An intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses an intelligent quantitative assessment system for adolescent scoliosis that integrates multimodal imaging. The system includes a multimodal data management module, a cross-modal registration fusion module, a dynamic quantitative analysis module, a risk prediction module, a three-dimensional reconstruction module, a report generation module, and a system management module. By setting a multimodal registration verification mechanism, the accurate matching of bony structure and soft tissue is ensured, and the calculation error of key parameters caused by registration failure is reduced, so that the three-dimensional reconstruction accuracy is greatly improved compared with traditional single-modal assessment. By introducing a dynamic breathing compensation module, the physiological changes of the spine in the respiratory movement cycle are automatically captured during the quantification stage of spinal deformation parameters, and the measurement values ​​are corrected based on an individualized dynamic model to distinguish between pathological progression and physiological fluctuations, thereby avoiding the risk of misjudgment caused by ignoring respiratory phase changes in traditional static measurements, and improving the reliability of scoliosis progression risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent quantitative assessment system for adolescent scoliosis integrating multimodal imaging. Background Art

[0002] Scoliosis is a common spinal deformity, also known as scoliosis. It is mainly manifested by the side of the spine deviating to one side, making the back present a C-shaped or S-shaped curve. A spinal X-ray shows that the spine has a lateral curvature greater than 10°, which can be diagnosed as scoliosis. This disease is common in adolescence, and the symptoms gradually worsen with age. In severe cases, it affects breathing and heart function, and may even cause spinal cord compression and paralysis.

[0003] Currently, there are defects in the application of multimodal imaging technology in the clinical evaluation of adolescent scoliosis: due to the misalignment of the anatomical coordinate systems of different modal images, it is impossible to detect the bone and soft tissue registration deviation in real time during three-dimensional spinal structure reconstruction. If the registration error exceeds 3mm, the calculation of key parameters will be distorted; at the same time, the existing evaluation system cannot monitor the impact of respiratory movement on spinal deformation in real time, and misjudges physiological deformation as pathological progression, resulting in a risk prediction accuracy reduction of more than 20%; in addition, due to the black box characteristics of the deep learning algorithm, the decision basis of the model cannot be visualized during key point detection, resulting in the inability of doctors to verify the reliability of AI markers, leading to the risk of clinical misdiagnosis; in the dynamic quantitative evaluation process, due to the lack of a multi-dimensional parameter fusion mechanism, the coordinated analysis of deformation parameters and biomechanical characteristics cannot be achieved, which further limits the accuracy of generating personalized diagnosis and treatment plans and affects the clinical applicability of scoliosis evaluation.

[0004] Therefore, in order to solve the above problems, the present invention provides an intelligent quantitative assessment system for adolescent scoliosis integrating multimodal imaging. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the shortcomings of the existing technology, the present invention provides an intelligent quantitative assessment system for adolescent scoliosis that integrates multimodal imaging, which solves the problems raised in the above-mentioned background technology.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention provides the following technical solutions: an intelligent quantitative assessment system for adolescent scoliosis integrating multimodal imaging, comprising:

[0009] Multimodal data management module:

[0010] An image acquisition unit equipped with a medical imaging communication protocol interface can acquire X-ray, CT, and MRI data;

[0011] Structured database unit that stores clinical parameters and supports batch import and export;

[0012] Cross-modal registration and fusion module:

[0013] Affine transformation unit, which performs X-ray-CT anatomical alignment;

[0014] GAN fusion unit, which realizes the fusion of spatial domain features of MRI and CT;

[0015] Dynamic quantitative analysis module:

[0016] Key point detection unit, integrating improved real-time multi-person pose estimation network model and nested U-shaped network;

[0017] Parameter calculation unit, dynamically calculates spinal curvature angle, AVR and sagittal balance parameters;

[0018] Risk prediction module:

[0019] Extreme gradient boosting tree ensemble learning unit, outputting scoliosis progress probability;

[0020] Interpretability processing unit, generating SHAP feature contribution graph;

[0021] 3D reconstruction module:

[0022] Point cloud processing unit, constructing a 3D mesh model of the spine;

[0023] Deformation simulation unit, which visualizes spinal deformation under different loads;

[0024] Report generation module:

[0025] Multi-view fusion unit, integrating 3D models, heat maps and time series curves;

[0026] The early warning trigger unit generates red, yellow, and blue warning signals according to the risk level;

[0027] System management module:

[0028] A permission control unit based on the role-based access control model distinguishes the operation permissions of doctors, patients, and administrators;

[0029] Audit log unit, records data access and algorithm call behavior.

[0030] Preferably, the method comprises the following steps:

[0031] S1. Collecting multimodal medical imaging data and clinical parameters of the patient, wherein the multimodal medical imaging includes spinal X-rays, computed tomography scans, and magnetic resonance imaging, and the clinical parameters include gender, age, skeletal maturity index, spinal curvature angle, and scoliosis classification system classification;

[0032] S2. Cross-modal registration and fusion processing of multimodal medical images:

[0033] Affine transformation was used to align the coronal and sagittal anatomical structures of the X-ray and CT images;

[0034] Fusion of MRI soft tissue information to CT bone structure coordinate system based on generative adversarial network;

[0035] Generate standardized three-dimensional spinal fusion models;

[0036] S3. Perform multi-plane keypoint detection on the fusion model:

[0037] The improved real-time multi-person pose estimation model is used to detect vertebral corners and endplate landmarks in the coronal plane;

[0038] A nested U-shaped network was used to identify the sacral tilt angle and thoracolumbar turning point in the sagittal plane;

[0039] S4. Dynamic quantification of spinal deformation parameters:

[0040] Calculate the major and minor spinal curvature angles based on the key points of the coronal plane:

[0041]

[0042] in are the normal vectors of the upper and lower vertebral endplates, respectively;

[0043] Calculate the thoracic kyphosis angle and lumbar lordosis angle based on the key points in the sagittal plane;

[0044] S5. Construct a multi-factor risk prediction model:

[0045] The clinical parameters and deformation parameters are input into the extreme gradient boosting ensemble learning framework;

[0046] Analyze the feature contribution by adding Shapley and explaining the value, and output the probability of scoliosis progression ;

[0047] S6. Generate dynamic quantitative evaluation report:

[0048] Fusion of 3D reconstruction model, key point heat map, and parameter time series comparison curve;

[0049] Early warning signals are triggered based on the progression probability level.

[0050] Preferably, the cross-modality registration in step S2 specifically includes:

[0051] S21, performing edge enhancement and grayscale normalization on X-ray film;

[0052] S22, extracting bony landmarks from CT images using scale-invariant feature transformation;

[0053] S23. Calculate the projection transformation matrix between X-ray and CT by random sampling consensus algorithm :

[0054]

[0055] in is the affine transformation function, is the coordinate of the i-th bony landmark point in the CT image, is the coordinate of the i-th projection point matched in the X-ray film, is the number of matching feature point pairs;

[0056] S24. Use a recurrent generative adversarial network to map MRI T2-weighted images to the CT coordinate system:

[0057]

[0058] in Generate adversarial networks for MRI to CT, is the original MRI image, is a synthetic pseudo CT image.

[0059] Preferably, the key point detection in step S3 adopts a cascade optimization mechanism:

[0060] S31, localize the vertebral region of interest through the spatial attention module;

[0061] S32, Extracting coronal plane key point coordinates using a high-resolution network with adaptive sampling ;

[0062] S33, Constructing vertebral topology constraints based on graph convolutional networks;

[0063] S34. Optimizing sagittal plane key point sequences via bidirectional long short-term memory networks .

[0064] Preferably, the parameter calculation in step S4 includes a dynamic compensation mechanism:

[0065] S41, segmenting CT sequence frames according to respiratory phase;

[0066] S42, calculate the change in spinal curvature angle at each phase ;

[0067] S43. Establish deformation-respiration cycle correlation function:

[0068]

[0069] in is the deformation amplitude coefficient, For the respiratory phase, is the duration of a complete breathing cycle, is the baseline offset.

[0070] Preferably, the model building in step S5 includes:

[0071] S51. Dealing with imbalanced distribution of clinical data by synthetic minority class oversampling technique;

[0072] S52, using Bayesian optimization to search for extreme gradient boosting tree hyperparameter combinations;

[0073] S53. Construct multi-task output layer:

[0074] Task 1: Probability of scoliosis progression ;

[0075] Task 2: Classification labeling of surgical indications .

[0076] Preferably, the method further includes an algorithm adaptive optimization step:

[0077] S61. Establish an algorithm library:

[0078]

[0079] Contains key point detection algorithms and prediction models;

[0080] S62, real-time monitoring of the area under the curve and F1 score of each algorithm on the validation set;

[0081] S63, when Less than -σ triggers algorithm switching:

[0082] Using the improved Osprey search algorithm to select the optimal alternative algorithm :

[0083]

[0084] in is a collection of algorithm libraries, is the clinical weight factor, is the area under the curve, is the F1 score.

[0085] Preferably, a three-dimensional reconstruction step is added after step S2:

[0086] S71. Construct a spine point cloud model based on the registration results;

[0087] S72. Generate topology optimized 3D mesh by Poisson surface reconstruction;

[0088] S73. Calculate the vertebral rotation AVR on the mesh model:

[0089]

[0090] in is the average vertebral rotation, is the number of vertebrae involved in the calculation, is the normal vector of the i-th vertebral cross section, is the reference plane normal vector.

[0091] Preferably, the step S5 introduces interpretability processing:

[0092] S81, generating a key point detection heat map through gradient weighted class activation mapping;

[0093] S82. Use a local interpretable model-independent explanation algorithm to explain the extreme gradient boosting tree prediction results;

[0094] S83. Visualize the feature contribution ranking in the evaluation report.

[0095] Preferably, the cloud platform deployment step is also included:

[0096] S91. Use containerization technology to encapsulate the evaluation algorithm;

[0097] S92. Dynamically allocate computing resources through the container orchestration system.

[0098] S93. Implement hierarchical authority control for doctors, patients, and administrators based on open authorization protocols.

[0099] (3) Beneficial effects

[0100] Compared with the existing technology, the present invention provides an intelligent quantitative assessment system for adolescent scoliosis that integrates multimodal imaging, which has the following beneficial effects:

[0101] 1. In the present invention, by setting up a multimodal registration verification mechanism, the anatomical structure alignment deviation of X-ray, CT and MRI images is detected in real time during the fusion process of the spinal three-dimensional model. When the error exceeds a reasonable range, the calibration process is automatically triggered and a spatial error heat map is generated to ensure the precise matching of bony structure and soft tissue, reduce the calculation error of key parameters caused by registration failure, and significantly improve the three-dimensional reconstruction accuracy compared with traditional single-modality evaluation.

[0102] 2. In the present invention, by introducing a dynamic breathing compensation module, the physiological changes of the spine during the respiratory movement cycle are automatically captured during the quantification stage of spinal deformation parameters. The measurement values ​​are corrected based on an individualized dynamic model to distinguish between pathological progression and physiological fluctuations, thereby avoiding the risk of misjudgment caused by ignoring respiratory phase changes in traditional static measurements and improving the reliability of scoliosis progression risk assessment.

[0103] 3. In the present invention, by constructing an adaptive algorithm optimization engine, the performance degradation of the core algorithm is continuously monitored when the system is running. When a decrease in diagnostic accuracy is detected, the optimal alternative algorithm is automatically switched to solve the problem of algorithm failure caused by differences in imaging equipment and special body shapes, and enhance the system's adaptability to different clinical scenarios.

[0104] 4. In the present invention, by integrating a dual-channel explainable analysis module, the decision-making process of deep learning is converted into a visual heat map and feature contribution map, and the generation basis of key parameters is intuitively displayed in the three-dimensional spinal model, allowing doctors to trace the decision logic of AI and significantly reduce the risk of clinical misdiagnosis caused by the black box characteristics of the algorithm.

[0105] 5. In the present invention, by establishing a multi-dimensional closed-loop diagnosis and treatment system, integrating biomechanical characteristics, dynamic deformation parameters and clinical classification data, a quantitative evaluation report and a personalized correction plan are generated simultaneously, realizing full-link optimization from image analysis to clinical decision-making, and improving the diagnosis and treatment efficiency of scoliosis and the accuracy of plan matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 This is a structural diagram of an intelligent quantitative assessment system for adolescent scoliosis that integrates multimodal imaging according to the present invention. DETAILED DESCRIPTION

[0107] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0108] See also Figure 1 This is an intelligent quantitative assessment system for adolescent scoliosis that integrates multimodal imaging, including:

[0109] Multimodal data management module:

[0110] An image acquisition unit equipped with a medical imaging communication protocol interface can acquire X-ray, CT, and MRI data;

[0111] Structured database unit that stores clinical parameters and supports batch import and export;

[0112] Cross-modal registration and fusion module:

[0113] Affine transformation unit, which performs X-ray-CT anatomical alignment;

[0114] GAN fusion unit, which realizes the fusion of spatial domain features of MRI and CT;

[0115] Dynamic quantitative analysis module:

[0116] Key point detection unit, integrating improved real-time multi-person pose estimation network model and nested U-shaped network;

[0117] Parameter calculation unit, dynamically calculates spinal curvature angle, AVR and sagittal balance parameters;

[0118] Risk prediction module:

[0119] Extreme gradient boosting tree ensemble learning unit, outputting scoliosis progress probability;

[0120] Interpretability processing unit, generating SHAP feature contribution graph;

[0121] 3D reconstruction module:

[0122] Point cloud processing unit, constructing a 3D mesh model of the spine;

[0123] Deformation simulation unit, which visualizes spinal deformation under different loads;

[0124] Report generation module:

[0125] Multi-view fusion unit, integrating 3D models, heat maps and time series curves;

[0126] The early warning trigger unit generates red, yellow, and blue warning signals according to the risk level;

[0127] System management module:

[0128] A permission control unit based on the role-based access control model distinguishes the operation permissions of doctors, patients, and administrators;

[0129] Audit log unit, records data access and algorithm call behavior.

[0130] The following steps are involved:

[0131] S1. Collecting multimodal medical imaging data and clinical parameters of the patient, wherein the multimodal medical imaging includes spinal X-rays, computed tomography scans, and magnetic resonance imaging, and the clinical parameters include gender, age, skeletal maturity index, spinal curvature angle, and scoliosis classification system classification;

[0132] S2. Cross-modal registration and fusion processing of multimodal medical images:

[0133] Affine transformation was used to align the coronal and sagittal anatomical structures of the X-ray and CT images;

[0134] Fusion of MRI soft tissue information to CT bone structure coordinate system based on generative adversarial network;

[0135] Generate standardized three-dimensional spinal fusion models;

[0136] S3. Perform multi-plane keypoint detection on the fusion model:

[0137] The improved real-time multi-person pose estimation model is used to detect vertebral corners and endplate landmarks in the coronal plane;

[0138] A nested U-shaped network was used to identify the sacral tilt angle and thoracolumbar turning point in the sagittal plane;

[0139] S4. Dynamic quantification of spinal deformation parameters:

[0140] Calculate the major and minor spinal curvature angles based on the key points of the coronal plane:

[0141]

[0142] in are the normal vectors of the upper and lower vertebral endplates, respectively;

[0143] Calculate the thoracic kyphosis angle and lumbar lordosis angle based on the key points in the sagittal plane;

[0144] S5. Construct a multi-factor risk prediction model:

[0145] The clinical parameters and deformation parameters are input into the extreme gradient boosting ensemble learning framework;

[0146] Analyze the feature contribution by adding Shapley and explaining the value, and output the probability of scoliosis progression ;

[0147] S6. Generate dynamic quantitative evaluation report:

[0148] Fusion of 3D reconstruction model, key point heat map, and parameter time series comparison curve;

[0149] Early warning signals are triggered based on the progression probability level.

[0150] The cross-modality registration in step S2 specifically includes:

[0151] S21, performing edge enhancement and grayscale normalization on X-ray film;

[0152] S22, extracting bony landmarks from CT images using scale-invariant feature transformation;

[0153] S23. Calculate the projection transformation matrix between X-ray and CT by random sampling consensus algorithm :

[0154]

[0155] in is the affine transformation function, is the coordinate of the i-th bony landmark point in the CT image, is the coordinate of the i-th projection point matched in the X-ray film, is the number of matching feature point pairs;

[0156] S24. Use a recurrent generative adversarial network to map MRI T2-weighted images to the CT coordinate system:

[0157]

[0158] in Generate adversarial networks for MRI to CT, is the original MRI image, is a synthetic pseudo CT image.

[0159] The key point detection in step S3 adopts a cascade optimization mechanism:

[0160] S31, localize the vertebral region of interest through the spatial attention module;

[0161] S32, Extracting coronal plane key point coordinates using a high-resolution network with adaptive sampling ;

[0162] S33, Constructing vertebral topology constraints based on graph convolutional networks;

[0163] S34. Optimizing sagittal plane key point sequences via bidirectional long short-term memory networks .

[0164] The parameter calculation of step S4 includes a dynamic compensation mechanism:

[0165] S41, segmenting CT sequence frames according to respiratory phase;

[0166] S42, calculate the change in spinal curvature angle at each phase ;

[0167] S43. Establish deformation-respiration cycle correlation function:

[0168]

[0169] in is the deformation amplitude coefficient, For the respiratory phase, is the duration of a complete breathing cycle, is the baseline offset.

[0170] The model building in step S5 includes:

[0171] S51. Dealing with imbalanced distribution of clinical data by synthetic minority class oversampling technique;

[0172] S52, using Bayesian optimization to search for extreme gradient boosting tree hyperparameter combinations;

[0173] S53. Construct multi-task output layer:

[0174] Task 1: Probability of scoliosis progression ;

[0175] Task 2: Classification labeling of surgical indications .

[0176] It also includes algorithm adaptive optimization steps:

[0177] S61. Establish an algorithm library:

[0178]

[0179] Contains key point detection algorithms and prediction models;

[0180] S62, real-time monitoring of the area under the curve and F1 score of each algorithm on the validation set;

[0181] S63, when Less than -σ triggers algorithm switching:

[0182] Using the improved Osprey search algorithm to select the optimal alternative algorithm :

[0183]

[0184] in is a collection of algorithm libraries, is the clinical weight factor, is the area under the curve, is the F1 score.

[0185] Add a 3D reconstruction step after step S2:

[0186] S71. Construct a spine point cloud model based on the registration results;

[0187] S72. Generate topology optimized 3D mesh by Poisson surface reconstruction;

[0188] S73. Calculate the vertebral rotation AVR on the mesh model:

[0189]

[0190] in is the average vertebral rotation, is the number of vertebrae involved in the calculation, is the normal vector of the i-th vertebral cross section, is the reference plane normal vector.

[0191] In step S5, explainability processing is introduced:

[0192] S81, generating a key point detection heat map through gradient weighted class activation mapping;

[0193] S82. Use a local interpretable model-independent explanation algorithm to explain the extreme gradient boosting tree prediction results;

[0194] S83. Visualize the feature contribution ranking in the evaluation report.

[0195] Also includes cloud platform deployment steps:

[0196] S91. Use containerization technology to encapsulate the evaluation algorithm;

[0197] S92. Dynamically allocate computing resources through the container orchestration system.

[0198] S93. Implement hierarchical authority control for doctors, patients, and administrators based on open authorization protocols.

[0199] Example 1: Multimodal Registration Fusion and 3D Reconstruction

[0200] During system implementation, spinal X-rays, CT scans, and MRI scans of the patient's spine were acquired via a DICOM interface. Cross-modality registration was performed using the SIFT algorithm to extract bony landmarks of the vertebral body margins and pedicles from the CT images. Simultaneously, a modified CycleGAN network was used to map the T2-weighted MRI images to the CT coordinate system to generate pseudo-CT images. When the spatial registration module detected a 2.3mm misalignment between the CT and MRI planes of the sacral endplate, a re-registration process was automatically triggered: the affine transformation matrix was recalculated using the RANSAC algorithm, and the registration error was displayed as a heat map overlaid on the 3D model.

[0201] After registration, the point cloud processing unit fuses multimodal data to construct a three-dimensional mesh model of the spine, and generates a topologically optimized digital twin of the spine through Poisson surface reconstruction. The calculation accuracy of the vertebral rotation reaches the sub-millimeter level, significantly eliminating the anatomical structure distortion caused by a single imaging modality.

[0202] Example 2: Dynamic Quantification and Risk Prediction Closed Loop

[0203] During the dynamic evaluation stage, the system automatically segments the respiratory phase based on the 4D-CT sequence, detects the coronal vertebral angle points of each phase through the adaptively sampled HRNet network, and optimizes the sagittal plane key point sequence in combination with bidirectional LSTM. The parameter calculation unit applies a dynamic compensation function: the spinal flexion angle measurement value is corrected in real time according to the patient's individual respiratory cycle duration, eliminating the 6.1° pseudo-deviation caused by spinal extension in the inspiratory phase. The corrected parameters are input into the XGBoost multi-task model, and the SMOTE-enhanced clinical database is processed synchronously.

[0204] When the main spinal curvature angle of a 15-year-old female patient is detected to be 31° and the sagittal thoracolumbar turning point offset is identified, the system links the biomechanical simulation unit to calculate the intervertebral disc stress distribution, generate the probability of scoliosis progression and surgical indication warning, and simultaneously trigger the three-dimensional correction scheme simulation module to output the personalized brace mechanical parameters.

[0205] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0206] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent quantitative assessment system for adolescent scoliosis integrating multimodal imaging, characterized by: include: Multimodal data management module: An image acquisition unit equipped with a medical imaging communication protocol interface can acquire X-ray, CT, and MRI data; Structured database unit that stores clinical parameters and supports batch import and export; Cross-modal registration and fusion module: Affine transformation unit, which performs X-ray-CT anatomical alignment; GAN fusion unit, which realizes the fusion of spatial domain features of MRI and CT; Dynamic quantitative analysis module: Key point detection unit, integrating improved real-time multi-person pose estimation network model and nested U-shaped network; Parameter calculation unit, dynamically calculates spinal curvature angle, AVR and sagittal balance parameters; Risk prediction module: Extreme gradient boosting tree ensemble learning unit, outputting scoliosis progress probability; Interpretability processing unit, generating SHAP feature contribution graph; 3D reconstruction module: Point cloud processing unit, constructing a 3D mesh model of the spine; Deformation simulation unit, which visualizes spinal deformation under different loads; Report generation module: Multi-view fusion unit, integrating 3D models, heat maps and time series curves; The early warning trigger unit generates red, yellow, and blue warning signals according to the risk level; System management module: A permission control unit based on the role-based access control model distinguishes the operation permissions of doctors, patients, and administrators; Audit log unit, records data access and algorithm call behavior.

2. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 1, characterized in that: The following steps are involved: S1. Collecting multimodal medical imaging data and clinical parameters of the patient, wherein the multimodal medical imaging includes spinal X-rays, computed tomography scans, and magnetic resonance imaging, and the clinical parameters include gender, age, skeletal maturity index, spinal curvature angle, and scoliosis classification system classification; S2. Cross-modal registration and fusion processing of multimodal medical images: Affine transformation was used to align the coronal and sagittal anatomical structures of the X-ray and CT images; Fusion of MRI soft tissue information to CT bone structure coordinate system based on generative adversarial network; Generate standardized three-dimensional spinal fusion models; S3. Perform multi-plane keypoint detection on the fusion model: The improved real-time multi-person pose estimation model is used to detect vertebral corners and endplate landmarks in the coronal plane; A nested U-shaped network was used to identify the sacral tilt angle and thoracolumbar turning point in the sagittal plane; S4. Dynamic quantification of spinal deformation parameters: Calculate the major and minor spinal curvature angles based on the key points of the coronal plane: in are the normal vectors of the upper and lower vertebral endplates, respectively; Calculate the thoracic kyphosis angle and lumbar lordosis angle based on the key points in the sagittal plane; S5. Construct a multi-factor risk prediction model: The clinical parameters and deformation parameters are input into the extreme gradient boosting ensemble learning framework; Analyze the feature contribution by adding Shapley and explaining the value, and output the probability of scoliosis progression ; S6. Generate dynamic quantitative evaluation report: Fusion of 3D reconstruction model, key point heat map, and parameter time series comparison curve; Early warning signals are triggered based on the progression probability level.

3. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: The cross-modality registration in step S2 specifically includes: S21, performing edge enhancement and grayscale normalization on X-ray film; S22, extracting bony landmarks from CT images using scale-invariant feature transformation; S23. Calculate the projection transformation matrix between X-ray and CT by random sampling consensus algorithm : in is the affine transformation function, is the coordinate of the i-th bony landmark point in the CT image, is the coordinate of the i-th projection point matched in the X-ray film, is the number of matching feature point pairs; S24. Use a recurrent generative adversarial network to map MRI T2-weighted images to the CT coordinate system: in Generate adversarial networks for MRI to CT, is the original MRI image, is a synthetic pseudo CT image.

4. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: The key point detection in step S3 adopts a cascade optimization mechanism: S31, localize the vertebral region of interest through the spatial attention module; S32, Extracting coronal plane key point coordinates using a high-resolution network with adaptive sampling ; S33, Constructing vertebral topology constraints based on graph convolutional networks; S34. Optimizing sagittal plane key point sequences via bidirectional long short-term memory networks .

5. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: The parameter calculation of step S4 includes a dynamic compensation mechanism: S41, segmenting CT sequence frames according to respiratory phase; S42, calculate the change in spinal curvature angle at each phase ; S43. Establish deformation-respiration cycle correlation function: in is the deformation amplitude coefficient, For the respiratory phase, is the duration of a complete breathing cycle, is the baseline offset.

6. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: The model building in step S5 includes: S51. Dealing with imbalanced distribution of clinical data by synthetic minority class oversampling technique; S52, using Bayesian optimization to search for extreme gradient boosting tree hyperparameter combinations; S53. Construct multi-task output layer: Task 1: Probability of scoliosis progression ; Task 2: Classification labeling of surgical indications .

7. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: It also includes algorithm adaptive optimization steps: S61. Establish an algorithm library: Contains key point detection algorithms and prediction models; S62, real-time monitoring of the area under the curve and F1 score of each algorithm on the validation set; S63, when Less than -σ triggers algorithm switching: Using the improved Osprey search algorithm to select the optimal alternative algorithm : in is a collection of algorithm libraries, is the clinical weight factor, is the area under the curve, for Fraction.

8. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: Add a 3D reconstruction step after step S2: S71. Construct a spine point cloud model based on the registration results; S72. Generate topology optimized 3D mesh by Poisson surface reconstruction; S73. Calculate the vertebral rotation AVR on the mesh model: in is the average vertebral rotation, is the number of vertebrae involved in the calculation, is the normal vector of the i-th vertebral cross section, is the reference plane normal vector.

9. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 2, characterized in that: The step S5 introduces the explainability process: S81, generating a key point detection heat map through gradient weighted class activation mapping; S82. Use a local interpretable model-independent explanation algorithm to explain the extreme gradient boosting tree prediction results; S83. Visualize the feature contribution ranking in the evaluation report.

10. The intelligent quantitative assessment system for adolescent scoliosis integrated with multimodal imaging according to claim 1, characterized in that: Also includes cloud platform deployment steps: S91. Use containerization technology to encapsulate the evaluation algorithm; S92. Dynamically allocate computing resources through the container orchestration system. S93. Implement hierarchical authority control for doctors, patients, and administrators based on open authorization protocols.

Images