Femoral head necrosis scoring system and method based on CT image

By using a CT image-based femoral head necrosis scoring system, and combining data preprocessing and a three-plane fusion module with a Video Swin Transformer network, the problems of strong subjectivity, low efficiency, and lack of quantitative indicators in existing technologies are solved, thus achieving accurate quantitative assessment of femoral head necrosis risk and improving the robustness of the model.

CN121861345APending Publication Date: 2026-04-14TIANJIN HOSPITAL +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511928766.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-19
Publication Date
2026-04-14

AI Technical Summary

Technical Problem

Existing technologies for assessing avascular necrosis of the femoral head are highly subjective, inefficient, and lack quantitative indicators. Existing deep learning models cannot fully utilize information from different planes of CT images, resulting in insufficient assessment accuracy.

Method used

A CT-based femoral head necrosis scoring system was adopted. Through data preprocessing, enhancement, and three-plane fusion modules, combined with the Video Swin Transformer network, a continuous risk score for femoral head necrosis was extracted. The three-plane feature fusion and quantitative scoring design were then utilized.

Benefits of technology

It enables quantitative assessment of the risk of avascular necrosis of the femoral head, improves the accuracy of the assessment and the robustness of the model, and provides accurate risk reference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121861345A_ABST
    Figure CN121861345A_ABST
Patent Text Reader

Abstract

The invention discloses a femoral head necrosis scoring system and method based on a CT image, and belongs to the technical field of image processing. By collecting preoperative hip joint CT data of a femoral neck fracture patient, through preprocessing operations of unifying resolution, calibrating an area of interest and the like, and in combination with multiple data enhancement means, sample diversity is improved; the method comprises the following steps: firstly, extracting a plurality of images, then converting the images into sagittal plane views, axial plane views and coronal plane views, fusing features, extracting multi-scale spatial features based on Video Swin Transformer, outputting continuous risk scores between 0 and 1, and finally training the model by adopting a corresponding loss function and an optimizer until convergence; according to the femoral head necrosis scoring system and method based on the CT image, the problems that a traditional method subjectively depends on experience and lacks quantitative indexes are solved, the scoring accuracy and efficiency are improved, and an objective basis is provided for clinical risk assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a scoring system and method for femoral head necrosis based on CT images. Background Technology

[0002] Avascular necrosis of the femoral head is a major complication following internal fixation of femoral neck fractures, leading to loss of joint function and severely impacting patients' quality of life. Currently, the prediction of avascular necrosis of the femoral head mainly relies on the experience and judgment of clinicians, with imaging analysis primarily performed through manual observation of X-ray, CT, or MRI images. This manual assessment method is highly dependent on the doctor's professional background and practical experience, significantly influenced by subjective factors. Different doctors may arrive at significantly different conclusions, making it difficult to ensure consistency in assessment. Furthermore, the manual analysis process is cumbersome, time-consuming, and inefficient, failing to meet the needs of rapid clinical evaluation.

[0003] With the development of deep learning technology, computer vision technology has been increasingly applied to the prediction of avascular necrosis of the femoral head. Existing methods primarily employ deep learning models such as convolutional neural networks and Transformers to automatically extract image features and output classification results, reducing manual intervention to some extent. However, these methods mostly focus on classification tasks, only determining whether a patient is at risk of necrosis, lacking quantitative indicators of risk severity, and thus failing to provide accurate references for clinical treatment planning and prognostic assessment. Furthermore, existing models do not fully utilize information from different planes of CT images, and the comprehensiveness of feature extraction is insufficient, leading to room for improvement in prediction accuracy. Therefore, a technical solution capable of quantifying necrosis risk and improving assessment accuracy is urgently needed to address the shortcomings of existing technologies. Summary of the Invention

[0004] The purpose of this invention is to provide a scoring system and method for femoral head necrosis based on CT images, which solves the problems of subjectivity, inefficiency and lack of quantitative indicators in the existing technology.

[0005] To achieve the above objectives, the present invention provides a CT image-based scoring system for femoral head necrosis, comprising: Data acquisition module: used to collect preoperative hip CT scan data of patients with femoral neck fractures; Data preprocessing module: used to perform resolution unification and normalization on CT scan data; Data augmentation module: Used to perform data augmentation operations on preprocessed data, increase sample diversity and improve model robustness; Three-plane fusion module: used to fuse features from the sagittal, axial, and coronal planes of CT images; Feature extraction module: Based on Video Swin Transformer backbone network encoding CT image features; Necrosis scoring module: used to output the encoded features as a continuous risk score for femoral head necrosis.

[0006] Preferably, the data preprocessing module adjusts the thickness of CT data slices by downsampling, identifies the region of interest covering the femoral head and its underlying cortical region, and normalizes the data resolution to a preset size.

[0007] Preferably, data augmentation operations include random rotation, translation, brightness and contrast adjustment, affine shearing, and Gaussian noise injection.

[0008] Preferably, in the three-plane fusion module, the sagittal, axial, and coronal planes are obtained by spatially rearranging the CT images along their depth, height, and width directions.

[0009] This invention also provides a scoring method for femoral head necrosis based on CT images, comprising the following steps: Step 1: Collect preoperative CT image data of patients with femoral neck fractures from the hospital database; Step 2: Preprocess CT data, unify resolution and slice thickness, identify regions of interest and normalize them; Step 3: Perform data enhancement operations on the CT data; Step 4: Through spatial rearrangement, the CT images are arranged along their depth, height, and width directions to obtain CT views in the sagittal, axial, and coronal planes, respectively. After 3D block embedding layer mapping, feature stitching, and convolutional fusion, the final fused features are obtained. The specific process is as follows: S41. Represent consecutive slices from CT scans as... , where D, H, and W represent depth (number of slices), height, and width, respectively, and the channel dimension is 1; S42. Perform a replacement operation on the CT scan, Convert to sagittal and coronal plane representations: and ; S43. Input the three planes into a shared 3D block embedding layer, and map each view to a common token space of dimension C through 3D convolution, as shown in the following expression: ; ; ; in, , , p is the size of the patch. This represents the token feature after the axial plane image is mapped by the embedding layer. This represents the token features after the sagittal image is mapped by the embedding layer. This represents the token features of the coronal image after being mapped by the embedding layer; S44, will and Reshape as and with splicing along the channel axis and Reshape as and with The splicing is performed along the channel axis, as shown in the following expression: ; in, Indicates splicing, This represents the feature formed by splicing three-plane token features along the channel axis; S45. Use a 1×1×1 convolution to fuse the features of the three planes to obtain a fused feature containing complementary anatomical information. The expression is as follows: ; in, Indicates an embedding operation; Step 5: Using Video Swin Transformer as the backbone network for feature extraction, the fusion features obtained in Step 4 are hierarchically encoded by stacking multiple stages of the W-MSA and SW-MSA modules of the backbone network to obtain a high-dimensional feature map for prediction. Step 6: Perform global pooling and flattening operations on the high-dimensional feature map obtained in Step 5 to convert it into a one-dimensional feature vector. Map this feature to a low-dimensional space through a linear layer, and use the Sigmoid function to convert the output into a probability value between 0 and 1, which is used as the continuous risk score for femoral head necrosis. Step 7: Use the binary cross-entropy loss function as the model optimization objective, calculate the loss between the model's predicted results and the actual results, and update the model's parameters through backpropagation using the AdamW optimizer, training until the preset convergence condition is met.

[0010] Preferably, the backbone network consists of four consecutive stages, each stage consisting of multiple Video Swin Transformer blocks. The specific content within each Video Swin Transformer block is as follows: ; ; ; ; in, This indicates a window-based multi-head self-attention module. This represents a multi-head self-attention module for a translation window. Representation layer normalization, It is a two-layer feedforward neural network. This represents the input feature map of the current block. This indicates the multi-head self-attention module passing through the window. and the intermediate feature map obtained after concatenating with the residuals of the input feature map. Indicates the first Intermediate output feature map after submodule This indicates a multi-head self-attention module after translation window. and the intermediate feature map obtained after residual connection, This indicates that the Video Swin Transformer block is passed through the second... The final output feature map after processing by the submodule.

[0011] Preferably, step 6 is performed as follows: S61. Perform global pooling and flattening operations on the high-dimensional feature map obtained in step 5 to obtain a one-dimensional feature vector. The feature vector is then mapped through a linear layer to obtain intermediate variables. The expression is as follows: ; In the formula, This represents the weight vector of the linear layer. For the bias term of the linear layer, and All are learnable parameters; S62. Using the Sigmoid function to handle intermediate variables The probability of risk of femoral head necrosis is mapped to the following expression: ; in, The model outputs a risk score for avascular necrosis of the femoral head.

[0012] Preferably, the binary cross-entropy loss function in step 7 The expression is as follows: ; In the formula, Output the necrosis risk probability for the model of the nth sample. This represents the true label of the nth sample.

[0013] Preferably, the expression for updating the deep learning model parameters through backpropagation using the AdamW optimizer is as follows: ; In the formula, This represents the gradient of the loss function. Represents the set of model parameters. Indicates the number of iterations.

[0014] Therefore, the present invention employs the above-mentioned CT image-based scoring system and method for femoral head necrosis, which has the following beneficial effects: (1) It outputs continuous risk scores, realizing the quantitative assessment of the risk of femoral head necrosis, and providing a more accurate risk reference compared with traditional classification methods; (2) By integrating the three-plane features of CT images and making full use of anatomical information from different angles, the accuracy of the scoring has been significantly improved; (3) The robustness of the model was improved through data preprocessing and enhancement, and it was adapted to the differences in CT scan data from different hospitals.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of a CT-based scoring method for femoral head necrosis according to the present invention. Figure 2 This is a flowchart illustrating the feature fusion process in an embodiment of the present invention. Figure 3 This is a diagram illustrating the feature extraction process in an embodiment of the present invention. Detailed Implementation

[0017] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0018] A CT-based scoring system for avascular necrosis of the femoral head, comprising: Data acquisition module: used to collect preoperative hip CT scan data of patients with femoral neck fractures; The data preprocessing module is used to unify and normalize the resolution of CT scan data. In addition, the data preprocessing module adjusts the thickness of CT data slices by downsampling, identifies the region of interest covering the femoral head and its underlying cortical area, and unifies the data resolution to the preset size after normalization. Data augmentation module: Used to perform data augmentation operations on preprocessed data to increase sample diversity and improve model robustness; data augmentation operations include random rotation, translation, brightness and contrast adjustment, affine shearing, and Gaussian noise injection. The three-plane fusion module is used to fuse features from the sagittal, axial, and coronal planes of CT images. The sagittal, axial, and coronal planes are obtained by spatially rearranging the CT images along their depth, height, and width directions. Feature extraction module: Based on Video Swin Transformer backbone network encoding CT image features; Necrosis scoring module: used to output the encoded features as a continuous risk score for femoral head necrosis.

[0019] Please see Figure 1 A scoring method for femoral head necrosis based on CT images includes the following steps: Step 1: Collect preoperative CT image data of patients with femoral neck fractures from the hospital database; Step 2: Preprocess CT data, unify resolution and slice thickness, label regions of interest and normalize; data with different slice thicknesses are downsampled to ensure consistent slice thickness; the region of interest for each CT scan is labeled according to the anatomical structure of the femoral head to ensure coverage of the femoral head and its underlying cortical bone; normalize all CT data to a uniform resolution for easier model processing. Step 3: Perform data augmentation operations on the CT data, including random rotation, translation, brightness and contrast adjustment, affine shearing, and Gaussian noise injection, to ensure the diversity of training data and the robustness of the model. Step 4: Through spatial rearrangement, the CT images are arranged along their depth, height, and width directions to obtain sagittal, axial, and coronal CT views, respectively. After 3D block embedding layer mapping, feature stitching, and convolutional fusion, the final fused features are obtained, such as... Figure 2 As shown, the specific process is as follows: S41. Represent consecutive slices from CT scans as... , where D, H, and W represent depth (number of slices), height, and width, respectively, and the channel dimension is 1; S42. Perform a replacement operation on the CT scan, Convert to sagittal and coronal plane representations: and ; S43. Input the three planes into a shared 3D block embedding layer, and map each view to a common token space of dimension C through 3D convolution, as shown in the following expression: ; ; ; in, , , p is the size of the patch. This represents the token feature after the axial plane image is mapped by the embedding layer. This represents the token features after the sagittal image is mapped by the embedding layer. This represents the token features of the coronal image after being mapped by the embedding layer; S44, will and Reshape as and with splicing along the channel axis and Reshape as and with The splicing is performed along the channel axis, as shown in the following expression: ; in, Indicates splicing, This represents the feature formed by splicing three-plane token features along the channel axis; S45. Use a 1×1×1 convolution to fuse the features of the three planes to obtain a fused feature containing complementary anatomical information. The expression is as follows: ; in, Indicates an embedding operation; Step 5: Using the Video Swin Transformer as the backbone network for feature extraction, and constructing it through multiple stacked stages of the W-MSA and SW-MSA modules, the fused features obtained in Step 4 are subjected to hierarchical feature encoding to obtain a high-dimensional feature map for prediction, such as... Figure 3As shown, a hierarchical window self-attention mechanism is utilized to achieve efficient multi-scale spatial modeling. This mechanism reduces the high computational cost of global self-attention computation by calculating self-attention within local windows, while preserving important local anatomical information. The backbone network consists of four consecutive stages, each consisting of multiple Video Swin Transformer blocks. The specific content within each Video Swin Transformer block is as follows: ; ; ; ; in, This indicates a window-based multi-head self-attention module. This represents a multi-head self-attention module for a translation window. Representation layer normalization, It is a two-layer feedforward neural network. This represents the input feature map of the current block. This indicates the multi-head self-attention module passing through the window. and the intermediate feature map obtained after concatenating with the residuals of the input feature map. Indicates the first Intermediate output feature map after submodule This indicates a multi-head self-attention module after translation window. and the intermediate feature map obtained after residual connection, This indicates that the Video Swin Transformer block is passed through the second... The final output feature map after processing by the submodule; In this process, the input feature map is first segmented into multiple non-overlapping fixed-size local windows, and multi-head self-attention is independently computed within each window; assuming each window contains There are 10 tokens; within each window, each token is linearly mapped to form a vector of query Q, key K, and value V; the self-attention within this window is calculated as follows: ; Where d is the feature dimension, and T represents the matrix transpose operation; To enhance interaction between windows First, translate the feature map along the depth, height, and width axes. Then, the translated feature map is segmented into non-overlapping parts. Window, and compute self-attention independently within each window; Step 6: Perform global pooling and flattening operations on the high-dimensional feature map obtained in Step 5 to convert it into a one-dimensional feature vector. Then, map this feature vector to a low-dimensional space using a linear layer, and use the Sigmoid function to convert the output into a probability value between 0 and 1, which is then used as a continuous risk score for avascular necrosis of the femoral head. The specific process is as follows: S61. Perform global pooling and flattening operations on the high-dimensional feature map obtained in step 5 to obtain a one-dimensional feature vector. The feature vector is then mapped through a linear layer to obtain intermediate variables. The expression is as follows: ; In the formula, This represents the weight vector of the linear layer. For the bias term of the linear layer, and All are learnable parameters; S62. Using the Sigmoid function to handle intermediate variables The probability of risk of femoral head necrosis is mapped to the following expression: ; in, The model outputs a risk score for avascular necrosis of the femoral head.

[0020] Step 7: Use the binary cross-entropy loss function as the model optimization objective (the core model corresponding to the binary cross-entropy loss function as the optimization objective is a femoral head necrosis risk scoring model built based on the Video Swin Transformer). Calculate the loss between the model's predicted results and the actual results, and update the model's parameters through backpropagation using the AdamW optimizer, training until the preset convergence condition is met. The binary cross-entropy loss function... The expression is as follows: ; In the formula, Output the necrosis risk probability for the model of the nth sample. This represents the true label of the nth sample.

[0021] The expression for updating the parameters of a deep learning model via backpropagation using the AdamW optimizer is as follows: ; In the formula, This represents the gradient of the loss function. Represents the set of model parameters. This indicates the number of iterations; through multiple iterations of optimization, the model stops training when the validation set metrics no longer improve significantly or when the preset number of training rounds is reached.

[0022] Example This invention uses a retrospective dataset for performance validation. This dataset includes preoperative CT images of 441 patients with femoral neck fractures, of whom 90 were diagnosed with avascular necrosis of the femoral head (ONFH) and 351 were not. To ensure the objectivity of model training, tuning, and evaluation, the dataset was randomly divided into training, validation, and test sets in a 7:1:2 ratio, used for model parameter learning, hyperparameter tuning, and final performance testing, respectively.

[0023] To fully verify the superiority of the model in this invention, three representative 3D deep learning models were selected as benchmarks, covering mainstream technical paradigms in the field of volumetric data analysis: 1) 3D ResNet-18 (3D convolutional network paradigm); 2) Video Vision Transformer (ViViT, global self-attention paradigm); 3) Video Swin Transformer (hierarchical local attention paradigm). All models were evaluated on the same test set for predicting femoral head necrosis. Core evaluation metrics included precision, recall, F1 score, accuracy, and area under the curve (AUC).

[0024] Experimental results show that the model proposed in this invention significantly outperforms the comparison model in all core metrics: precision 0.7308, recall 0.6786, F1 score 0.7037, accuracy 0.7143, and AUC 0.7488. Specific performance comparison data for each model are shown in Table 1. Table 1

[0025] The comparison results show that the model of this invention, which adds a three-plane fusion module and a necrosis scoring module to the Video Swin Transformer, significantly outperforms other comparative models in terms of precision, accuracy, and AUC. This fully demonstrates the effectiveness of the three-plane feature fusion and quantitative scoring design, as well as the accuracy and reliability of the model in the femoral head necrosis risk assessment task.

[0026] Therefore, this invention employs the aforementioned CT image-based femoral head necrosis scoring system and method. It collects preoperative hip CT scan data from patients with femoral neck fractures, performs preprocessing such as slice thickness standardization, region of interest labeling, and normalization, and then enriches the samples through various data enhancement operations such as random rotation and translation. Subsequently, the CT images are converted into sagittal, axial, and coronal plane views, and comprehensive volumetric features are obtained through feature mapping, stitching, and fusion. The fused features are input into a Video Swin Transformer network, which uses a hierarchical window self-attention mechanism to extract multi-scale spatial features. Finally, a continuous risk score between 0 and 1 is output through linear transformation and probability transformation. The model is trained to convergence using the corresponding loss function and optimizer. This scheme, through three-plane feature fusion and quantitative scoring design, solves the problems of subjectivity, inefficiency, and lack of quantitative indicators in traditional methods. Experimental verification shows that the model performs excellently on key evaluation indicators, providing an objective and accurate risk assessment tool for femoral head necrosis in clinical practice.

[0027] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A scoring system for femoral head necrosis based on CT images, characterized in that, include: Data acquisition module: used to collect preoperative hip CT scan data of patients with femoral neck fractures; Data preprocessing module: used to perform resolution unification and normalization on CT scan data; Data augmentation module: Used to perform data augmentation operations on preprocessed data; Three-plane fusion module: used to fuse features from the sagittal, axial, and coronal planes of CT images; Feature extraction module: Based on Video Swin Transformer backbone network encoding CT image features; Necrosis scoring module: used to output the encoded features as a continuous risk score for femoral head necrosis.

2. The femoral head necrosis scoring system based on CT images according to claim 1, characterized in that: The data preprocessing module adjusts the thickness of CT data slices by downsampling, identifies the region of interest covering the femoral head and its underlying cortical bone, and normalizes the data resolution to a preset size.

3. The femoral head necrosis scoring system based on CT images according to claim 2, characterized in that: Data augmentation operations include random rotation, translation, brightness and contrast adjustment, affine shearing, and Gaussian noise injection.

4. The femoral head necrosis scoring system based on CT images according to claim 3, characterized in that: In the three-plane fusion module, the sagittal, axial, and coronal planes are obtained by spatially rearranging CT images along their depth, height, and width directions.

5. A method for scoring femoral head necrosis based on CT images, using the femoral head necrosis scoring system based on CT images as described in any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Collect preoperative CT image data of patients with femoral neck fractures from the hospital database; Step 2: Preprocess CT data, unify resolution and slice thickness, identify regions of interest and normalize them; Step 3: Perform data enhancement operations on the CT data; Step 4: Through spatial rearrangement, the CT images are arranged along their depth, height, and width directions to obtain CT views in the sagittal, axial, and coronal planes, respectively. After 3D block embedding layer mapping, feature stitching, and convolutional fusion, the final fused features are obtained. The specific process is as follows: S41. Represent consecutive slices from CT scans as... Where D, H, and W represent depth, height, and width, respectively, and the channel dimension is 1; S42. Perform a replacement operation on the CT scan, Convert to sagittal and coronal plane representations: and ; S43. Input the three planes into a shared 3D block embedding layer, and map each view to a common token space of dimension C through 3D convolution, as shown in the following expression: ; ; ; in, , , p is the size of the patch. This represents the token feature after the axial plane image is mapped by the embedding layer. This represents the token features after the sagittal image is mapped by the embedding layer. This represents the token feature after the coronal image is mapped by the embedding layer; S44, will and Reshape as and with splicing along the channel axis and Reshape as and with The splicing is performed along the channel axis, as shown in the following expression: ; in, Indicates splicing, This represents the feature formed by splicing three-plane token features along the channel axis; S45. Use a 1×1×1 convolution to fuse the features of the three planes to obtain a fused feature containing complementary anatomical information. The expression is as follows: ; in, Indicates an embedding operation; Step 5: Using Video Swin Transformer as the backbone network for feature extraction, the fusion features obtained in Step 4 are hierarchically encoded by stacking multiple stages of the W-MSA and SW-MSA modules of the backbone network to obtain a high-dimensional feature map for prediction. Step 6: Perform global pooling and flattening operations on the high-dimensional feature map obtained in Step 5 to convert it into a one-dimensional feature vector. Map this feature to a low-dimensional space through a linear layer, and use the Sigmoid function to convert the output into a probability value between 0 and 1, which is used as the continuous risk score for femoral head necrosis. Step 7: Use the binary cross-entropy loss function as the model optimization objective, calculate the loss between the model's predicted results and the actual results, and update the model's parameters through backpropagation using the AdamW optimizer, training until the preset convergence condition is met.

6. The method for scoring femoral head necrosis based on CT images according to claim 5, characterized in that: The backbone network consists of four consecutive stages, each of which comprises multiple VideoSwin Transformer blocks. The specific content within each VideoSwin Transformer block is as follows: ; ; ; ; in, This indicates a window-based multi-head self-attention module. This represents a multi-head self-attention module for a translation window. Representation layer normalization, It is a two-layer feedforward neural network. This represents the input feature map of the current block. This indicates the multi-head self-attention module passing through the window. and the intermediate feature map obtained after concatenating with the residuals of the input feature map. Indicates the first Intermediate output feature map after submodule This indicates a multi-head self-attention module after translation window. and the intermediate feature map obtained after residual connection, This indicates that the VideoSwin Transformer block is passed through the second... The final output feature map after processing by the submodule.

7. A scoring method for femoral head necrosis based on CT images according to claim 6, characterized in that, Step 6 is as follows: S61. Perform global pooling and flattening operations on the high-dimensional feature map obtained in step 5 to obtain a one-dimensional feature vector. The feature vector is then mapped through a linear layer to obtain intermediate variables. The expression is as follows: ; In the formula, This represents the weight vector of the linear layer. For the bias term of the linear layer, and All are learnable parameters; S62. Using the Sigmoid function to handle intermediate variables The probability of risk of femoral head necrosis is mapped to the following expression: ; in, The model outputs a risk score for avascular necrosis of the femoral head.

8. A scoring method for femoral head necrosis based on CT images according to claim 7, characterized in that, Step 7 Binary cross-entropy loss function The expression is as follows: ; In the formula, Output the necrosis risk probability for the model of the nth sample. This represents the true label of the nth sample.

9. A scoring method for femoral head necrosis based on CT images according to claim 8, characterized in that, The expression for updating the parameters of a deep learning model via backpropagation using the AdamW optimizer is as follows: ; In the formula, This represents the gradient of the loss function. Represents the set of model parameters. Indicates the number of iterations.