Rigo typing prediction model construction method and system based on gradient fuzzy decision tree

Abstract

This invention discloses a method and system for constructing a Rigo classification prediction model based on a gradient fuzzy decision tree. The method includes acquiring an emphyseal X-ray image of a patient, identifying the thoracic, lumbar, and sacral regions within the image, obtaining Rigo classification parameters based on the bounding boxes corresponding to each region, standardizing these parameters, combining them into a membership vector, and inputting this vector into a gradient fuzzy decision tree to determine the patient's scoliosis type. This invention provides a method for rapidly locating the thoracic, lumbar, and sacral regions in an emphyseal X-ray image, extracting Rigo classification parameters, improving the speed and accuracy of key feature recognition, and quickly and accurately inferring the patient's Rigo classification, while reducing subjective errors from manual judgment.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnosis and treatment technology, specifically to a method and system for constructing a Rigo classification prediction model based on gradient fuzzy decision trees. Background Technology

[0002] Wearing an orthotic brace is the primary non-surgical treatment option for scoliosis patients, and the design principles of the brace are closely related to the classification methods of scoliosis. The Rigo classification is currently the mainstream classification system, but existing Rigo classifications have the following problems and limitations: 1. In Rigo classification, imaging features are mainly extracted manually. However, manual extraction of imaging features is subject to subjective errors and is time-consuming. 2. In clinical cases, there is often ambiguity in the judgment rules that multiple characteristics correspond to different subtypes, making it difficult to accurately determine the Rigo scoliosis type of the patient.

[0003] The aforementioned problems prevent orthodontists from quickly and accurately determining the type of orthodontist and designing braces accordingly. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method and system for constructing a Rigo classification prediction model based on gradient fuzzy decision trees. This method can quickly locate the thoracic spine, lumbar spine, and sacral region in X-ray emphyseal images, extract various judgment parameters for Rigo classification, improve the speed and accuracy of key feature recognition, and quickly and accurately infer the Rigo classification of patients, reducing the subjective error of manual judgment.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to a first aspect of this application, a method for constructing a Rigo classification prediction model based on gradient fuzzy decision trees is provided, comprising: Obtain emphyseal X-ray images of the patient; Identify the thoracic, lumbar, and sacral regions in emphyseal X-ray images; Extract the target bounding box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; Based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively, the various judgment parameters of Rigo classification are obtained; Standardize all judgment parameters; The standardized judgment parameters are combined into a membership vector, which is then input into a gradient fuzzy decision tree to determine the patient's scoliosis type.

[0007] In some embodiments of this application, based on the aforementioned scheme, the YOLOv11 deep learning network is used to identify the thoracic spine region, lumbar spine region, and sacral region.

[0008] In some embodiments of this application, based on the foregoing scheme, the judgment parameters are Cobb angle, thoracic and lumbar curvature, thoracic apex cone, lumbar apex cone, transition point, HT1Tp, HT1CSL line, HTpCSL line, VT1Tp, L4 tilt state, and L4-L5 compensation state parameters. The method for obtaining Cobb angles is as follows: Identify the uppermost and lowermost vertebral bodies in the thoracic or lumbar curvature region. Draw the first parallel line corresponding to the upper endplate of the uppermost vertebra and the second parallel line corresponding to the lower endplate of the lowermost vertebra; The angle formed between the first parallel lines is taken as the Cobb angle corresponding to the thoracic or lumbar flexure region. The method for obtaining the thoracic and lumbar vertebral curvature is as follows: Based on the target detection box data corresponding to each thoracic and lumbar vertebra, the geometric center coordinates of each thoracic and lumbar vertebra are obtained. Transition points are inserted between the geometric center coordinates of adjacent vertebrae. The curvature of the thoracic and lumbar vertebrae is obtained by fitting all the geometric center coordinates of all vertebrae and all transition points. The method for obtaining the apex cone of the thoracic curve is as follows: Iterate through the coordinates of the geometric center of each thoracic vertebra, and define the thoracic vertebra with the largest horizontal offset of its geometric center as the thoracic apex cone. The method for obtaining the top cone of the waist bend is as follows: Iterate through the coordinates of the geometric center of each lumbar vertebra and define the lumbar vertebra with the largest horizontal offset of its geometric center as the lumbar apex cone. The method for obtaining HT1Tp, HT1CSL line, HTpCSL line, and VT1Tp is as follows: The transition point between the first and second thoracic vertebrae is designated as the predetermined transition point Tp. HT1Tp is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp. The HT1CSL line is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the CSL line. The HTpCSL line is the horizontal distance between the predetermined transition point Tp and the CSL line; VT1Tp is the vertical distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp. The method for obtaining the L4 tilt state is as follows: Obtain the vector difference between the geometric center coordinates of the vertebral body corresponding to the thoracic vertebral apex and the geometric center coordinates of the vertebral body corresponding to the fourth lumbar vertebra, and take the angle between the L4 end cone vector and the vector difference as the first angle; If the first included angle is obtuse, it indicates tilting towards the convex side of the thoracic spine, and the L4 tilt state is 1; if the first included angle is acute, it indicates tilting towards the concave side of the thoracic spine, and the L4 tilt state is -1; if the first included angle is right, it indicates no tilting, and the L4 tilt state is 0. The method for obtaining L4-L5 compensated state parameters is as follows: The angle between the lower conical end of the fourth thoracic vertebra and the horizontal plane is taken as the second angle, and the angle between the upper conical end of the fifth thoracic vertebra and the horizontal plane is taken as the third angle. If the difference between the second angle and the third angle is less than the difference threshold, the L4-L5 compensation state parameter is positive compensation and the L4-L5 compensation state parameter is 1. If the difference between the second angle and the third angle is greater than or equal to the difference threshold, the L4-L5 compensation state parameter is negative compensation and the L4-L5 compensation state parameter is -1.

[0009] In some embodiments of this application, based on the foregoing scheme, the standardization processing of each judgment parameter includes: Principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter to obtain the dimensionality-reduced judgment parameters. An improved fuzzy probability C-means clustering method is used to process the dimensionality reduction judgment parameters to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center. All membership degrees are combined into a membership degree vector, and the membership degree vector is used as the standardized judgment parameters.

[0010] In some embodiments of this application, based on the aforementioned scheme, principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter to obtain dimensionality-reduced judgment parameters, including: Construct using several samples including various judgment parameters Qualitative data matrix of dimension Qualitative data matrix Transform into dimensional indicator matrix ,in, For the number of samples, To determine the number of parameters, For the first The total number of categories for each judgment parameter; right dimensional score matrix and Quantization matrix of dimension Alternating updates make the loss function Gradually converges, where the loss function The calculation formula is:

[0011] In the formula, for An identity matrix of dimensionality; Principal component analysis was used to obtain the eigenvalues ​​of each judgment parameter. The variance contribution rate of each judgment parameter is calculated using the following formula:

[0012] In the formula, For the first The feature values ​​of each judgment parameter For the first The feature values ​​of each judgment parameter No. The feature values ​​of each judgment parameter For the first The characteristic values ​​of each judgment parameter; The variance contribution rates are summed in descending order until the cumulative contribution rate exceeds the contribution threshold. The number of items to retain is then determined. The judgment parameters are used as dimensionality reduction judgment parameters; In some embodiments of this application, based on the foregoing scheme, the step of processing the dimensionality reduction judgment parameters using the improved fuzzy probability C-means clustering method to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center, combining all membership degrees into a membership degree vector, and using the membership degree vector as each judgment parameter after standardization, includes: Using the dimensionality reduction judgment parameters as input, the improved fuzzy probability C-means clustering method is used to cluster the data. The objective function is:

[0013]

[0014] In the formula, It is an exponential decay factor. Density adjustment factor, , For fuzzy index, For the sample To the cluster center Fuzzy index-based The square of the Euclidean distance, and These are the membership degree and uniqueness matrix elements, respectively. Dynamic weights; Dynamic weights Defined as:

[0015]

[0016] In the formula, For the sample To the cluster center The square of the Euclidean distance; During the iteration process, the membership degree is updated alternately. Unique matrix elements With cluster center This continues until the objective function converges, ultimately yielding the center of each sample across all clusters. Membership vector on .

[0017] In some embodiments of this application, based on the foregoing scheme, the step of combining the standardized judgment parameters into a membership vector, and inputting the membership vector into a gradient fuzzy decision tree to determine the patient's scoliosis type includes: Membership vector As input, it replaces the probability defined by quantity proportion in the CART decision tree. The Gini coefficients are obtained as follows:

[0018] In the formula, Total number of categories; Node weights and These are defined as the ratio of the sum of the fuzzy membership degrees of the left child nodes to the sum of the fuzzy membership degrees of the parent node, and the ratio of the sum of the fuzzy membership degrees of the right child nodes to the sum of the fuzzy membership degrees of the parent node, respectively:

[0019] In the formula, and Representing samples respectively Fuzzy membership degree of left and right child nodes, The total number of samples; Obtaining the fuzzy Gini information gain:

[0020] in, The Gini coefficient of the left child node. is the Gini coefficient of the right child node; Based on fuzzy Gini information gain, a base fuzzy classifier is constructed. Through an ensemble learning framework, a gradient boosting mechanism is used to linearly combine several base fuzzy classifiers to construct a gradient fuzzy decision tree.

[0021] According to a second aspect of this application, a Rigo classification prediction model construction system based on gradient fuzzy decision trees is provided, comprising: The image acquisition module is used to acquire X-ray images of the patient from the front view. The recognition module is used to identify the thoracic spine region, lumbar spine region, and sacral region in X-ray emphyseal images; The extraction module is used to extract the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The judgment parameter acquisition module is used to obtain various judgment parameters of Rigo classification based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The processing module is used to standardize the various judgment parameters; The determination module is used to combine the standardized judgment parameters into a membership vector, and input the membership vector into the gradient fuzzy decision tree to determine the patient's scoliosis type.

[0022] According to a third aspect of this application, a computer-readable storage medium is provided that stores a computer program thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0023] According to a fourth aspect of this application, an electronic device is provided, comprising: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, cause the one or more processors to implement the method described above.

[0024] The beneficial effects of this application are as follows: The Rigo classification prediction model construction method and system provided in this application, based on gradient fuzzy decision trees, quickly locates the thoracic, lumbar, and sacral regions by identifying the target detection box data corresponding to the thoracic, lumbar, and sacral regions in X-ray emphyseal images. Then, based on the target detection box data corresponding to the thoracic, lumbar, and sacral regions, it obtains various judgment parameters for Rigo classification, standardizes and transforms these parameters to improve the speed and accuracy of key feature recognition, and finally combines the standardized judgment parameters into a membership vector. This membership vector is then input into a gradient fuzzy decision tree to determine the patient's scoliosis type. This method can quickly and accurately infer the patient's Rigo classification, reducing the subjective error of manual judgment.

[0025] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are intended to explain the invention, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of a Rigo classification prediction model construction method based on gradient fuzzy decision tree according to the present invention; Figure 2 This refers to the thoracic spine region, lumbar spine region, and sacral spine region in an X-ray emphyseal image identified in a specific embodiment of the present invention. Figure 3 This is a schematic diagram of the various judgment parameters identified in a specific embodiment of the present invention; Figure 4 This is a Rigo typing result identified in a specific embodiment of the present invention; Figure 5 This is a flowchart of a Rigo classification prediction model construction system based on gradient fuzzy decision tree according to the present invention; Figure 6 This is a schematic diagram of an electronic device according to the present invention. Detailed Implementation

[0027] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0028] It should be understood that the terms "comprising" and other similar expressions in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, or apparatus that includes a series of steps or units and is not limited to the listed steps or units. Furthermore, "first" and "second" are used to distinguish different objects and are not intended to describe a specific order.

[0029] According to the first aspect of this application, please refer to Figure 1 This embodiment provides a method for constructing a Rigo classification prediction model based on gradient fuzzy decision trees, including: Step S1: Obtain the patient's X-ray anteroposterior image.

[0030] In this embodiment, X-ray images of the patient are acquired using a digital X-ray device.

[0031] Step S2: Identify the thoracic spine region, lumbar spine region, and sacral region in the X-ray anteroposterior image.

[0032] In some embodiments of this example, the YOLOv11 deep learning network is used to identify the thoracic, lumbar, and sacral regions.

[0033] In this embodiment, the YOLOv11 deep learning network is used to identify the thoracic spine region, lumbar spine region, and sacral region, including: Several complete X-ray emphyseal images were used as the original dataset to ensure that each X-ray emphyseal image contained complete image data from the head to the sacrum. Data augmentation methods such as translation, cropping, and noise injection are used to expand the original dataset to obtain the recognition dataset, thereby improving the model's generalization ability and robustness. A fully trained YOLOv11 deep learning network was loaded, and rotational target detection was used to process the emphyseal images in the X-ray emphyseal images in the recognition dataset to fit the actual shape of the patient's vertebrae. This identified 12 thoracic vertebrae corresponding to the thoracic region, 5 lumbar vertebrae corresponding to the lumbar region, and the sacral region. For example... Figure 2 The image shown is a thoracic spine region, lumbar spine region, and sacral region identified in an X-ray anteroposterior image according to a specific embodiment of the present invention.

[0034] Step S3: Extract the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively.

[0035] In some embodiments of this example, the process of extracting target detection box information is as follows: output the coordinate information of the target detection box, the first column is the target category index, and the second and third columns, the fourth and fifth columns, the sixth and seventh columns, and the eighth and ninth columns are the horizontal and vertical coordinates of the four target detection box vertices, respectively, arranged in counterclockwise order.

[0036] Step S4: Based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively, obtain the various judgment parameters of Rigo classification.

[0037] Among existing scoliosis classification systems, the Rigo classification has received widespread attention due to its high consistency and excellent practicality in clinical diagnosis.

[0038] The Rigo classification integrates clinical signs and radiological imaging parameters, and is based on four imaging assessment indicators: (1) the SRS curve pattern of the International Society for the Study of Scoliosis, (2) transition point deviation, (3) T1 bone deviation, and (4) positive / negative tilt of L4 and L5. It subdivides scoliosis into six basic types: (a) type A1 (tri-curve type - subtype A1), (b) type A2+A3 (tri-curve type - subtypes A2 and A3), (c) type B1+B2 (quadri-curve type), (d) type C1+C2 (neither tri-curve nor quadri-curve type), (e) type E1 (single lumbar vertebra type), and (f) type E2 (single thoracolumbar vertebra type).

[0039] In some embodiments of this example, the judgment parameters are Cobb angle, thoracic curvature, lumbar curvature, thoracic apex cone, lumbar apex cone, transition point, HT1Tp, HT1CSL line, HTpCSL line, VT1Tp, L4 tilt state, and L4-L5 compensation state parameters. Using these parameters can cover all accurate judgments of the Rigo classification.

[0040] In some embodiments of this example, the Cobb angle is a key indicator for assessing the severity of scoliosis. During measurement, the uppermost and lowermost vertebrae in the thoracic or lumbar curvature region are first identified. A first parallel line is drawn corresponding to the superior endplate of the uppermost vertebra, and a second parallel line is drawn corresponding to the inferior endplate of the lowermost vertebra. The angle formed between these two parallel lines is the Cobb angle. In the clinical assessment system for scoliosis, the Cobb angle value shows a significant correlation with the pathological severity. Internationally accepted classification standards divide scoliosis into three grades: a Cobb angle between 10° and 20° is defined as mild scoliosis; 20° to 40° is moderate scoliosis; and above 40° is severe scoliosis. Doctors will select appropriate treatment plans based on the Cobb angle measurement data, combined with factors such as the patient's bone age and scoliosis type.

[0041] In one specific embodiment, the detection result of the Cobb angle is as follows: Figure 3 As shown, the Cobb angle (Cobb1) corresponding to the thoracic spine region is 18.86°, and the Cobb angle (Cobb2) corresponding to the lumbar spine region is 37.18°.

[0042] In some embodiments of this example, thoracic and lumbar curvatures are important geometric parameters reflecting the degree of curvature in the chest and lumbar regions. The Levenberg-Marquardt method is used to fit the circular arc and calculate the curvature in reverse. At the same time, transition points are inserted at the geometric centers of adjacent vertebrae to enhance the smoothness of the curve and achieve a quantitative assessment of scoliosis.

[0043] In this embodiment, the method for obtaining the thoracic vertebral curvature and lumbar vertebral curvature is as follows: Based on the target detection box data corresponding to each thoracic and lumbar vertebra, the geometric center of each thoracic and lumbar vertebra is obtained. Transition points are inserted between adjacent vertebral geometric centers. Arcs are fitted to the geometric centers of all thoracic vertebrae and all the transition points inserted in the geometric centers of all thoracic vertebrae, as well as the geometric centers of all lumbar vertebrae and all the transition points inserted in the geometric centers of all lumbar vertebrae. The curvature of the thoracic vertebrae and the curvature of the lumbar vertebrae are obtained based on the arcs.

[0044] Specifically, the OBB rotating target detection box based on the YOLOv11 deep learning network is used to identify the thoracic and lumbar vertebrae, accurately locating the anatomical structures of 17 vertebrae in the thoracic and lumbar vertebrae.

[0045] The coordinates of the four vertices of the quadrilateral of each target detection box are obtained. The coordinates of the geometric center of the vertices are determined by the arithmetic mean of the coordinates of the four vertices. In the continuous line segment formed by adjacent geometric centers of the vertices, 1-2 transition points are introduced to participate in the arc curve fitting operation together with the geometric center of the vertices to reduce local fluctuations and improve fitting accuracy and stability.

[0046] The center coordinates of the arc are determined by fitting the transition point with the geometric center of the vertebral body. and radius Based on the standard equation of a circle (1), an error function is established as shown in equation (2). ,Will Defined as the minimum sum of squared distances from all transition points and the geometric center of the cone to the radial direction of the fitted circular arc: (1) (2) In the formula, Represents the coordinates of the center of the circle. Let the radius be the circle. To fit the coordinates of the center of the arc, For the first A transition point and the geometric center of the vertebral body.

[0047] The Levenberg-Marquardt optimization method is used to solve equation (2) using nonlinear least squares. The parameter iterative update formula is as follows: (3) In the formula, It is a Jacobian matrix. It is the damping factor. It is the identity matrix. It is the residual vector. During the iteration process, the method compares the norm of the residuals before and after the update. The effectiveness of parameter adjustments is evaluated as follows: when the residual norm decreases, parameter updates are accepted and the damping coefficient is decreased to accelerate convergence; when the residual norm does not decrease, parameter updates are rejected and the damping coefficient is increased to enhance method stability. After multiple iterative calculations, the optimal radius estimate is finally obtained. Optimal radius estimate The reciprocal of is the curvature value of the measured profile. The curvature value of the measured contour This refers to the curvature of the thoracic spine or the curvature of the lumbar spine.

[0048] In one specific embodiment, the detection results of thoracic or lumbar curvature are as follows: Figure 3 As shown, the thoracic curvature (Curve1) is 1.6 (1 / dm) and the lumbar curvature (Curve2) is 2.7 (1 / dm).

[0049] In some embodiments of this example, the thoracic apex vertebra and the lumbar apex vertebra refer to the thoracic and lumbar vertebrae with the greatest curvature in the thoracic and lumbar segments of the spine, respectively. To accommodate the numerical input data requirements of the decision tree model, the thoracic and lumbar vertebrae need to be sequentially encoded: the initial thoracic vertebra T1 is encoded as 1, and the terminal thoracic vertebra T12 is encoded as 12. The initial vertebra L1 of the lumbar spine is coded as 13, and the final vertebra L5 is coded as 17.

[0050] The method for obtaining the apex cone of the thoracic curve is as follows: Iterate through the geometric centers of each thoracic vertebra and define the thoracic vertebra with the largest horizontal offset from its geometric center as the thoracic apex cone.

[0051] The method for obtaining the top cone of the waist bend is as follows: Traverse the geometric center of each lumbar vertebra and define the lumbar vertebra with the largest horizontal offset of its geometric center as the lumbar apex cone.

[0052] To address the special case of single-bend morphology in the Rigo classification system, the following criterion is added: when the maximum horizontal displacement of the geometric center of the thoracic or lumbar vertebral body in the coronal plane is less than 50% of the sagittal length of the vertebral body, the thoracic or lumbar vertebra is determined not to have formed an effective bend. The corresponding apical vertebra's coding value is then set to 0.

[0053] In one specific embodiment, such as Figure 3 As shown, the thoracic Apex is 7 and the lumbar Apex is 13.

[0054] In some embodiments of this example, the method for obtaining HT1Tp, HT1CSL line, HTpCSL line, and VT1Tp is as follows: The transition point between the first thoracic vertebra T1 and the second thoracic vertebra T2 is defined as the predetermined transition point Tp. HT1Tp is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp. The HT1CSL line is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the CSL line, and the CSL line is the vertical axis drawn from the center of the sacrum. The HTpCSL line is the horizontal distance between the predetermined transition point Tp and the CSL line; VT1Tp is the vertical distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp.

[0055] In some embodiments of this example, the values ​​of HT1Tp, HT1CSL, and HTpCSL are divided by the average width of the vertebral body; the value of VT1Tp is divided by the total length of the vertical projection of the spine, in order to eliminate the influence of individual differences on the dimensions of the input parameters.

[0056] In one specific embodiment, such as Figure 3 As shown, there are 11 transition points: HT1Tp is 0.9, HT1CSL is 0.3, HTpCSL is 1.1, and VT1Tp is 0.6.

[0057] In some embodiments of this example, the L4 tilt state is defined as the tilt state of the fourth lumbar vertebra relative to the thoracic curve, which can be divided into tilting towards the convex side of the thoracic spine and tilting towards the concave side of the thoracic spine. The method for obtaining the L4 tilt state is as follows: Calculate the L4 end cone vector The angle with the horizontal plane, if If the angle is less than 5°, the tilt parameter is determined to be 0; if the parallelism is not achieved, further judgment is performed. Obtain the vector difference between the geometric center coordinates of the vertebral body corresponding to the thoracic vertebral apex and the geometric center coordinates of the vertebral body corresponding to the fourth lumbar vertebra. Take the angle between the L4 apex vector and the vector difference as the first angle. If the first angle is obtuse, it indicates tilting towards the convex side of the thoracic vertebra, and if it is acute, it indicates tilting towards the concave side of the thoracic vertebra.

[0058] L4 tilt is represented by a numerical value, where 1 indicates tilt towards the convex side of the thoracic spine, 0 indicates a horizontal state, and -1 indicates tilt towards the concave side of the thoracic spine.

[0059] In one specific embodiment, such as Figure 3 As shown, the L4 incline is -1.

[0060] In some embodiments of this example, the L4-L5 compensation state parameters are defined as follows: Negative compensation occurs when the tilt direction and angle of the fourth and fifth lumbar vertebrae L5 are consistent with L4, and the endplates of the fourth and fifth lumbar vertebrae L5 are parallel to L4; positive compensation occurs when the tilt direction and angle of the fourth and fifth lumbar vertebrae L5 are consistent with L4, but the amplitude of the endplates of the fourth and fifth lumbar vertebrae L5 is significantly reduced. The method for obtaining the L4-L5 compensation state parameters is as follows: The angle between the lower conical end of the fourth thoracic vertebra and the horizontal plane is taken as the second angle, and the angle between the upper conical end of the fifth thoracic vertebra and the horizontal plane is taken as the third angle. If the difference between the second angle and the third angle is less than the difference threshold, the L4-L5 compensation state parameter is positive compensation; otherwise, it is negative compensation.

[0061] The L4-L5 compensation state parameters use 1 to represent positive compensation and -1 to represent negative compensation.

[0062] In one specific embodiment, such as Figure 3 As shown, the L4-L5 compensation state parameter (L4-L5) is 1.

[0063] Table 1 shows the range of each parameter.

[0064] Table 1 Range of each parameter

[0065] Step S5: Standardize the various judgment parameters.

[0066] In some implementations of this embodiment, the standardization processing of various judgment parameters includes: Principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter to obtain the dimensionality-reduced judgment parameters. An improved fuzzy probability C-means clustering method is used to process the dimensionality reduction judgment parameters to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center, and all membership degrees are combined into a membership degree vector.

[0067] In some embodiments of this example, principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter, resulting in dimensionality-reduced judgment parameters, including: Construct using several samples including various judgment parameters Qualitative data matrix of order Qualitative data matrix Convert to indicator matrix ,in, For the number of samples, To determine the number of parameters, define 3D matrix As shown in the following formula: (4) In the formula, Indicates the first The first sample The feature variable corresponding to the judgment parameter is in the th _ ... The discrete classification value of the category, which is determined by threshold classification.

[0068] Through the above operations, continuous quantitative data is transformed into qualitative data with clear category classification.

[0069] After converting quantitative data into qualitative data, based on the qualitative data matrix structure Dimensional indicator matrix As shown below: (5) In the formula, For the first The total number of categories for each judgment parameter, matrix elements The value is a binary variable, when the first... The sample belongs to the first The first judgment parameter The value is 1 for categories and 0 otherwise. By qualitative data matrix The discrete classification information is converted into Boolean-type attribution identifiers, and a membership matrix between samples and feature categories is established.

[0070] right dimensional score matrix and Quantization matrix of dimension Alternating updates make the loss function Gradually converges, where the loss function The calculation formula is: (6) in, for A dimensional score matrix, for An identity matrix of dimension 1 yes Quantization matrix of dimension The column vectors correspond to the categories of each variable in Quantization weights in dimensional space.

[0071] By alternating updates to the score matrix Quantization matrix The parameters are set to allow the loss function to converge to a steady state. During the initialization phase, the score matrix needs to be adjusted based on the standard normal distribution information. With quantization matrix Parameters are assigned. A fixed score matrix is ​​executed alternately. Update quantization matrix and fixed quantization matrix Update the score matrix The iterative operation ensures that the parameter vector approaches the optimal solution in each iteration.

[0072] To achieve the desired dimensionality reduction, an overestimation method was used in the initial setting phase. The value is set to the feature threshold of the original feature number to avoid missing the main feature information.

[0073] In one specific embodiment, the feature threshold can be selected as 80%.

[0074] Principal component analysis (PCA) was used for data dimensionality reduction. Principal components were selected based on eigenvalue criteria to ensure that the retained components represent the maximum variance information in the dataset. The characteristic equation is defined as follows: (7) In the formula, For the first The eigenvalues ​​of the judgment parameters are positively correlated with the amount of information carried by the principal components. It is an identity matrix.

[0075] It is all The correlation matrix of the combined linear scores of the matrix set is defined as follows: (8) (9) In the formula, To transform the data matrix, by column vectors Composition; matrix, column vector Reflecting the The mapping relationship between the category information of each judgment parameter and the sample score. for A column vector, used to record the first... The quantized weights of each category for each judgment parameter.

[0076] After obtaining the eigenvalues ​​of each judgment parameter, the variance contribution rate of each judgment parameter is calculated. The variance contribution rate is used to evaluate the degree of contribution of each principal component to the data variance. The calculation formula is as follows: (10) in, To determine the total number of parameters, No. The feature values ​​of each judgment parameter For the first The characteristic values ​​of each judgment parameter; The variance contribution rates are accumulated in descending order until the cumulative contribution rate exceeds the contribution threshold, and the number of items to be retained is determined. The judgment parameters are used as low-dimensional output, and the output size is dimensional score matrix .

[0077] In one specific embodiment, the contribution threshold can be selected as 90%.

[0078] Thus, due to the large number of judgment parameters, this embodiment uses principal component analysis based on alternating least squares to reduce the dimensionality of each judgment parameter, thereby reducing redundancy, minimizing the impact of multicollinearity, and extracting key judgment parameters.

[0079] In some implementations of this embodiment, an improved fuzzy probability C-means clustering method is used to process the dimensionality reduction judgment parameters to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center. All membership degrees are then combined into a membership degree vector, including: Using the aforementioned low-dimensional output as input, the data is clustered using the Modified Fuzzy Possibilistic C Means (MFPCM) method. The objective function of MFPCM is: (11) (12) In the formula, It is an exponential decay factor. Density adjustment factor, , For fuzzy index, For the sample To the cluster center Fuzzy index-based The square of the Euclidean distance, and These are the membership degree and uniqueness matrix elements, respectively. For dynamic weights.

[0080] Dynamic weights Defined as: (13) (14) In the formula, For the sample To the cluster center The square of the Euclidean distance.

[0081] In this embodiment, MFPCM uses an exponential decay factor. and density regulation factor This method suppresses noise and outliers, causing data points far from the cluster center to rapidly decrease their weights during iterative updates, thus improving the clustering's compactness and robustness. During iteration, membership degrees are updated alternately. Unique matrix elements With cluster center This continues until the objective function converges, ultimately yielding the center of each sample across all clusters. Membership vector on .

[0082] Thus, the dimensionality-reduced judgment parameters obtained after dimensionality reduction may retain the main information but have reduced dimensionality. At this time, the improved fuzzy probability C-means clustering method can better explore the internal structure of the dimensionality-reduced judgment parameters, generate membership vectors as new features, and improve the effect of subsequent analysis.

[0083] Step S6: Combine the standardized judgment parameters into a membership vector, input the membership vector into the gradient fuzzy decision tree, and determine the patient's scoliosis type.

[0084] Membership vector As input, it replaces the probability defined by quantity proportion in the CART decision tree. The Gini coefficients are obtained as follows: (15) In the formula, This represents the total number of categories.

[0085] And replace the node weights determined by the proportion of the sample size. and , node weight and These are defined as the ratio of the sum of the fuzzy membership degrees of the left child nodes to the sum of the fuzzy membership degrees of the parent node, and the ratio of the sum of the fuzzy membership degrees of the right child nodes to the sum of the fuzzy membership degrees of the parent node, respectively: (16) in, and Representing samples respectively Fuzzy membership degree of left and right child nodes, The total number of samples. Gradient fuzzy decision trees use fuzzy Gini information gain as the node splitting factor. A larger fuzzy Gini information gain indicates a greater proportion of the input's influence on the result. The fuzzy Gini information gain is: (17) in, The Gini coefficient of the left child node. is the Gini coefficient of the right child node.

[0086] Based on fuzzy Gini information gain, a base fuzzy classifier is constructed. Through an ensemble learning framework, a gradient boosting mechanism is used to linearly combine several base fuzzy classifiers to construct a gradient fuzzy decision tree.

[0087] This embodiment utilizes a phased optimization strategy to progressively improve the performance of the gradient fuzzy decision tree. In the initial stage, the gradient fuzzy decision tree parameters are initialized, with the goal of minimizing the loss function on the training sample set to determine the initial predicted values. During iteration, the pseudo-residual between the current predicted value and the actual value is first calculated, and then the gradient fuzzy decision tree is used to fit and model the pseudo-residual. After each base fuzzy classifier completes training, the parameters of the overall gradient fuzzy decision tree are updated based on the gradient descent direction output by the base fuzzy classifier. This process is repeated until a preset number of iterations is reached, resulting in a trained gradient fuzzy decision tree.

[0088] Through the iterative process of gradient boosting described above, a trained gradient fuzzy decision tree is finally constructed. Using this trained gradient fuzzy decision tree, the samples are pushed down to the leaf nodes, and Rigo classification prediction is performed based on the category distribution of the leaf nodes, achieving high-precision automated classification of X-ray emphyseal images.

[0089] Thus, the membership vector As input, it replaces the probability defined by quantity proportion in the CART decision tree. Membership vectors provide probabilistic fuzzy features, while gradient fuzzy decision trees effectively utilize these features through gradient optimization and fuzzy rules, thereby improving the accuracy of Rigo classification prediction.

[0090] like Figure 4 As shown, in one specific embodiment, the Rigo classification prediction model construction method based on gradient fuzzy decision tree provided in this embodiment can predict that the lateral curvature type of the X-ray emphyseal image is type B1.

[0091] According to the second aspect of this application, such as Figure 5 As shown in the figure, this embodiment provides a Rigo classification prediction model construction system based on gradient fuzzy decision trees, including: The image acquisition module is used to acquire X-ray images of the patient from the front view. The recognition module is used to identify the thoracic spine region, lumbar spine region, and sacral region in X-ray emphyseal images; The extraction module is used to extract the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The judgment parameter acquisition module is used to obtain various judgment parameters of Rigo classification based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The processing module is used to standardize the various judgment parameters; The determination module is used to combine the standardized judgment parameters into a membership vector, and input the membership vector into the gradient fuzzy decision tree to determine the patient's scoliosis type.

[0092] Specifically, this embodiment corresponds one-to-one with the above method embodiments. The functions of each module have been described in detail in the corresponding method embodiments, so they will not be repeated here.

[0093] According to a third aspect of this application, this embodiment provides a computer-readable storage medium having a computer program stored thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0094] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0095] According to a fourth aspect of this application, an electronic device is provided, comprising: One or more processors; Memory is used to store executable instructions for the processor, which, when executed by one or more processors, cause one or more processors to implement the methods described above.

[0096] Electronic devices are manifested in the form of general-purpose computing devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including memory and processor).

[0097] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of a computer system, connecting all parts of the computer system through various interfaces and lines.

[0098] Memory can be used to store computer programs and / or modules. The processor implements various functions of the computer system by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0099] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and memory) containing computer-usable program code.

[0100] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a Rigo classification prediction model based on gradient fuzzy decision trees, characterized in that, include: Obtain emphyseal X-ray images of the patient; Identify the thoracic, lumbar, and sacral regions in emphyseal X-ray images; Extract the target bounding box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; Based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively, the various judgment parameters of Rigo classification are obtained; Standardize all judgment parameters; The standardized judgment parameters are combined into a membership vector, which is then input into a gradient fuzzy decision tree to determine the patient's scoliosis type. Specifically: Membership vector As input, replacing the probability defined by quantity proportion in the CART decision tree, we obtain the Gini coefficient: In the formula, Total number of categories; Node weights and These are defined as the ratio of the sum of the fuzzy membership degrees of the left child nodes to the sum of the fuzzy membership degrees of the parent node, and the ratio of the sum of the fuzzy membership degrees of the right child nodes to the sum of the fuzzy membership degrees of the parent node, respectively: In the formula, and Representing samples respectively Fuzzy membership degree of left and right child nodes, The number of samples; Obtaining the fuzzy Gini information gain: In the formula, The Gini coefficient of the left child node. The Gini coefficient of the right child node; Based on fuzzy Gini information gain, a base fuzzy classifier is constructed. Through an ensemble learning framework, a gradient boosting mechanism is used to linearly combine several base fuzzy classifiers to construct a gradient fuzzy decision tree.

2. The method for constructing a Rigo classification prediction model based on gradient fuzzy decision trees according to claim 1, characterized in that: The YOLOv11 deep learning network was used to identify the thoracic spine region, lumbar spine region, and sacral region.

3. The method for constructing a Rigo classification prediction model based on gradient fuzzy decision trees according to claim 1, characterized in that: The parameters for each judgment are Cobb angle, thoracic and lumbar curvature, thoracic apex cone, lumbar apex cone, transition point, HT1Tp, HT1CSL line, HTpCSL line, VT1Tp, L4 tilt state, and L4-L5 compensation state parameters. The method for obtaining Cobb angles is as follows: Identify the uppermost and lowermost vertebral bodies in the thoracic or lumbar curvature region. Draw the first parallel line corresponding to the upper endplate of the uppermost vertebra and the second parallel line corresponding to the lower endplate of the lowermost vertebra. The angle formed between the first parallel lines is taken as the Cobb angle corresponding to the thoracic or lumbar flexure region. The method for obtaining the thoracic and lumbar vertebral curvature is as follows: Based on the target detection box data corresponding to each thoracic and lumbar vertebra, the geometric center coordinates of each thoracic and lumbar vertebra are obtained. Transition points are inserted between the geometric center coordinates of adjacent vertebrae. The curvature of the thoracic and lumbar vertebrae is obtained by fitting all the geometric center coordinates of all vertebrae and all transition points. The method for obtaining the apex cone of the thoracic curve is as follows: Iterate through the coordinates of the geometric center of each thoracic vertebra, and define the thoracic vertebra with the largest horizontal offset of its geometric center as the thoracic apex cone. The method for obtaining the top cone of the waist bend is as follows: Iterate through the coordinates of the geometric center of each lumbar vertebra and define the lumbar vertebra with the largest horizontal offset of its geometric center as the lumbar apex cone. The method for obtaining HT1Tp, HT1CSL line, HTpCSL line, and VT1Tp is as follows: The transition point between the first and second thoracic vertebrae is designated as the predetermined transition point Tp. HT1Tp is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp. The HT1CSL line is the horizontal distance between the geometric center coordinates of the first thoracic vertebra and the CSL line. The HTpCSL line is the horizontal distance between the predetermined transition point Tp and the CSL line; VT1Tp is the vertical distance between the geometric center coordinates of the first thoracic vertebra and the predetermined transition point Tp. The method for obtaining the L4 tilt state is as follows: Obtain the vector difference between the geometric center coordinates of the vertebral body corresponding to the thoracic vertebral apex and the geometric center coordinates of the vertebral body corresponding to the fourth lumbar vertebra, and take the angle between the L4 end cone vector and the vector difference as the first angle; If the first included angle is obtuse, it indicates tilting towards the convex side of the thoracic spine, and the L4 tilt state is 1; if the first included angle is acute, it indicates tilting towards the concave side of the thoracic spine, and the L4 tilt state is -1; if the first included angle is right, it indicates no tilting, and the L4 tilt state is 0. The method for obtaining L4-L5 compensated state parameters is as follows: The angle between the lower conical end of the fourth thoracic vertebra and the horizontal plane is taken as the second angle, and the angle between the upper conical end of the fifth thoracic vertebra and the horizontal plane is taken as the third angle. If the difference between the second angle and the third angle is less than the difference threshold, the L4-L5 compensation state parameter is positive compensation and the L4-L5 compensation state parameter is 1. If the difference between the second angle and the third angle is greater than or equal to the difference threshold, the L4-L5 compensation state parameter is negative compensation and the L4-L5 compensation state parameter is -1.

4. The method for constructing a Rigo classification prediction model based on a gradient fuzzy decision tree according to claim 1, characterized in that, The standardization processing of each judgment parameter includes: Principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter to obtain the dimensionality-reduced judgment parameters. An improved fuzzy probability C-means clustering method is used to process the dimensionality reduction judgment parameters to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center. All membership degrees are combined into a membership degree vector, and the membership degree vector is used as the standardized judgment parameters.

5. The method for constructing a Rigo classification prediction model based on a gradient fuzzy decision tree according to claim 4, characterized in that: Principal component analysis based on alternating least squares is used to reduce the dimensionality of each judgment parameter, resulting in the following dimensionality-reduced judgment parameters: Construct using several samples including various judgment parameters Qualitative data matrix of dimension Qualitative data matrix Transform into dimensional indicator matrix ,in, To determine the number of parameters, For the first The total number of categories for each judgment parameter; right dimensional score matrix and dimensional quantization matrix Alternate updates To achieve the desired dimensionality reduction, the loss function is... Gradually converges, where the loss function The calculation formula is: In the formula, for An identity matrix of dimensionality; Principal component analysis was used to obtain the eigenvalues ​​of each judgment parameter. ; The variance contribution rate of each judgment parameter is calculated using the following formula: In the formula, For the first The feature values ​​of each judgment parameter No. The feature values ​​of each judgment parameter For the first The characteristic values ​​of each judgment parameter; The variance contribution rates are summed in descending order until the cumulative contribution rate exceeds the contribution threshold. The number of items to retain is then determined. The judgment parameters are used as dimensionality reduction judgment parameters.

6. The method for constructing a Rigo classification prediction model based on a gradient fuzzy decision tree according to claim 4, characterized in that, The improved fuzzy probability C-means clustering method is used to process the dimensionality reduction judgment parameters to obtain the membership degree of each dimensionality reduction judgment parameter to the corresponding cluster center. All membership degrees are combined into a membership degree vector, and the membership degree vector is used as the standardized judgment parameters, including: Using the dimensionality reduction judgment parameters as input, the improved fuzzy probability C-means clustering method is used to cluster the data. The objective function is: In the formula, It is an exponential decay factor. Density adjustment factor, , For fuzzy index, For the sample To the cluster center Fuzzy index-based The square of the Euclidean distance, and These are the membership degree and uniqueness matrix elements, respectively. Dynamic weights; Dynamic weights Defined as: In the formula, For the sample To the cluster center The square of the Euclidean distance; During the iteration process, the membership degree is updated alternately. Unique matrix elements With cluster center This continues until the objective function converges, ultimately yielding the center of each sample across all clusters. Membership vector on .

7. A Rigo classification prediction model construction system based on gradient fuzzy decision trees, applied to the Rigo classification prediction model construction method based on gradient fuzzy decision trees as described in any one of claims 1-6, characterized in that, include: The image acquisition module is used to acquire X-ray images of the patient in frontal view. The recognition module is used to identify the thoracic spine region, lumbar spine region, and sacral region in X-ray emphyseal images; The extraction module is used to extract the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The judgment parameter acquisition module is used to obtain various judgment parameters of Rigo classification based on the target detection box data corresponding to the thoracic spine region, lumbar spine region and sacral region respectively; The processing module is used to standardize the various judgment parameters; The determination module is used to combine the standardized judgment parameters into a membership vector, and input the membership vector into the gradient fuzzy decision tree to determine the patient's scoliosis type.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program includes executable instructions that, when executed by a processor, implement the method of any one of claims 1-6.

9. An electronic device, characterized in that, include: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, cause the one or more processors to perform the method according to any one of claims 1-6.

Images