Spine sagittal plane form intelligent classification method based on multi-scale feature extraction

Abstract

The invention discloses a spine sagittal plane form intelligent classification method based on multi-scale feature extraction. The method comprises the steps of image preprocessing and feature extraction, key feature screening, data balance and optimization and random forest classification. According to the method, automatic and high-precision classification of spine-pelvis sagittal plane X-ray images of adults can be realized, scientific decision support is provided for clinical diagnosis, large-scale screening, personalized prevention and surgical plan formulation of spine diseases, and spine morphological biomechanical characteristics are deeply represented through multi-scale characteristics; through combination of rigorous feature screening and model optimization strategies, the problems that a traditional method is high in subjectivity and single in feature are effectively solved.

Description

Technical Field

[0001] This invention relates to an intelligent classification method for the sagittal morphology of the spine, specifically an intelligent classification method for the sagittal morphology of the spine based on multi-scale feature extraction, belonging to the interdisciplinary field of medical image processing and artificial intelligence. Background Technology

[0002] As the core supporting structure of the human body, the accurate classification of the spine's sagittal plane morphology is the key basis for assessing spinal health, diagnosing spinal diseases (such as intervertebral disc herniation, scoliosis, lumbar spondylolisthesis, etc.) and formulating treatment plans. Traditional classification methods mainly rely on doctors manually measuring several static anatomical parameters on X-ray films, such as the pelvic angle of incidence, sacral tilt angle, and lumbar lordosis angle, and classifying them according to empirical rules.

[0003] Traditional classification of spinal sagittal morphology relies on manual measurement of key parameters, which has significant limitations. 1. Large subjective error; existing literature: P. Roussouly, S. Gollogly, E. Berthonnaud, et al., Classification of the normal variation in the sagittal alignment of the human lumbar spine and pelvis in the standing position. Spine (Phila Pa 1976), 2005, 30(3):346-353. The proposed classification method is based only on a few static parameters such as sacral tilt angle (SS) and lumbar lordosis morphology, without considering the dynamic changes of the overall curve of the spine and regional coordination. In the state of spinal degeneration and other diseases, the reduction of lumbar lordosis will lead to the distortion of SS value and the significant decrease in classification accuracy. 2. Narrow scope of application; existing literature: CSLee, SSChung, KCKang, et al., Normal patterns of sagittal alignment of the spine in young adults radiological analysis in a Korean population. Spine (Phila Pa 1976), 2011, 36(25): E1648-1654. The proposed grading system is only for the single disease of lumbar spinal stenosis, relies on high-resolution MRI images, and is limited by equipment conditions, so it cannot be extended to the classification of the overall sagittal morphology of the spine. 3. Low level of intelligence; existing literature: S. Mohanty, F.M. Hassan, L. G.L. Enke, et al., Machine learning clustering of adult spinal deformity patients identifies four prognostic phenotypes: a multicenter prospective cohort analysis with singlesurgeon extralateral validation. Spine J, 2024, 24(6): 1095-1108; Although it reduces human intervention, the selection of kernel function and parameter tuning are complex, and the generalization ability is weak when processing high-dimensional spinal imaging data, making it difficult to adapt to large-scale and diverse clinical data; 4. Poor population adaptability: Most classification systems are based on data from foreign populations, ignoring the differences in spinal sagittal morphological development among different races and regions, and lacking a specific classification model for Chinese adults. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent classification method for spinal sagittal morphology based on multi-scale feature extraction to solve at least one of the above-mentioned technical problems. This method can automatically and accurately process spinal sagittal images and extract multi-dimensional features that reflect the biomechanical essence from the images. It also has strong generalization ability and robustness against class imbalance, thus overcoming the technical defects of existing technologies such as large subjective error, incomplete feature representation, weak model generalization ability, and poor population adaptability.

[0005] This invention achieves the above objective through the following technical solution: an intelligent classification method for spinal sagittal morphology based on multi-scale feature extraction, comprising the following steps:

[0006] Step 1: Image Preprocessing and Feature Extraction: For sagittal X-ray images of the spine and pelvis, a multi-task deep learning method using an existing hollow spatial pyramid pooling module and cross-task attention mechanism is employed to detect key points at the four vertebrae of each vertebra. After solving for the centroid coordinates of the vertebrae using the vertex coordinates, a 12th-order polynomial is used to fit the centroid coordinate sequence to generate a fitting curve. Based on this curve, multi-scale features that can characterize the dynamic changes in the morphology of the spine are extracted to form an initial feature set.

[0007] Step 2: Key Feature Screening: Based on statistical correlation analysis, a subset of key features that are strongly correlated with the spine classification results is screened from the initial feature set;

[0008] Step 3, Data Balancing and Optimization: Divide the key feature subset into training and testing sets, perform class balancing, and optimize the hyperparameters of the random forest;

[0009] Step 4: Random Forest Classification: Based on the key feature subset, construct an optimized random forest model to classify the sagittal morphology of the spine, and output the classification results and performance indicators.

[0010] As a further aspect of the present invention: in step one, the key points include 24 key points of the spinal vertebrae; the centroid coordinates of the spinal vertebrae are calculated using the polygon centroid formula; the 12th-order polynomial fitting employs the least squares method, with the sum of squared errors... The objective is to minimize the coefficients of the 12th-order polynomial, and the coefficients are solved using the Vandermonde matrix. The ordinate of the feature point is... Predict the ordinate of the fitted curve to ensure a good fit between the fitted curve and the centroid coordinates;

[0011] Multiscale features include the first derivative, second derivative, and curvature of the centroids of the 24 vertebral bodies corresponding to the fitted curves.

[0012] As a further aspect of the present invention: the first derivative reflects the vertebral displacement rate, the second derivative reflects the degree of joint curvature, and the curvature reflects the degree of spinal load concentration.

[0013] As a further aspect of the present invention: In step two, the statistical correlation analysis method is Spearman rank correlation analysis, and features with an absolute value of correlation coefficient not less than 0.6 are selected; the key feature subset includes L4 first derivative, L3 first derivative, T12 second derivative, L2 first derivative, L1 second derivative, L3 second derivative, L2 second derivative, T11 second derivative, T10 first derivative, T9 first derivative, L3 curvature, and T11 curvature.

[0014] As a further aspect of this invention: the Spearman rank correlation analysis calculation formula is as follows:

[0015] ;

[0016] in, The rank correlation coefficient has a value range of [-1, 1]. Let n be the rank difference between paired samples, and n be the sample size.

[0017] As a further aspect of the present invention: In step three, the class balancing process adopts the SMOTE-ENN algorithm, wherein the SMOTE oversampling ratio is 1:3, and the ENN algorithm removes noisy samples based on 3 nearest neighbor samples; the hyperparameters of the random forest are performed by the GridSearchCV method for grid search, and the hyperparameters of the random forest include at least one of the following: number of decision trees (100 / 200 / 300), maximum depth of decision trees (None / 10 / 20 / 30), maximum number of features (sqrt, log2, None), minimum number of samples for node split (2 / 5 / 10), minimum number of samples for leaf nodes (1 / 2 / 4), autonomous sampling (True, False), and class weight (balanced, None).

[0018] As a further step in this invention, the optimal combination of hyperparameters for the random forest classification model is as follows: number of decision trees n_estimators=200, maximum number of features max_features=sqrt, minimum number of samples for node splits min_samples_split=2, bootstrap=False, class weight class_weight = balanced, maximum depth of decision trees max_depth = None, and minimum number of samples for leaf nodes min_samples_leaf=2.

[0019] As a further aspect of the present invention: In step four, the sagittal plane morphology of the spine is divided into four categories based on the position of the horizontal lumbar spine line relative to the lumbar vertebral body and intervertebral space:

[0020] Type I, HLL is located below the L4 / 5 intervertebral space; Type II, HLL passes through the L4 vertebral body and the L4 / 5 intervertebral space; Type III, HLL passes through the L3 vertebral body and the L3 / 4 intervertebral space; Type IV, HLL is located above the superior endplate of L3.

[0021] As a further aspect of this invention, the training process of the random forest model employs a weighted cross-entropy loss function.

[0022] A multi-scale feature extraction-based intelligent classification system for spinal sagittal morphology includes an image preprocessing module, a feature extraction module, a feature filtering module, a data optimization module, and a random forest classification module that are sequentially connected and work together to complete the classification of spinal sagittal morphology.

[0023] The image preprocessing module and feature extraction module are used to perform image preprocessing and feature extraction in step one; the feature selection module is used to perform key feature selection in step two; the data optimization module is used to perform data balancing and optimization in step three; and the random forest classification module is used to perform random forest classification in step four.

[0024] The beneficial effects of this invention are:

[0025] 1) This invention extracts the coordinates of multiple vertebral centroids and performs high-order polynomial fitting, and then calculates the first derivative, second derivative and curvature. It innovatively constructs a multi-scale feature set from both geometric and biomechanical perspectives, which can comprehensively and deeply characterize the dynamic changes, bending morphology and load distribution characteristics of the spinal curve. It overcomes the limitations of traditional methods that rely on a few static parameters, and lays a rich and physically meaningful feature foundation for accurate classification.

[0026] 2) This invention employs rigorous statistical correlation analysis to automatically select a subset of key features from high-dimensional initial features, effectively eliminating redundant information, reducing model complexity and enhancing interpretability, while avoiding the risk of overfitting. It addresses the common class imbalance problem in clinical data by combining a hybrid algorithm of SMOTE oversampling and ENN denoising, and finely tunes the hyperparameters of the random forest model using grid search, significantly improving the model's ability to identify minority classes and its overall generalization performance.

[0027] 3) Based on the optimized random forest model, this invention achieves accurate four-class classification of the sagittal morphology of the spine in Chinese adults, with high classification accuracy and good stability. The method is fully automated, which greatly reduces human intervention and subjective error. It is specifically trained and optimized for Chinese population data and has excellent population adaptability. It has important clinical application value for the auxiliary diagnosis, large-scale screening and personalized treatment planning of spinal diseases.

[0028] 4) This invention can realize automated and high-precision classification of sagittal X-ray images of the spine and pelvis in adults, providing scientific decision support for clinical diagnosis, large-scale screening, personalized prevention and surgical planning of spinal diseases. Furthermore, by deeply characterizing the morphological and biomechanical properties of the spine through multi-scale features, combined with rigorous feature selection and model optimization strategies, it effectively overcomes the problems of strong subjectivity and single features in traditional methods. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall process structure of the present invention;

[0030] Figure 2 This is a schematic diagram illustrating the calculation of the centroid of the vertebral body according to the present invention;

[0031] Figure 3 This is a schematic diagram of the structure of the 12th-order polynomial curve fitting result of the present invention;

[0032] Figure 4 This is a schematic diagram illustrating the physical meaning of the first derivative of this invention;

[0033] Figure 5This is a schematic diagram illustrating the physical meaning of the second derivative in this invention;

[0034] Figure 6 This is a schematic diagram illustrating the physical meaning of curvature in this invention;

[0035] Figure 7 This is a schematic diagram of the training process structure of the random forest model of the present invention;

[0036] Figure 8 This is a schematic diagram of the four types of spines in the improved Lee classification of this invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example 1, as Figures 1 to 8 As shown, this embodiment provides an intelligent classification method for spinal sagittal morphology based on multi-scale feature extraction. This intelligent classification method for spinal sagittal morphology includes the following steps:

[0039] Step 1: Image Preprocessing and Feature Extraction: For sagittal X-ray images of the spine and pelvis, a multi-task deep learning method using an existing hollow spatial pyramid pooling module and cross-task attention mechanism is employed to detect key points at the four vertebrae of each vertebra. After solving for the centroid coordinates of the vertebrae using the vertex coordinates, a 12th-order polynomial is used to fit the centroid coordinate sequence to generate a fitting curve. Based on this curve, multi-scale features that characterize the dynamic changes in the morphology of the spine are extracted to form an initial feature set.

[0040] Step 2: Key Feature Screening: Based on statistical correlation analysis, a subset of key features that are strongly correlated with the spine classification results is screened from the initial feature set;

[0041] Step 3: Data balancing and optimization. The key feature subset is divided into training and testing sets, and class balancing and optimization of the random forest hyperparameters are performed.

[0042] Step 4: Random Forest Classification. Based on the key feature subset, an optimized random forest model is constructed to classify the sagittal morphology of the spine, and the classification results and performance indicators are output.

[0043] Example 2, in addition to all the technical features included in Example 1, also includes:

[0044] In step one: the key points include 24 key points of the spinal vertebrae;

[0045] like Figure 2 As shown, the centroid coordinates of the vertebral bodies are calculated using the polygon centroid formula.

[0046] like Figure 3 As shown, the 12th-order polynomial fitting uses the least squares method, with the sum of squared errors as the basis of measurement. The objective is to minimize the coefficients of the 12th-order polynomial, and the coefficients are solved using the Vandermonde matrix. The ordinate of the feature point is... Predict the ordinate of the fitted curve to ensure a good fit between the fitted curve and the centroid coordinates;

[0047] like Figure 4 , Figure 5 and Figure 6 As shown, the multi-scale features include the first derivative, second derivative, and curvature of the centroid of the 24 vertebral bodies corresponding to the fitted curve. The first derivative reflects the displacement rate of the vertebral bodies, the second derivative reflects the degree of joint bending, and the curvature reflects the degree of load concentration on the spine.

[0048] In step two: the statistical correlation analysis method is Spearman rank correlation analysis, and features with an absolute correlation coefficient of not less than 0.6 are selected; the key feature subset includes L4 first derivative, L3 first derivative, T12 second derivative, L2 first derivative, L1 second derivative, L3 second derivative, L2 second derivative, T11 second derivative, T10 first derivative, T9 first derivative, L3 curvature, and T11 curvature.

[0049] Furthermore, the formula for Spearman's rank correlation analysis is as follows:

[0050] ;

[0051] in, is the rank correlation coefficient (value range [-1,1]). Let n be the rank difference between paired samples, and n be the sample size.

[0052] In step three, the class balancing process uses the SMOTE-ENN algorithm, where the SMOTE oversampling ratio is 1:3 and the ENN algorithm removes noisy samples based on the three nearest neighbor samples.

[0053] The hyperparameters of the random forest are obtained by using the GridSearchCV method for grid search. The hyperparameters of the random forest include at least one of the following: number of decision trees (100 / 200 / 300), maximum depth of decision trees (None / 10 / 20 / 30), maximum number of features (sqrt, log2, None), minimum number of samples for node split (2 / 5 / 10), minimum number of samples for leaf nodes (1 / 2 / 4), autonomous sampling (True, False), and class weights (balanced, None).

[0054] In step four, the hyperparameter combination of the random forest classification model is as follows: number of decision trees n_estimators=200, maximum number of features max_features=sqrt, minimum number of samples for node splits min_samples_split=2, bootstrap=False, class weight=balanced, maximum depth of decision trees max_depth = None, minimum number of samples for leaf nodes min_samples_leaf=2.

[0055] Based on the position of the horizontal lumbar spine line relative to the lumbar vertebral bodies and intervertebral spaces, the sagittal plane morphology of the spine is divided into four categories:

[0056] Type I (HLL is located below the L4 / 5 intervertebral space);

[0057] Type II (HLL passes through the L4 vertebral body and L4 / 5 intervertebral space);

[0058] Type III (HLL passes through the L3 vertebral body and L3 / 4 intervertebral space);

[0059] Type IV (HLL is located above the L3 upper endplate).

[0060] like Figure 7 As shown, the complete process of data preprocessing, SMOTE-ENN balancing, GridSearchCV hyperparameter optimization, and cross-validation is demonstrated. The training process of the random forest model adopts the weighted cross-entropy loss function and assigns higher weights to the minority class samples of Type I and Type IV.

[0061] Example 3: This example provides a spinal sagittal morphology intelligent classification system based on multi-scale feature extraction, used to implement the spinal sagittal morphology intelligent classification method in Example 1. The spinal sagittal morphology intelligent classification system includes an image preprocessing module, a feature extraction module, a feature filtering module, a data optimization module, and a random forest classification module that are sequentially connected and work together to complete the spinal sagittal morphology classification.

[0062] The image preprocessing module and feature extraction module are used to perform image preprocessing and feature extraction; the feature selection module is used to perform key feature selection; the data optimization module is used to perform data balancing and optimization; and the random forest classification module is used to perform random forest classification.

[0063] Example 4: An intelligent classification method for spinal sagittal morphology based on multi-scale feature extraction. This method utilizes four core modules working collaboratively: image preprocessing and feature extraction, key feature selection, data balancing and optimization, and random forest classification. The process is as follows:

[0064] First, image preprocessing and feature extraction: The sagittal X-ray images of the spine and pelvis are preprocessed to extract the key points of the four vertebral vertebrae and calculate the centroid of the vertebral body. Multi-scale features (first derivative, second derivative, curvature) are generated by fitting a 12th-order polynomial curve.

[0065] The image preprocessing and feature extraction module specifically includes:

[0066] 1) Image preprocessing: Receive sagittal X-ray images of the adult spine-pelvis and extract the coordinates of four key points.

[0067] 2) Keypoint detection and centroid calculation: Utilizing existing hollow space pyramid pooling modules and cross-task attention mechanisms, a multi-task deep learning method is used to detect keypoints at the four vertebrae of 24 vertebrae in the spinal column (C1-C7, T1-T12, L1-L5). The centroid coordinates of each vertebra (quadrilateral structure) are calculated using the polygon centroid formula, serving as feature points for curve fitting. The formula is as follows:

[0068] ;

[0069] in, Let be the area of ​​the quadrilateral of the cone. The coordinates of the vertices of the pyramidal quadrilateral are given.

[0070] 3) Multi-scale feature calculation: Minimize the sum of squared errors for 24 feature points using the least squares method. ; The ordinate of the feature point is... To predict the ordinate of the fitted curve, a 12th-order polynomial curve fitting is performed.

[0071] 4) Multi-scale feature parameters were calculated based on the fitted curve, generating a total of 72 initial features, including: a First derivative:

[0072] It reflects the rate of change of the position of key points in the spine. Its physical meaning is the dynamic mechanical transmission characteristics of the vertebral body. The larger the absolute value, the higher the relative displacement rate of adjacent vertebral bodies.

[0073] b Second derivative:

[0074] It reflects the degree of curvature and concavity / convexity of the spinal joints. The second derivative is >0 in the convex region and <0 in the convex region. The larger the absolute value, the more significant the curvature.

[0075] c Curvature:

[0076] It reflects the degree to which the spine deviates from a straight line and is directly related to the mechanical load-bearing capacity of the spine. The greater the curvature, the more concentrated the local load.

[0077] Second, key feature screening: Based on Spearman rank correlation analysis, feature parameters that are strongly correlated with spine type are screened and redundant information is eliminated.

[0078] The key features were selected based on Spearman rank correlation analysis. The specific steps are as follows:

[0079] 1) Correlation calculation:

[0080] The Spearman rank correlation coefficients between the 72 initial features and the spine type (Type I-Type IV based on the improved Lee classification) are calculated using the following formula:

[0081] ;

[0082] in, is the rank correlation coefficient (value range [-1,1]). For the rank difference of paired samples, For the sample size (in this study) ).

[0083] 2) Feature Filtering:

[0084] The screening threshold was set as follows: the absolute value of the correlation coefficient ≥ 0.6, and 12 key features were ultimately retained, as shown in Table 1.

[0085] Table 1 shows the 12 key features retained after feature selection.

[0086] Third, data balancing and optimization: The SMOTE-ENN algorithm is used to handle the data class imbalance problem, and the model hyperparameters are optimized in combination with GridSearchCV.

[0087] The data balancing and optimization module specifically includes:

[0088] 1) Data partitioning and class balancing: 439 samples were stratified into a training set (351 samples) and a test set (88 samples) using an 8:2 ratio to ensure consistent class distribution between the two datasets (Type I 20 samples, Type II 211 samples, Type III 171 samples, Type IV 37 samples). The SMOTE algorithm was used to oversample the minority class samples (Type I and Type IV) in the training set (sampling ratio 1:3), and then the ENN algorithm was used to remove noise samples introduced by oversampling (samples that are different from the three nearest neighbor samples were removed), finally resulting in 390 training samples.

[0089] 2) Model hyperparameter optimization: The GridSearchCV method will be used to traverse the hyperparameter combinations of the random forest, and the hyperparameter search range will be set.

[0090] Number of decision trees (n_estimators): 100, 200, 300;

[0091] Maximum number of features (max_features): sqrt, log2;

[0092] Minimum number of samples for node split (min_samples_split): 2, 5;

[0093] Minimum number of samples per leaf node (min_samples_leaf): 1, 2;

[0094] Maximum tree depth (max_depth): None, 10, 20.

[0095] 3) Using the average F1 score of five-fold cross-validation as the evaluation metric, determine the optimal hyperparameter combination: n_estimators=200, max_features=sqrt, min_samples_split=2, min_samples_leaf=1, max_depth=None, bootstrap=False, class_weight=balanced.

[0096] Fourth, Random Forest Classification: Construct an optimized random forest model to achieve accurate classification of spine types (Type I-Type IV) and output classification results and performance metrics.

[0097] Model construction and spine type classification: Using 12 key features as input, a random forest classification model is constructed: 200 decision trees are generated based on Bootstrap sampling, and the Gini coefficient is used as the node splitting criterion (to measure the purity of the dataset). The final classification result is determined by majority voting.

[0098] The dataset is divided into training and testing sets, with 80% of the data used for training and 20% used for testing, to perform random forest classification prediction. Selected parameter metrics are applied to the random forest classification prediction model.

[0099] The definitions and sample proportions of the four spinal types are as follows:

[0100] Type I: The horizontal lumbar line (HLL) is located below the L4 / 5 intervertebral space, with PI=34.7°, SS=20.2°, and LL=26.8°, accounting for 3.6%;

[0101] Type II: HLL penetrates the L4 vertebral body and L4 / 5 intervertebral space, PI=38.1°, SS=30.0°, accounting for 39.3%;

[0102] Type III: HLL penetrates the L3 vertebral body and L3 / 4 intervertebral space, PI=46.9°, SS=36.8°, accounting for 45.7%;

[0103] Type IV: HLL is located above the upper endplate of L3, LL=47.0°, and thoracic kyphosis (TK) is significantly increased, accounting for 11.4%.

[0104] Performance validation and results output: Hierarchical five-fold cross-validation is used to evaluate the model's generalization ability, and the classification results (spine type) and performance indicators (accuracy, precision, recall, F1 score) are output to ensure that the model's accuracy on the test set is ≥99% and the average accuracy of cross-validation is ≥96.5%.

[0105] Example 5: A method for intelligent classification of spinal sagittal morphology based on multi-scale feature extraction, specifically including:

[0106] Data preparation: Sagittal X-ray images of the spine and pelvis of 439 Han Chinese adults (aged 18-49, male-to-female ratio 1:1.37) were selected and were exempted from ethical review by Xuzhou Central Hospital (IRB approval number: XZCH-2024-032).

[0107] Feature extraction: Extract 30 feature points according to the above steps, fit a 12th-order polynomial curve, calculate 72 initial features, and select 12 key features;

[0108] Data optimization: The training set contains 390 samples after processing with SMOTE-ENN, and the optimal hyperparameters are determined using GridSearchCV.

[0109] Model validation: The validation results on the test set are shown in Table 2. The overall accuracy reached 99%, and the average accuracy of the five-fold cross-validation was 96.46%, which meets clinical needs.

[0110] Table 2 shows the validation results of the test set.

[0111] Sagittal X-ray images of the adult spine and pelvis were acquired. After preprocessing, 24 key points of the vertebral bodies and 6 key points of the sacrum and coccyx were detected. The centroid coordinates were calculated using the polygon centroid formula. A 12th-order polynomial was used to fit curves to the centroids, and the first derivative (reflecting the displacement rate of the key points), second derivative (reflecting the degree of joint flexion), and curvature (reflecting the degree of load concentration) of the fitted curves were extracted to generate 72 initial features. 12 key features (absolute correlation coefficient ≥ 0.6) were selected based on Spearman rank correlation analysis. The SMOTE-ENN algorithm was used to handle data class imbalance, and the hyperparameters of the random forest were optimized by combining GridSearchCV (optimal combination: n_estimators=200, max_features=sqrt, etc.) to construct a classification model. The final output is the Type I-Type IV spine types based on the improved Lee classification. Figure 8 As shown, it simultaneously provides performance metrics such as accuracy and F1 score.

[0112] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0113] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent classification of spinal sagittal plane morphology based on multi-scale feature extraction, characterized in that, Includes the following steps: Step 1: Image Preprocessing and Feature Extraction: For sagittal X-ray images of the spine and pelvis, a multi-task deep learning method using an existing hollow spatial pyramid pooling module and cross-task attention mechanism is employed to detect key points at the four vertebrae of each vertebra. After solving for the centroid coordinates of the vertebrae using the vertex coordinates, a 12th-order polynomial is used to fit the centroid coordinate sequence to generate a fitting curve. Based on this curve, multi-scale features that can characterize the dynamic changes in the morphology of the spine are extracted to form an initial feature set. Step 2: Key Feature Screening: Based on statistical correlation analysis, a subset of key features that are strongly correlated with the spine classification results is screened from the initial feature set; Step 3, Data Balancing and Optimization: Divide the key feature subset into training and testing sets, perform class balancing, and optimize the hyperparameters of the random forest. Step 4: Random Forest Classification: Based on the aforementioned key feature subset, construct an optimized random forest model to classify the sagittal morphology of the spine, and output the classification results and performance indicators.

2. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 1, characterized in that, In step one: The key points include 24 key points of the spinal vertebrae; The centroid coordinates of the spinal vertebral body are calculated using the polygon centroid formula. The 12th-order polynomial fitting employs the least squares method, using the sum of squared errors. The objective is to minimize the coefficients of the 12th-order polynomial, and the coefficients are solved using the Vandermonde matrix. The ordinate of the feature point is... Predict the ordinate of the fitted curve; The multi-scale features include the first derivative, second derivative, and curvature of the centroids of the 24 vertebral bodies corresponding to the fitted curves.

3. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 2, characterized in that: The first derivative reflects the displacement rate of key points in the vertebral body; the second derivative reflects the degree of joint curvature; and the curvature reflects the degree of load concentration on the spine.

4. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 1, characterized in that, In step two: The statistical correlation analysis method is Spearman rank correlation analysis, and features with an absolute correlation coefficient of not less than 0.6 are selected. The key feature subset includes the first derivative of L4, the first derivative of L3, the second derivative of T12, the first derivative of L2, the second derivative of L1, the second derivative of L3, the second derivative of L2, the second derivative of T11, the first derivative of T10, the first derivative of T9, the curvature of L3, and the curvature of T11.

5. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 4, characterized in that: The formula for calculating the Spearman rank correlation analysis is as follows: ； in, The rank correlation coefficient has a value range of [-1, 1]. Let n be the rank difference between paired samples, and n be the sample size.

6. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 1, characterized in that, In step three: The category balancing process uses the SMOTE-ENN algorithm, where the SMOTE oversampling ratio is 1:3 and the ENN algorithm removes noisy samples based on the three nearest neighbor samples. The hyperparameters of the random forest are obtained by performing grid search using the GridSearchCV method, and the hyperparameters of the random forest include at least one of the following: number of decision trees, maximum depth of decision trees, maximum number of features, minimum number of samples for node splits, minimum number of samples for leaf nodes, autonomous sampling, and class weights.

7. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 6, characterized in that, In step four: The optimized random forest classification model has the following hyperparameters: number of decision trees n_estimators=200, maximum number of features max_features=sqrt, minimum number of samples for node splits min_samples_split=2, bootstrap=False, class weight=balanced, maximum depth of decision trees max_depth=None, and minimum number of samples for leaf nodes min_samples_leaf=2.

8. The intelligent classification method for sagittal plane morphology of the spine according to claim 1, characterized in that, In step four: The sagittal morphology of the spine is classified into four categories based on the position of the horizontal lumbar spine line relative to the lumbar vertebral body and intervertebral space: Type I, HLL is located below the L4 / 5 intervertebral space; Type II, HLL passes through the L4 vertebral body and the L4 / 5 intervertebral space; Type III, HLL passes through the L3 vertebral body and the L3 / 4 intervertebral space; Type IV, HLL is located above the L3 upper endplate.

9. The intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction according to claim 8, characterized in that: The training process of the random forest model uses a weighted cross-entropy loss function.

10. A system for implementing the intelligent classification method for spinal sagittal plane morphology based on multi-scale feature extraction as claimed in claim 1, characterized in that: The intelligent classification system for spinal sagittal morphology includes an image preprocessing module, a feature extraction module, a feature filtering module, a data optimization module, and a random forest classification module, which are connected in sequence and work together to complete the classification of spinal sagittal morphology. The image preprocessing module and the feature extraction module are used to perform image preprocessing and feature extraction in step one; The feature filtering module is used to perform the key feature filtering in step two. The data optimization module is used to perform data balancing and optimization in step three. The random forest classification module is used to perform the random forest classification in step four.

Images