preset correction scheme is used to determine the fracture degree of the target vertebral body.

[0033] Optionally, the step of, in response to the current overall compression of the target vertebral body, using a preset correction scheme to determine the fracture degree of the target vertebral body includes:

[0034] In response to the target recognition result indicating overall compression, obtain the first theoretical area of the previous vertebral body of the current target vertebral body and the second theoretical area of the next vertebral body;

[0035] Use the weighted average of the first theoretical area and the second theoretical area as the new theoretical area of the current target vertebral body;

[0036] Obtain the target real area of the current target vertebral body;

[0037] Calculate the first ratio of the target real area and the new theoretical area of the current target vertebral body;

[0038] Determine the fracture degree of the current target vertebral body based on the first ratio;

[0039] Among them, different fracture degrees correspond to different preset ratio ranges;

[0040] Or,

[0041] The step of, in response to the current overall compression of the target vertebral body, using a preset correction scheme to determine the fracture degree of the target vertebral body includes:

[0042] In response to the current overall compression of the target vertebral body, obtain the target real area and the target theoretical area of the current target vertebral body;

[0043] Calculate the second ratio of the target true area and the target theoretical area of the current target vertebral body, and multiply the second ratio by a preset coefficient to obtain a third ratio;

[0044] Determine the fracture classification of the current target vertebral body based on the third ratio;

[0045] Wherein, different fracture classifications correspond to different preset ratio ranges.

[0046] Optionally, the recognition method further includes:

[0047] In response to the target recognition result indicating that no overall compression has occurred, obtain the target true area and the target theoretical area corresponding to the current target vertebral body;

[0048] Calculate the fourth ratio of the target true area and the target theoretical area corresponding to each target vertebral body;

[0049] Determine the fracture classification of the current target vertebral body based on the fourth ratio;

[0050] Wherein, different fracture classifications correspond to different preset ratio ranges;

[0051] And / or,

[0052] The step of calculating the target true area of the target vertebral body includes:

[0053] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body to obtain a true edge line corresponding to each edge;

[0054] Generate a true edge graph corresponding to the vertebral body based on each true edge line;

[0055] Obtain the area corresponding to the true edge graph and use it as the target true area of the target vertebral body;

[0056] And / or,

[0057] The step of calculating the target theoretical area of the target vertebral body includes:

[0058] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body to obtain a true edge line corresponding to each edge;

[0059] Wherein, the true edge line includes a true front edge line, a true rear edge line, a true upper edge line, and a true lower edge line;

[0060] Form a closed graph based on each true edge line and use it as the true edge graph corresponding to the vertebral body;

[0061] Generate a circumscribed rectangle of the true edge graph based on the true leading edge line, the true trailing edge line, and the true edge graph, and use the circumscribed rectangle as the theoretical shape of the target vertebral body;

[0062] Obtain the area corresponding to the theoretical shape and use it as the target theoretical area of the target vertebral body.

[0063] The present disclosure also provides a training system for a vertebral compression fracture analysis model, and the training system includes:

[0064] A sample image acquisition module for acquiring a plurality of sample vertebral body images;

[0065] A sample training data acquisition module for acquiring sample vertebral body feature data corresponding to each sample vertebral body in each sample vertebral body image, and sample annotation data indicating whether overall compression has occurred in each sample vertebral body image; wherein, the sample vertebral body feature data includes sample area correlation data of the sample vertebral body.

[0066] A model training module for performing model training based on the sample vertebral body feature data of each sample vertebral body corresponding to each sample vertebral body image and the corresponding sample annotation data, so as to obtain the vertebral compression fracture analysis model for outputting whether the vertebral body in any target vertebral body image has undergone overall compression.

[0067] The present disclosure also provides a recognition system for the condition of vertebral compression fractures, and the recognition system includes:

[0068] A target image acquisition module for acquiring a target vertebral body image;

[0069] A target feature acquisition module for acquiring target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image;

[0070] Wherein, the target vertebral body feature data includes target area correlation data of the target vertebral body.

[0071] A target recognition result acquisition module for inputting the target area correlation data into the vertebral compression fracture analysis model to obtain a target recognition result indicating whether the vertebral body in the target vertebral body image has undergone overall compression;

[0072] Wherein, the vertebral compression fracture analysis model is obtained based on the training system for the vertebral compression fracture analysis model as described above.

[0073] The present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and configured to run on the processor. When the processor executes the computer program, the training method of the vertebral compression fracture analysis model as described above is implemented; or, the recognition method of the vertebral compression fracture situation as described above.

[0074] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the training method of the vertebral compression fracture analysis model as described above is implemented; or, the recognition method of the vertebral compression fracture situation as described above.

[0075] The present disclosure also provides a computer program product, including a computer program. When the computer program is executed by a processor, the training method of the vertebral compression fracture analysis model as described above is implemented; or, the recognition method of the vertebral compression fracture situation as described above.

[0076] On the basis of conforming to the common knowledge in the art, the above preferred conditions can be combined arbitrarily to obtain various preferred examples of the present disclosure.

[0077] The positive and progressive effects of the present disclosure are as follows:

[0078] In the present disclosure, the sample data is considered and designed skillfully and reasonably. By obtaining the sample vertebral feature data of each vertebral body in the sample vertebral image, especially calculating the sample area correlation data corresponding to the vertebral body, that is, the comparison results of the actual area, theoretical area of the vertebral body itself, and the theoretical area of adjacent vertebral bodies, etc., are used as the training sample data of the model to comprehensively and accurately present the characteristics of each vertebral body; thus, combined with the network model to train the pre-constructed training sample data to obtain a vertebral compression fracture analysis model, effectively ensuring the accuracy, efficiency, and reliability of the prediction result of whether the vertebral body in any target vertebral image has overall compression. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flowchart of the training method of the vertebral compression fracture analysis model according to Embodiment 1 of the present disclosure;

[0080] Figure 2 It is a flowchart of the training method of the vertebral compression fracture analysis model according to Embodiment 2 of the present disclosure;

[0081] Figure 3 It is a flowchart of obtaining the sample actual area of the sample vertebral body according to Embodiment 2 of the present disclosure;

[0082] Figure 4 It is the original schematic diagram of the vertebral image according to Embodiment 2 of the present disclosure;

[0083] Figure 5 Schematic diagram of the actual graph of the vertebral body image in Embodiment 2 of the present disclosure;

[0084] Figure 6 Schematic diagram of the posterior margin of the vertebral body image in Embodiment 2 of the present disclosure;

[0085] Figure 7 Schematic diagram showing bilateral sides at the upper and lower margins of the vertebral body image in Embodiment 2 of the present disclosure;

[0086] Figure 8 Schematic diagram of the midline at the upper and lower margins of the vertebral body image in Embodiment 2 of the present disclosure;

[0087] Figure 9 Schematic diagram of the anterior margin of the vertebral body image in Embodiment 2 of the present disclosure;

[0088] Figure 10 Flowchart for obtaining the sample theoretical area of the sample vertebral body in Embodiment 2 of the present disclosure;

[0089] Figure 11 Flowchart of the method for identifying vertebral body compression fractures in Embodiment 3 of the present disclosure;

[0090] Figure 12 Flowchart of the method for identifying vertebral body compression fractures in Embodiment 4 of the present disclosure;

[0091] Figure 13 Schematic diagram of the modules of the training system for the vertebral body compression fracture analysis model in Embodiment 5 of the present disclosure;

[0092] Figure 14 Schematic diagram of the modules of the training system for the vertebral body compression fracture analysis model in Embodiment 6 of the present disclosure;

[0093] Figure 15 Schematic diagram of the modules of the identification system for vertebral body compression fractures in Embodiment 7 of the present disclosure;

[0094] Figure 16 Schematic diagram of the modules of the identification system for vertebral body compression fractures in Embodiment 8 of the present disclosure;

[0095] Figure 17 Schematic diagram of the structure of the electronic device in Embodiment 9 of the present disclosure. Detailed implementation manners

[0096] The present disclosure will be further described below by way of embodiments, but the present disclosure is not limited to the scope of the described embodiments.

[0097] In the embodiments of the present disclosure, prefix words such as "first" and "second" are only used to distinguish different described objects, and have no restrictive effect on the position, order, priority, quantity, content, etc. of the described objects. In the embodiments of the present disclosure, the use of prefix words such as ordinal numbers for distinguishing described objects does not constitute a limitation on the described objects. For the statements of the described objects, refer to the description in the claims or the context of the embodiments. It should not constitute an unnecessary limitation due to the use of such prefix words. In addition, in the description of this embodiment, unless otherwise specified, the meaning of "a plurality" is two or more.

[0098] Embodiment 1

[0099] As Figure 1 shown, the training method of the vertebral compression fracture analysis model in this embodiment includes:

[0100] S101. Obtain a number of sample vertebral images;

[0101] Among them, the sample vertebral images can be lateral films of the thoracic and lumbar vertebrae generated based on an X-ray machine.

[0102] S102. Obtain the sample vertebral feature data corresponding to each sample vertebra in each sample vertebral image, and the sample annotation data indicating whether overall compression has occurred for each sample vertebral image;

[0103] Among them, the sample vertebrae are other vertebrae except the basic vertebra (i.e., the first vertebra), and the first vertebra is a vertebra without overall compression; the sample vertebral feature data includes the sample area correlation data of the sample vertebrae;

[0104] The sample area correlation data includes the sample theoretical area, the sample actual area, the difference in the sample theoretical area from the adjacent vertebra, the ratio of the sample theoretical area to the adjacent vertebra, the ratio of the sample actual area to the sample theoretical area, the convex hull area, the ratio of the convex hull area to the theoretical area, etc. Preferably, the difference in the sample theoretical area from the adjacent sample and the ratio of the sample theoretical area to the adjacent vertebra are the difference in the sample theoretical area and the ratio of the sample theoretical area with respect to the previous vertebra. Of course preferably, the sample vertebral feature data further includes the sample position data of the sample vertebrae, the sample length correlation data, etc.

[0105] Among them, the sample position data is the position of each vertebra in the spine, such as which vertebra; for example, the different vertebrae from top to bottom are: T4 - T12 - L1 - L4.

[0106] The sample length correlation data includes the length of the posterior edge of the vertebra, the ratio of the length of the posterior edge to the adjacent vertebra, the difference in the length of the posterior edge from the adjacent vertebra, etc.

[0107] In addition, for the sample annotation data corresponding to the vertebral body, a label of "0" indicates that there is no overall compression fracture in the vertebral body, and a label of "1" indicates that there is an overall compression fracture in the vertebral body.

[0108] S103. Based on the sample vertebral body feature data of each sample vertebral body corresponding to each sample vertebral body image and the corresponding sample annotation data, model training is performed to obtain a vertebral body compression fracture analysis model for outputting whether there is an overall compression in the vertebral body in any target vertebral body image.

[0109] For example, X-ray films of several typical patients are selected as sample vertebral body images. Each vertebral body (i.e., each vertebra) in these sample vertebral body images is segmented using image segmentation techniques (such as the yolo-xray image segmentation algorithm, etc.), and whether there is an overall compression in each vertebral body is annotated to obtain a label indicating whether there is an overall compression. In addition, features of each preset dimension of each vertebral body are obtained, including the position of the vertebral body in the spine (such as which vertebral body segment), area-related data features, and length-related data features of the posterior edge. These data are used as a sample data set, and then the sample data set is divided into an experimental group and a control group in a ratio of 8:2 for model training and verification.

[0110] Then, the sample vertebral body feature data of each vertebral body is used as the input, and the corresponding annotation data is used as the output to perform model training on a preset network to obtain a vertebral body compression fracture analysis model.

[0111] Among them, the preset network includes but is not limited to Logistic Regression, XGBoost, Random Forest, GBDT, Catboost, MLP (all are machine learning algorithms); of course, large models, etc. can also be used to implement it. There is no limit on which algorithm is specifically used for model training, as long as the obtained vertebral body compression fracture analysis model can accurately analyze whether there is an overall compression in the vertebral body in any input vertebral body image.

[0112] In the present disclosure, the sample data is considered and designed skillfully and reasonably. By obtaining the sample vertebral body feature data of each vertebral body in the sample vertebral body image, especially calculating the sample area-related data corresponding to the vertebral body, that is, the comparison results of the actual area, theoretical area of the vertebral body itself, and the theoretical area of adjacent vertebral bodies, etc., are used as the training sample data of the model to comprehensively and accurately present the characteristics of each vertebral body. In this way, combined with the network model, the pre-constructed training sample data is trained to obtain a vertebral body compression fracture analysis model, thereby effectively ensuring the accuracy, efficiency, and reliability of the prediction result of whether there is an overall compression in the vertebral body in any target vertebral body image output by the vertebral body compression fracture analysis model.

[0113] Embodiment 2

[0114] The training method of the vertebral compression fracture analysis model in this embodiment is a further improvement of Embodiment 1. Specifically:

[0115] In an implementable solution, as Figure 2 shown, after the step of obtaining the sample vertebral body feature data corresponding to each vertebral body and before model training, this training method further includes:

[0116] S1030. Perform feature screening processing on the sample vertebral body feature data corresponding to each vertebral body to obtain the sample vertebral body feature data after screening processing;

[0117] Specifically, the feature data related to the vertebral body includes the following:

[0118] cls: represents the position of the vertebral body, records the position of each vertebral body in the spine, such as which vertebral body; areaDiff: represents the theoretical area difference from the previous vertebral body; areaDiffRatio: represents the theoretical area ratio to the previous vertebral body; areaRatio: represents the ratio of the actual area to the theoretical area; boxArea: represents the theoretical area; convexArea: represents the convex hull area; convexRatio: represents the ratio of the convex hull area to the theoretical area; disBigendRatio: represents the length ratio to the posterior edge of the previous vertebral body; disBigendDiff: represents the length difference to the posterior edge of the previous vertebral body; countourArea: represents the actual area; disBigend: the length of the posterior edge of the vertebral body.

[0119] When performing feature screening, considering that Lasso regression has good interpretability and explainability, can automatically select important features related to the target variable, and at the same time set the coefficients of irrelevant or redundant features to zero. In this solution, Lasso regression is used for feature selection, and by introducing L1 regularization to promote the sparsity of some feature coefficients, that is, screening the obtained feature data of the vertebral body to screen out more reasonable, effective and reliable feature data for model training;

[0120] For example, the screening results are cls: represents the position of the vertebral body; areaDiff: represents the theoretical area difference from the previous vertebral body; areaDiffRatio: represents the theoretical area ratio to the previous vertebral body; boxArea: represents the theoretical area; convexArea: represents the convex hull area; disBigendRatio: represents the length ratio to the posterior edge of the previous vertebral body; disBigendDiff: represents the length difference to the posterior edge of the previous vertebral body; countourArea: represents the actual area; disBigend: the length of the posterior edge of the vertebral body.

[0121] In addition, the convex hull refers to the smallest convex polygon or convex polyhedron that contains all the points in a point set. In a two-dimensional plane, the convex hull is a simple polygon that encloses all the given points with the smallest perimeter. Any line segment connecting two points inside it lies entirely inside or on the boundary of the polygon, and the area of the convex hull is the area within this region.

[0122] When using the Andrew algorithm (also known as the monotone chain algorithm) to find the convex hull, the convex hull has the following characteristics: The leftmost and rightmost points must be on the convex hull. The convex hull is a convex polygon, that is, when walking counterclockwise from a point, the trajectory always turns left (cross product greater than 0). Once a right turn (cross product less than 0) or three points are collinear (cross product equal to 0) occurs, it means that this point is not on the convex hull. According to this characteristic, the Andrew convex hull algorithm divides the convex hull into two parts, namely the upper convex hull and the lower convex hull: The lower convex hull turns left from the leftmost point to the rightmost point, and the upper convex hull turns left from the rightmost point to the leftmost point. Specifically, sort all the points in descending order of the x value and then in descending order of the y value. Traverse all the points and find the points where the two connected sides turn left. All these points are the lower convex hull. Sort the points in reverse order, repeat the above steps, and the values found by turning left are the upper convex hull. Combine the upper and lower convex hulls and remove duplicate points.

[0123] In this solution, by further screening a preliminary set of sample vertebral body feature data, the accuracy and reliability of the input data during model training are ensured, so as to guarantee the reliability of the finally trained vertebral compression fracture analysis model. At the same time, the model input data is simplified, effectively reducing the amount of data processing during the model training stage, thereby effectively improving the training efficiency of the vertebral compression fracture analysis model.

[0124] In an implementable solution, as Figure 3 shown, the steps to calculate the sample true area of the sample vertebral body include:

[0125] S201. Adopt a preset edge processing rule to identify and process each edge in the sample vertebral body to obtain the true edge line corresponding to each edge;

[0126] Among them, obtain the edge association information of each edge of the vertebral body in the vertebral body image;

[0127] Among them, the edges of the vertebral body in the vertebral body image include the anterior edge, posterior edge, upper edge, and lower edge; the edge association information includes all information that can characterize each corresponding edge, including but not limited to the existence of multiple curves and curve position information.

[0128] Judge whether the edge association information meets the corresponding preset conditions. If it meets, execute the processing of the edge association information using the matching preset edge processing rule to generate the true edge line corresponding to each edge;

[0129] Among them, different edges correspond to different preset conditions and preset edge processing rules; that is, for the leading edge, trailing edge, upper edge, and lower edge, due to their different presented states, they each have their own characteristics, and the corresponding preset conditions are pre-constructed to automatically identify the actual situation of the corresponding edge, and then the matching preset edge processing rules are used for targeted processing to extract the true edge line corresponding to each edge.

[0130] S202. Generate the true edge graph corresponding to the vertebral body based on each true edge line;

[0131] According to the four true edge lines of the leading edge, trailing edge, upper edge, and lower edge, a closed quadrilateral is automatically formed to obtain the true edge graph corresponding to the vertebral body. As Figure 4 shown, it is the vertebral body image before processing; as Figure 5 shown, among them, the quadrilateral (black frame) is the true edge graph of the vertebral body.

[0132] S203. Obtain the area corresponding to the true edge graph and use it as the sample true area of the sample vertebral body;

[0133] Among them, the corresponding preset edge processing rules are used to process different edges of the vertebral body to automatically, quickly, and accurately extract each actual edge line of each vertebral body, ensuring the processing accuracy and efficiency of segmenting the actual shape of the vertebral body from the vertebral body image, and further ensuring the accuracy and efficiency of determining the true area of the vertebral body.

[0134] Furthermore, for the trailing edge of the vertebral body, the edge association information is the trailing edge information; among them, the trailing edge is the edge at the connection with the pedicle in the vertebral body.

[0135] The steps of judging whether the edge association information meets the corresponding preset conditions include:

[0136] Judge whether there is a ghosting area in the vertebral body image according to the trailing edge information. If so, execute the matching preset edge processing rule to process the edge association information to generate the true edge line corresponding to each edge;

[0137] Among them, whether there is a ghosting area is distinguished by judging whether there are multiple initial trailing edge curves. That is, once there are ≥2 initial trailing edge curves, there will be a ghosting area, that is, the actual position of the trailing edge cannot be directly distinguished.

[0138] The steps of using the matching preset edge processing rule to process the edge association information to generate the true edge line corresponding to each edge include:

[0139] Obtain multiple initial trailing edge curves;

[0140] Identify the pedicle region in the vertebral body image;

[0141] Use image recognition technology to analyze the vertebral body image to obtain the pedicle region and the region where the vertebral body is located.

[0142] Obtain an initial posterior edge curve that is the farthest from the pedicle region as the target posterior edge curve;

[0143] Specifically, each irregular initial posterior edge curve can be unified into a line segment shape, as Figure 6 shown, calculate the distances between different line segments and the pedicle region to obtain an initial posterior edge curve that is the farthest from the pedicle region.

[0144] Obtain the first upper endpoint and the first lower endpoint corresponding to the target posterior edge curve;

[0145] Among them, the intersection point of the target posterior edge curve and the innermost curve at the upper edge of the vertebral body is the first upper endpoint; the intersection point of the target posterior edge curve and the innermost curve at the lower edge of the vertebral body is the first lower endpoint; that is, the target posterior edge curve only intersects with other curves at the endpoints, and there is no intersection with other curves at other places.

[0146] Form a first line segment by connecting the first upper endpoint and the first lower endpoint;

[0147] Based on the first line segment, obtain the true posterior edge line of the vertebral body.

[0148] Furthermore, for the upper edge of the vertebral body, the edge association information is the upper edge information.

[0149] The steps of judging whether the edge association information meets the corresponding preset conditions include:

[0150] Judge whether there are multiple initial upper edge curves according to the upper edge information. If so, execute the step of processing the edge association information using the matching preset edge processing rule to generate the true edge line corresponding to each edge;

[0151] The steps of processing the edge association information using the matching preset edge processing rule to generate the true edge line corresponding to each edge include:

[0152] Obtain a first target curve within the first preset position range between the two initial upper edge curves located on the outermost and innermost sides, and obtain the true upper edge line based on the obtained first target curve.

[0153] Among them, the first target curve can be obtained by direct selection or automatic generation, etc.;

[0154] Furthermore, the steps of judging whether the edge association information meets the corresponding preset conditions include:

[0155] Judge whether there are multiple initial upper edge curves according to the upper edge information. If not, perform processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge;

[0156] The steps of performing processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge include:

[0157] If there is only one initial upper edge curve, obtain the true upper edge line based on the initial upper edge curve;

[0158] Furthermore, for the lower edge of the vertebral body, the edge association information is the lower edge information.

[0159] The steps of judging whether the edge association information meets the corresponding preset conditions include:

[0160] Judge whether there are multiple initial lower edge curves according to the lower edge information. If so, perform processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge;

[0161] The steps of performing processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge include:

[0162] Obtain a second target curve within the range of the second preset position between the two initial lower edge curves located at the innermost and outermost sides, and obtain the true lower edge line based on the obtained second target curve;

[0163] Among them, the second target curve can be obtained by direct selection or automatic generation, etc.;

[0164] Furthermore, the steps of judging whether the edge association information meets the corresponding preset conditions include:

[0165] When it is determined according to the lower edge information that there are no multiple initial lower edge curves, perform processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge;

[0166] The steps of performing processing on the edge association information using the matching preset edge processing rules to generate the true edge line corresponding to each edge include: If there is only one initial lower edge curve, obtain the true lower edge line based on the initial lower edge curve.

[0167] Furthermore, if there are two initial upper edge curves in the vertebral body image, form a first target curve within the position range from one-third to two-thirds between the two initial upper edge curves, and obtain the true upper edge line based on the formed first target curve;

[0168] Among them, the trend of the first target curve is between the trends of the two initial upper edge curves.

[0169] As Figure 7 shown, to determine whether there is a bilateral edge at the upper edge. When there are two initial upper edge curves (curves 1 and 2 marked in the figure), an arbitrary curve is generated between 1 / 3 and 2 / 3 of the two initial lower edge curves as the first target curve, and the true upper edge line of the vertebral body is finally determined based on this first target curve, avoiding determining the true upper edge line based on the curve located more outside or more inside, thus preventing the situation where the recognition accuracy of the true upper edge line cannot be guaranteed, and effectively ensuring the accuracy of obtaining the true upper edge line.

[0170] Furthermore, when two initial lower edge curves are included in the vertebral body image, a second target curve is formed within the position range of one-third to two-thirds between the two initial lower edge curves, and the true lower edge line is obtained based on the formed second target curve;

[0171] Among them, the trend of the second target curve is between the trends of the two initial lower edge curves.

[0172] As Figure 7 shown, to determine whether there is a bilateral edge at the lower edge. When there are two initial lower edge curves (curves 3 and 4 marked in the figure), an arbitrary curve is generated between 1 / 3 and 2 / 3 of the two initial lower edge curves as the second target curve, and the true lower edge line of the vertebral body is finally determined based on this second target curve, avoiding determining the true lower edge line based on the curve located more outside or more inside, thus preventing the situation where the recognition accuracy of the true lower edge line cannot be guaranteed, and effectively ensuring the accuracy of obtaining the true lower edge line.

[0173] Furthermore, the first target curve is the midline between the two initial upper edge curves;

[0174] As Figure 8 shown, the midline between the two initial upper edge curves (the curve in the middle of curves 1 and 2 marked in the figure) is used as the first target curve, and the overall undulating trend of the first target curve is between the two initial upper edge curves.

[0175] Furthermore, the second target curve is the midline between the two initial lower edge curves.

[0176] As Figure 8 shown, the midline between the two initial lower edge curves (the curve in the middle of curves 3 and 4 marked in the figure) is used as the second target curve, and the overall undulating trend of the second target curve is between the two initial lower edge curves.

[0177] Further, obtain the first intersection point between the straight line where the first line segment is located and the first target curve, and the second intersection point with the second target curve;

[0178] Take the second line segment between the first intersection point and the second intersection point as the true trailing edge line of the vertebral body;

[0179] Among them, the first intersection point is the first left end point of the true upper edge line, and the second intersection point is the second left end point of the true lower edge line.

[0180] Among them, based on the first line segment, the curve where the true upper edge line is located, and the curve where the true lower edge line is located, the intersection points between the straight line where the first line segment is located and the first target curve and the second target curve can be automatically obtained, so as to finally determine the two end points of the true trailing edge line, thus ensuring the rationality and accuracy of obtaining the true trailing edge line.

[0181] Further, for the leading edge of the vertebral body, the edge association information is the leading edge information;

[0182] The steps of judging whether the edge association information meets the corresponding preset conditions include:

[0183] Judge whether there is a bend on the initial leading edge curve with an inner concave curvature greater than the first set threshold. If so, execute the step of processing the edge association information with the matching preset edge processing rule to generate the true edge line corresponding to each edge;

[0184] The steps of processing the edge association information with the matching preset edge processing rule to generate the true edge line corresponding to each edge include:

[0185] Determine that there is osteophyte hyperplasia on the leading edge of the vertebral body;

[0186] For example, if there is a bend on the curve with an inner concave curvature greater than 15°, it is determined that there is osteophyte hyperplasia at this bend.

[0187] Based on the vertex of the bend, perform tangent processing with a preset slope to obtain the initial straight line;

[0188] Among them, as Figure 9 shown in the initial straight line, the preset slope is determined based on the first slope of the first line segment;

[0189] The initial straight line obtained thereby is used to cut off the part with osteophyte hyperplasia at the leading edge of the vertebral body to avoid the influence of osteophyte hyperplasia on the determination of the edge of the vertebral body.

[0190] Obtain the third intersection point between the initial straight line and the first target curve, and the fourth intersection point with the second target curve;

[0191] Take the third line segment between the third intersection point and the fourth intersection point as the true leading edge line of the vertebral body;

[0192] Among them, the third intersection point is the first right endpoint of the true upper edge line, and the fourth intersection point is the second right endpoint of the true lower edge line.

[0193] Furthermore, the difference between the preset slope and the first slope is less than a second set threshold; wherein, the second set threshold is a non-zero number;

[0194] Or, the preset slope is equal to the first slope.

[0195] Furthermore, the steps of determining whether the edge association information meets the corresponding preset conditions include:

[0196] When there is no bend on the initial leading edge curve with an inflection curvature greater than the first set threshold, then perform processing on the edge association information using a matching preset edge processing rule to generate the true edge line corresponding to each edge;

[0197] The steps of processing the edge association information using a matching preset edge processing rule to generate the true edge line corresponding to each edge include:

[0198] Determine that there is no bone hyperplasia on the leading edge of the vertebral body, and use the initial leading edge curve as the target leading edge curve;

[0199] Obtain the fifth intersection point of the target leading edge curve with the true upper edge line and the sixth intersection point with the true lower edge line;

[0200] Use the connecting curve between the fifth intersection point and the sixth intersection point as the true leading edge line of the vertebral body;

[0201] Among them, the fifth intersection point is the first right endpoint of the true upper edge line, and the sixth intersection point is the second right endpoint of the true lower edge line;

[0202] Furthermore, the steps of processing the edge association information using a matching preset edge processing rule to generate the true edge line corresponding to each edge include:

[0203] Obtain the second upper endpoint and the second lower endpoint corresponding to the target leading edge curve;

[0204] Use the fourth line segment formed by connecting the second upper endpoint and the second lower endpoint;

[0205] Obtain the seventh intersection point of the straight line where the fourth line segment is located with the true upper edge line and the eighth intersection point with the true lower edge line;

[0206] Use the fifth line segment between the seventh intersection point and the eighth intersection point as the true leading edge line of the vertebral body;

[0207] Among them, the seventh intersection point is the first right endpoint of the true upper edge line, and the eighth intersection point is the second right endpoint of the true lower edge line.

[0208] The steps of obtaining the true upper edge line based on the acquired first target curve include:

[0209] Taking the curve between the first left endpoint and the first right endpoint as the true upper edge line;

[0210] Further, the steps of obtaining the true lower edge line based on the acquired second target curve include:

[0211] Taking the curve between the second left endpoint and the second right endpoint as the true lower edge line.

[0212] Further, based on the true trailing edge line, true leading edge line, true upper edge line and true lower edge line, a closed quadrilateral is formed, and the quadrilateral is used as the true edge graph of the vertebral body.

[0213] In an implementable solution, as Figure 10 shown, the steps of calculating the sample theoretical area of the sample vertebral body include:

[0214] S301. Adopt a preset edge processing rule to identify and process each edge in the sample vertebral body to obtain the true edge line corresponding to each edge;

[0215] Among them, the true edge line includes the true leading edge line, true trailing edge line, true upper edge line and true lower edge line;

[0216] S302. Based on each true edge line, form a closed graph and use it as the true edge graph corresponding to the vertebral body;

[0217] S303. According to the true leading edge line, true trailing edge line and true edge graph, generate the circumscribed rectangle of the true edge graph, and use the circumscribed rectangle as the theoretical shape of the sample vertebral body;

[0218] S304. Obtain the area corresponding to the theoretical shape and use it as the sample theoretical area of the sample vertebral body.

[0219] Among them, different edges of the vertebral body are processed using corresponding preset edge processing rules to automatically, quickly and accurately extract each actual edge line of each vertebral body, ensuring the processing accuracy and efficiency of segmenting the actual shape of the vertebral body from the vertebral body image; furthermore, a corresponding circumscribed rectangle is formed with the true leading edge line, true trailing edge line, and the quadrilateral closed by the four true edge lines, and this is used as the theoretical shape corresponding to the vertebral body in the vertebral body image, thus ensuring the accuracy and rationality of determining the theoretical shape of the vertebral body, and further ensuring the precision of determining the theoretical area.

[0220] Further, step S303 includes:

[0221] Select the longer line segment among the true leading edge line and the true trailing edge line as the width side length of the circumscribed rectangle;

[0222] Among them, the true leading edge line and the true trailing edge line are parallel to each other;

[0223] Take the distance between the true leading edge line and the true trailing edge line as the length side of the circumscribed rectangle;

[0224] According to the width side and the length side, generate the circumscribed rectangle of the true edge figure, and take the circumscribed rectangle as the theoretical shape of the vertebral body.

[0225] The specific implementation processes in the other steps S301 and S302 are the same as those for calculating the sample true area of the vertebral body above, so they will not be elaborated here.

[0226] Embodiment 3

[0227] As Figure 11 shown, the method for identifying the vertebral body compression fracture situation in this embodiment includes:

[0228] S401. Obtain the target vertebral body image;

[0229] S402. Obtain the target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image;

[0230] Among them, the target vertebral body is other vertebral bodies except the basic vertebral body (that is, the first vertebral body), and the first vertebral body is a non-integrally compressed vertebral body; the target vertebral body feature data includes the target area correlation data of the target vertebral body;

[0231] S403. Input the target area correlation data into the vertebral body compression fracture analysis model to obtain the target recognition result of whether the vertebral body in the target vertebral body image has undergone integral compression;

[0232] Among them, the vertebral body compression fracture analysis model is obtained based on the training method of the vertebral body compression fracture analysis model as described above.

[0233] In this embodiment, the vertebral body compression fracture analysis model is used to analyze and process the input target vertebral body image to predict whether the vertebral body therein has undergone integral compression, ensuring the accuracy and efficiency of identifying the compression fracture situation in any target vertebral body image.

[0234] Embodiment 4

[0235] The method for identifying the vertebral body compression fracture situation in this embodiment is a further improvement of Embodiment 3. Specifically:

[0236] In an implementable solution, as Figure 12 shown, after step S403, it further includes:

[0237] S4041. In response to overall compression occurring in the current target vertebral body, determine the fracture degree of the target vertebral body using a preset correction scheme.

[0238] In this scheme, the vertebral body compression fracture analysis model is used to predict whether overall compression has occurred in the vertebral body. Once it is predicted that overall compression has occurred in a certain vertebral body, the default scheme for determining the fracture degree of the vertebral body is no longer used, and the correction scheme needs to be called in a timely manner to perform timely deviation correction on this vertebral body to ensure the accuracy of determining the fracture degree of the target vertebral body in a timely manner.

[0239] In an implementable scheme, step S4041 includes:

[0240] In response to the target recognition result indicating overall compression, obtain the first theoretical area of the vertebral body immediately preceding the current target vertebral body and the second theoretical area of the vertebral body immediately following the current target vertebral body;

[0241] Take the weighted average of the first theoretical area and the second theoretical area to obtain the third theoretical area as the new theoretical area of the current target vertebral body;

[0242] Obtain the target true area of the current target vertebral body;

[0243] Calculate the first ratio of the target true area and the new theoretical area of the current target vertebral body;

[0244] Determine the fracture degree of the current target vertebral body based on the first ratio;

[0245] Among them, different fracture degrees correspond to different preset ratio ranges;

[0246] First ratio = target true area of the vertebral body / new theoretical area of the vertebral body; different fracture degrees correspond to different preset ratio ranges: for r = 0 degree: [0, 10]; for r = 1 degree: (10, 20]; for r = 2 degree: (20, 40]; for r = 3 degree: (40, 100];

[0247] In this scheme, the value obtained by weighted averaging the theoretical areas of the adjacent vertebral bodies before and after is used as the theoretical area of the current target vertebral body, rather than using the theoretical area directly calculated for the current target vertebral body for subsequent fracture degree calculation, to ensure the accuracy of determining the fracture degree of the vertebral body with overall compression.

[0248] In an implementable scheme, step S4041 includes:

[0249] In response to overall compression occurring in the current target vertebral body, obtain the target true area and the target theoretical area of the current target vertebral body;

[0250] Calculate the second ratio of the target true area and the target theoretical area of the current target vertebral body, and multiply the second ratio by a preset coefficient to obtain a third ratio; wherein, the preset coefficient is determined according to experience and can also be re-determined or adjusted according to actual needs.

[0251] Determine the fracture classification of the current target vertebral body based on the third ratio;

[0252] Among them, different fracture classifications correspond to different preset ratio ranges.

[0253] The third ratio = the target true area of the vertebral body / the new theoretical area of the vertebral body; different fracture classifications correspond to different preset ratio ranges: when r is 0 degree: [0, 10]; when r is 1 degree: (10, 20]; when r is 2 degree: (20, 40]; when r is 3 degree: (40, 100];

[0254] Specifically, the third ratio = the true area of the vertebral body / (a * the theoretical area of the vertebral body), where a is the preset coefficient; the range of a is [1.19 - 1.3]. The fracture classification after correction can highly coincide with the gold standard classification result x of the vertebra with overall compression marked by experts.

[0255] In this solution, the theoretical area of the vertebral body with overall compression is directly corrected through a preset system, rather than using the theoretical area directly calculated for the current target vertebral body for subsequent fracture classification calculation, ensuring the accuracy of determining the fracture classification of the vertebral body with overall compression, simplifying the intermediate data processing process, and further improving the overall data processing efficiency.

[0256] In an implementable solution, the recognition method further includes:

[0257] In response to the target recognition result indicating no overall compression, obtain the target true area and the target theoretical area corresponding to the current target vertebral body;

[0258] Calculate the fourth ratio of the target true area and the target theoretical area corresponding to each target vertebral body;

[0259] Determine the fracture classification of the current target vertebral body based on the fourth ratio;

[0260] Among them, different fracture classifications correspond to different preset ratio ranges;

[0261] The fourth ratio = the target true area of the vertebral body / the target theoretical area of the vertebral body; different fracture classifications correspond to different preset ratio ranges: when r is 0 degree: [0, 10]; when r is 1 degree: (10, 20]; when r is 2 degree: (20, 40]; when r is 3 degree: (40, 100];

[0262] In this solution, a vertebral body compression fracture analysis model is used to predict whether overall compression occurs in the vertebral body. Once it is predicted that the vertebral body does not undergo overall compression, the default scheme for determining the fracture degree of the vertebral body is adopted for calculation and processing to ensure the efficiency and accuracy of determining the fracture degree of the target vertebral body in a timely manner.

[0263] In an implementable solution, the steps for calculating the target true area of the target vertebral body include:

[0264] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body to obtain the true edge line corresponding to each edge;

[0265] Based on each true edge line, generate the true edge graph corresponding to the vertebral body;

[0266] Obtain the area corresponding to the true edge graph and use it as the target true area of the target vertebral body;

[0267] In an implementable solution, the steps for calculating the target theoretical area of the target vertebral body include:

[0268] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body to obtain the true edge line corresponding to each edge;

[0269] Among them, the true edge line includes the true anterior edge line, the true posterior edge line, the true upper edge line, and the true lower edge line;

[0270] Based on each true edge line, form a closed graph and use it as the true edge graph corresponding to the vertebral body;

[0271] According to the true anterior edge line, the true posterior edge line, and the true edge graph, generate the circumscribed rectangle of the true edge graph and use the circumscribed rectangle as the theoretical shape of the target vertebral body;

[0272] Obtain the area corresponding to the theoretical shape and use it as the target theoretical area of the target vertebral body.

[0273] It should be noted that the process of calculating the target true area of the target vertebral body and the process of calculating the sample true area of the sample vertebral body, as well as the process of calculating the target theoretical area of the target vertebral body and the process of calculating the sample theoretical area of the sample vertebral body, are similar, so they will not be elaborated here.

[0274] Example 5

[0275] As Figure 13 shown, the training system of the vertebral body compression fracture analysis model in this embodiment includes:

[0276] A sample image acquisition module 1 for acquiring a number of sample vertebral body images;

[0277] A sample training data acquisition module 2, configured to obtain sample vertebral feature data corresponding to each sample vertebra in each sample vertebral image, and sample annotation data indicating whether overall compression has occurred in each sample vertebral image; wherein, the sample vertebral feature data includes sample area correlation data of the sample vertebra.

[0278] A model training module 3, configured to perform model training based on the sample vertebral feature data corresponding to each sample vertebra in each sample vertebral image and the corresponding sample annotation data, so as to obtain a vertebral compression fracture analysis model for outputting whether overall compression has occurred in the vertebra in any target vertebral image.

[0279] In the present disclosure, the sample data is considered and designed skillfully and reasonably. By obtaining the sample vertebral feature data of each vertebra in the sample vertebral image, especially calculating the sample area correlation data corresponding to the vertebra, that is, the true area of the vertebra itself, the theoretical area, and the comparison result with the theoretical area of the adjacent vertebra, etc., as the training sample data of the model, so as to comprehensively and accurately present the characteristics of each vertebra; thus, combining the network model to train the pre-constructed training sample data to obtain a vertebral compression fracture analysis model, thereby effectively ensuring the accuracy, efficiency, and reliability of the prediction result of whether overall compression has occurred in the vertebra in any target vertebral image output by the vertebral compression fracture analysis model.

[0280] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, where the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution.

[0281] Embodiment 6

[0282] The training system of the vertebral compression fracture analysis model in this embodiment is a further improvement of Embodiment 5. Specifically:

[0283] In an implementable solution, as Figure 14 shown, the training system further includes:

[0284] A feature screening module 4, configured to perform feature screening processing on the sample vertebral feature data corresponding to each vertebra to obtain the sample vertebral feature data after screening processing.

[0285] In this solution, by further screening a set of initially determined sample vertebral body feature data, the accuracy and reliability of the input data during model training are ensured, so as to ensure the reliability of the vertebral compression fracture analysis model obtained through final training. At the same time, the input data of the model is simplified, effectively reducing the data processing volume during the model training stage, thereby effectively improving the training efficiency of the vertebral compression fracture analysis model. In

[0286] In an implementable solution, such as Figure 3 shown, the steps for calculating the sample true area of the sample vertebral body include:

[0287] Adopt a preset edge processing rule to identify and process each edge in the sample vertebral body to obtain the true edge line corresponding to each edge;

[0288] Based on each true edge line, generate the true edge graph corresponding to the vertebral body;

[0289] Obtain the area corresponding to the true edge graph and use it as the sample true area of the sample vertebral body;

[0290] In an implementable solution, the steps for calculating the sample theoretical area of the sample vertebral body include:

[0291] Adopt a preset edge processing rule to identify and process each edge in the sample vertebral body to obtain the true edge line corresponding to each edge;

[0292] Among them, the true edge line includes the true front edge line, the true rear edge line, the true upper edge line, and the true lower edge line;

[0293] Based on each true edge line, form a closed graph and use it as the true edge graph corresponding to the vertebral body;

[0294] According to the true front edge line, the true rear edge line, and the true edge graph, generate the circumscribed rectangle of the true edge graph and use the circumscribed rectangle as the theoretical shape of the sample vertebral body;

[0295] Obtain the area corresponding to the theoretical shape and use it as the sample theoretical area of the sample vertebral body.

[0296] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present disclosure.

[0297] Embodiment 7

[0298] As Figure 15 shown, the recognition system for vertebral compression fracture conditions in this embodiment includes:

[0299] A target image acquisition module 5, configured to acquire a target vertebral body image;

[0300] A target feature acquisition module 6, configured to acquire target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image;

[0301] Among them, the target vertebral body feature data includes target area correlation data of the target vertebral body;

[0302] A target recognition result acquisition module 7, configured to input the target area correlation data into a vertebral compression fracture analysis model to obtain a target recognition result on whether the vertebral body in the target vertebral body image has undergone overall compression;

[0303] Among them, the vertebral compression fracture analysis model is obtained based on the training system of the vertebral compression fracture analysis model in Embodiment 5 or 6.

[0304] In this embodiment, the vertebral compression fracture analysis model is used to analyze and process the input target vertebral body image to predict whether the vertebral body therein has undergone overall compression, ensuring the accuracy and efficiency of the recognition of compression fracture conditions in any target vertebral body image.

[0305] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components as units may or may not be physical units, that is, they may be located in one place or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution.

[0306] Embodiment 8

[0307] As Figure 16 shown, the recognition system for vertebral compression fracture conditions in this embodiment is a further improvement of Embodiment 7. Specifically:

[0308] In an implementable solution, the recognition system further includes:

[0309] A fracture grading determination module 8, configured to, in response to the current target vertebral body having undergone overall compression, determine the fracture grading of the target vertebral body by using a preset correction scheme.

[0310] In an implementable solution, the fracture grading determination module 8 is configured to:

[0311] In response to the occurrence of overall compression in the target recognition result, obtain the first theoretical area of the vertebra immediately preceding the current target vertebra and the second theoretical area of the vertebra immediately following the current target vertebra;

[0312] Take the third theoretical area obtained by the weighted average of the first theoretical area and the second theoretical area as the new theoretical area of the current target vertebra;

[0313] Obtain the target true area of the current target vertebra;

[0314] Calculate the first ratio of the target true area and the new theoretical area of the current target vertebra;

[0315] Determine the fracture grade of the current target vertebra based on the first ratio;

[0316] Among them, different fracture grades correspond to different preset ratio ranges;

[0317] In an implementable solution, the fracture grade determination module 8 is used for:

[0318] In response to the occurrence of overall compression in the current target vertebra, obtain the target true area and the target theoretical area of the current target vertebra;

[0319] Calculate the second ratio of the target true area and the target theoretical area of the current target vertebra, and multiply the second ratio by a preset coefficient to obtain a third ratio; where the preset coefficient is determined according to experience and can also be re-determined or adjusted according to actual needs.

[0320] Determine the fracture grade of the current target vertebra based on the third ratio;

[0321] Among them, different fracture grades correspond to different preset ratio ranges.

[0322] In an implementable solution, the fracture grade determination module 8 is used for:

[0323] In response to the target recognition result indicating no overall compression, obtain the target true area and the target theoretical area corresponding to the current target vertebra;

[0324] Calculate the fourth ratio of the target true area and the target theoretical area corresponding to each target vertebra;

[0325] Determine the fracture grade of the current target vertebra based on the fourth ratio;

[0326] Among them, different fracture grades correspond to different preset ratio ranges;

[0327] In an implementable solution, the step of calculating the target true area of the target vertebra includes:

[0328] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body, so as to obtain the true edge line corresponding to each edge;

[0329] Generate a true edge graph corresponding to the vertebral body based on each true edge line;

[0330] Obtain the area corresponding to the true edge graph and use it as the target true area of the target vertebral body;

[0331] In an implementable solution, the steps of calculating the target theoretical area of the target vertebral body include:

[0332] Adopt a preset edge processing rule to identify and process each edge in the target vertebral body, so as to obtain the true edge line corresponding to each edge;

[0333] Among them, the true edge line includes a true leading edge line, a true trailing edge line, a true upper edge line, and a true lower edge line;

[0334] Based on each true edge line, form a closed graph and use it as the true edge graph corresponding to the vertebral body;

[0335] Generate a circumscribed rectangle of the true edge graph according to the true leading edge line, the true trailing edge line, and the true edge graph, and use the circumscribed rectangle as the theoretical shape of the target vertebral body;

[0336] Obtain the area corresponding to the theoretical shape and use it as the target theoretical area of the target vertebral body.

[0337] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiments described above are merely illustrative, where the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution.

[0338] Embodiment 9

[0339] Figure 17 It is a schematic structural diagram of an electronic device shown in an exemplary embodiment of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored on the memory and configured to run on the processor. When the processor executes the computer program, the method described in any of the above embodiments is implemented. Figure 17 The displayed electronic device 90 is only an example and should not impose any restrictions on the functions and usage scope of the embodiments of the present disclosure.

[0340] As Figure 17As shown, the electronic device 90 may be presented in the form of a general-purpose computing device. For example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one of the above-mentioned processors 91, at least one of the above-mentioned memories 92, and a bus 93 that connects different system components (including the memory 92 and the processor 91).

[0341] The bus 93 includes a data bus, an address bus, and a control bus.

[0342] 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.

[0343] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924. Such program modules 924 include, but are 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 the implementation of a network environment.

[0344] The processor 91 executes various functional applications and data processing by running computer programs stored in the memory 92, such as the methods provided in any of the above embodiments.

[0345] The electronic device 90 may also communicate with one or more external devices 94 (such as a keyboard, a pointing device, etc.). Such communication may be carried out through an input / output (I / O) interface 95. And, the electronic device 90 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 96. As shown in the figure, the network adapter 96 communicates with other modules of the electronic device 90 through the bus 93. It should be understood that although not shown in the figure, other hardware and / or software modules may be used in combination with the 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, etc.

[0346] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, such a division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-mentioned units / modules may be embodied in one unit / module. Conversely, the features and functions of one unit / module described above may be further divided and embodied by multiple units / modules.

[0347] Embodiment 10

[0348] Embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the method provided in any of the above embodiments is implemented.

[0349] Among them, the readable storage medium may more specifically include but is not limited to: portable disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0350] Embodiment 11

[0351] Embodiments of the present disclosure also provide a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any of the above is implemented.

[0352] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on the user device, partially on the user device, executed as an independent software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0353] Although the specific implementation manners of the present disclosure have been described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Without departing from the principles and essence of the present disclosure, those skilled in the art can make various changes or modifications to these implementation manners, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A training method for a vertebral compression fracture analysis model, characterized in that The training method includes: Obtaining a plurality of sample vertebral body images; Obtaining sample vertebral body feature data corresponding to each sample vertebral body in each of the sample vertebral body images, and sample annotation data indicating whether overall compression has occurred for each of the sample vertebral body images; wherein, the sample vertebral body feature data includes sample area correlation data of the sample vertebral body; Based on the sample vertebral body feature data of each sample vertebral body corresponding to each of the sample vertebral body images and the corresponding sample annotation data, performing model training to obtain the vertebral body compression fracture analysis model for outputting whether the vertebral body in any target vertebral body image has undergone overall compression.

2. The training method of the vertebral compression fracture analysis model according to claim 1, characterized in that The sample area correlation data includes at least one of sample theoretical area, sample actual area, difference in sample theoretical area from adjacent vertebral bodies, ratio of sample theoretical area to adjacent vertebral bodies, ratio of the sample actual area to the sample theoretical area, convex hull area, and ratio of the convex hull area to the theoretical area; and / or The sample vertebral body feature data further includes sample position data and / or sample length correlation data of the sample vertebral body; wherein, the sample length correlation data includes at least one of the length of the posterior edge of the vertebral body, ratio of the length to the posterior edge of the adjacent vertebral body, and difference in the length from the posterior edge of the adjacent vertebral body.

3. The training method of the vertebral compression fracture analysis model according to claim 2, characterized in that After the step of obtaining the sample vertebral body feature data corresponding to each vertebral body and before model training, the training method further includes: Performing feature screening processing on the sample vertebral body feature data corresponding to each vertebral body to obtain the screened sample vertebral body feature data; and / or, the step of calculating the sample actual area of the sample vertebral body includes: Using a preset edge processing rule to identify each edge in the sample vertebral body to obtain a true edge line corresponding to each edge; Based on each true edge line, generating a true edge graph corresponding to the vertebral body; Obtaining the area corresponding to the true edge graph and using it as the sample actual area of the sample vertebral body; and / or The step of calculating the sample theoretical area of the sample vertebral body includes: Using a preset edge processing rule to identify each edge in the sample vertebral body to obtain a true edge line corresponding to each edge; wherein, the true edge line includes a true anterior edge line, a true posterior edge line, a true upper edge line, and a true lower edge line; Based on each true edge line, forming a closed graph and using it as the true edge graph corresponding to the vertebral body; Generating a circumscribed rectangle of the true edge graph according to the true anterior edge line, the true posterior edge line, and the true edge graph, and using the circumscribed rectangle as the theoretical shape of the sample vertebral body; Obtaining the area corresponding to the theoretical shape and using it as the sample theoretical area of the sample vertebral body.

4. A method for identifying a vertebral compression fracture condition, characterized in that, The recognition method includes: Obtaining a target vertebral body image; Obtaining target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image; wherein, the target vertebral body feature data includes target area correlation data of the target vertebral body; Input the target area correlation data into the vertebral compression fracture analysis model to obtain a target recognition result on whether the vertebra in the target vertebral image has undergone overall compression; Among them, the vertebral compression fracture analysis model is obtained based on the training method of the vertebral compression fracture analysis model according to any one of claims 1 to 3.

5. The method for identifying the condition of vertebral compression fracture according to claim 4, wherein In response to the current overall compression of the target vertebra, use a preset correction scheme to determine the fracture degree of the target vertebra.

6. The method for identifying the condition of vertebral compression fracture according to claim 5, wherein, The step of, in response to the current overall compression of the target vertebra, using a preset correction scheme to determine the fracture degree of the target vertebra includes: In response to the target recognition result indicating overall compression, obtain the first theoretical area of the previous vertebra of the current target vertebra and the second theoretical area of the next vertebra. Use the weighted average of the first theoretical area and the second theoretical area as the new theoretical area of the current target vertebra. Obtain the target real area of the current target vertebra. Calculate the first ratio of the target real area and the new theoretical area of the current target vertebra. Determine the fracture degree of the current target vertebra based on the first ratio. Among them, different fracture degrees correspond to different preset ratio ranges. Or, The step of, in response to the current overall compression of the target vertebra, using a preset correction scheme to determine the fracture degree of the target vertebra includes: In response to the current overall compression of the target vertebra, obtain the target real area and the target theoretical area of the current target vertebra. Calculate the second ratio of the target real area and the target theoretical area of the current target vertebra, and multiply the second ratio by a preset coefficient to obtain a third ratio. Determine the fracture degree of the current target vertebra based on the third ratio. Among them, different fracture degrees correspond to different preset ratio ranges.

7. The method for identifying the condition of vertebral compression fracture according to claim 6, characterized in that, The recognition method further includes: In response to the target recognition result indicating no overall compression, obtain the target real area and the target theoretical area corresponding to the current target vertebra. Calculate the fourth ratio of the target real area and the target theoretical area corresponding to each target vertebra. Determine the fracture degree of the current target vertebra based on the fourth ratio. Among them, different fracture degrees correspond to different preset ratio ranges. And / or, The step of calculating the target real area of the target vertebra includes: Adopt a preset edge processing rule to identify each edge in the target vertebra to obtain a real edge line corresponding to each edge. Generate a real edge graph corresponding to the vertebra based on each real edge line. Obtain the area corresponding to the real edge graph and use it as the target real area of the target vertebra. And / or, The step of calculating the target theoretical area of the target vertebra includes: Adopt a preset edge processing rule to identify each edge in the target vertebra to obtain a real edge line corresponding to each edge. Among them, the real edge lines include a real leading edge line, a real trailing edge line, a real upper edge line, and a real lower edge line; Based on each of the real edge lines, a closed figure is formed and used as the real edge figure corresponding to the vertebral body; According to the real leading edge line, the real trailing edge line, and the real edge figure, a circumscribed rectangle of the real edge figure is generated, and the circumscribed rectangle is used as the theoretical shape of the target vertebral body; The area corresponding to the theoretical shape is obtained and used as the target theoretical area of the target vertebral body.

8. A training system for a vertebral compression fracture analysis model, characterized in that, The training system includes: A sample image acquisition module, configured to acquire a plurality of sample vertebral body images; A sample training data acquisition module, configured to acquire sample vertebral body feature data corresponding to each sample vertebral body in each sample vertebral body image, and sample annotation data indicating whether overall compression has occurred in each sample vertebral body image; among them, the sample vertebral body feature data includes sample area correlation data of the sample vertebral body; A model training module, configured to perform model training based on the sample vertebral body feature data of each sample vertebral body corresponding to each sample vertebral body image and the corresponding sample annotation data, so as to obtain a vertebral body compression fracture analysis model for outputting whether the vertebral body in any target vertebral body image has undergone overall compression.

9. An identification system for vertebral compression fracture conditions, characterized in that, The recognition system includes: A target image acquisition module, configured to acquire a target vertebral body image; A target feature acquisition module, configured to acquire target vertebral body feature data corresponding to each target vertebral body in the target vertebral body image; Among them, the target vertebral body feature data includes target area correlation data of the target vertebral body; A target recognition result acquisition module, configured to input the target area correlation data into the vertebral body compression fracture analysis model to obtain a target recognition result indicating whether the vertebral body in the target vertebral body image has undergone overall compression; Among them, the vertebral body compression fracture analysis model is obtained based on the training system of the vertebral body compression fracture analysis model as claimed in claim 8.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and configured to run on the processor, characterized in that, When the processor executes the computer program, it implements the training method of the vertebral body compression fracture analysis model as claimed in any one of claims 1 to 3; or, the recognition method of the vertebral body compression fracture situation as claimed in any one of claims 4 to 7.

11. 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 training method of the vertebral body compression fracture analysis model as claimed in any one of claims 1 to 3; or, the recognition method of the vertebral body compression fracture situation as claimed in any one of claims 4 to 7.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method of the vertebral body compression fracture analysis model as claimed in any one of claims 1 to 3; or, the recognition method of the vertebral body compression fracture situation as claimed in any one of claims 4 to 7.

Images