Construction method of knee joint osteophyte quantitative analysis model and computer equipment
By constructing a quantitative analysis model for knee osteophytes and using deep learning technology to automate the analysis of knee X-ray images, the problem of insufficient diagnostic accuracy and consistency in existing technologies has been solved, and efficient and accurate quantitative assessment of osteophytes has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LINGYU YINGNUO (BEIJING) TECH CO LTD
- Filing Date
- 2024-03-05
- Publication Date
- 2026-04-17
AI Technical Summary
Existing methods for analyzing knee osteophytes rely on manual interpretation, resulting in insufficient diagnostic accuracy and consistency, and the presence of human error.
A quantitative analysis model for knee osteophytes was constructed. This was achieved by acquiring and preprocessing knee X-ray images, dividing the knee into regions using a pre-trained detection model, training a corresponding quantitative analysis model for each region, and then using deep learning technology for automated analysis.
It enables automated quantitative analysis of knee osteophytes, reduces human error, improves analytical accuracy and consistency, reduces workload, and improves the efficiency of medical services.
Smart Images
Figure CN121883344A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical technology, and in particular to a method for constructing a quantitative analysis model of knee osteophytes and a computer device. Background Technology
[0002] Knee osteophytes are bony protrusions that form on the surface of the knee joint bones, usually caused by factors such as bone hyperplasia or arthritis. Osteophytes typically form on the surface of the bone at the joint margins and can be small, sharp protrusions or larger bone fragments. Furthermore, the number and distribution of osteophytes directly relate to the severity of knee osteophytes; quantitative analysis of osteophytes can be used to monitor disease progression and treatment effectiveness.
[0003] Currently, doctors or researchers grade osteophytes in images according to a set of predetermined standards, classifying them into different levels or degrees. This method is usually based on the doctor's or researcher's visual judgment and experience, and the scoring criteria may vary from research institution to research institution. This can lead to inconsistent assessments of the same patient's condition by doctors or researchers, making it difficult to guarantee the accuracy and consistency of diagnosis. With the development of computer technology, some computer-aided image analysis software is used for quantitative analysis of osteophytes. This method can measure the morphology and number of osteophytes through digital processing and calculation of images. However, this method still requires human intervention for interpretation and confirmation, which also introduces human error into osteophyte analysis, affecting accuracy and consistency. Summary of the Invention
[0004] This application provides a method and computer device for constructing a quantitative analysis model of knee osteophytes, aiming to solve the technical problem that existing osteophyte analysis methods cannot guarantee diagnostic accuracy and consistency.
[0005] Firstly, a method for constructing a quantitative analysis model for knee osteophytes includes:
[0006] Acquire knee joint X-ray images and preprocess the knee joint X-ray images;
[0007] Based on a pre-trained knee joint detection model, the knee joint X-ray image is divided into regions, and images of different regions in the knee joint X-ray image are obtained.
[0008] Using images of different regions in the knee joint X-ray, a corresponding quantitative analysis model for osteophytes is trained for each region.
[0009] Optionally, in the above scheme, the preprocessing of the knee joint X-ray image includes:
[0010] The knee joint X-ray image is finely adjusted to make the knee joint structure more prominent in the knee joint X-ray image;
[0011] The knee joint X-ray image is adjusted according to the preset window width and preset window level;
[0012] The knee joint X-ray image was converted into JPEG format to obtain a knee joint JPEG image;
[0013] The knee joint JPEG image was downsampled;
[0014] Improve the contrast of the knee joint JPEG image and remove noise from the knee joint JPEG image;
[0015] The size of the knee joint JPEG image is adjusted according to a preset size.
[0016] In the above scheme, optionally, the knee joint X-ray image is converted to JPEG format using Pydicom; the knee joint JPEG image is downsampled using bilinear interpolation; gamma correction is applied to improve the contrast of the knee joint JPEG image; and Gaussian blur is used to remove noise from the knee joint X-ray image.
[0017] In the above scheme, optionally, the knee joint detection model adopts the YOLO detection model, and the knee joint detection model selects the Dice loss function as the training objective of the model; when training the knee joint detection model, the Adam optimizer is selected, the learning rate parameter is set to 0.001, and the batch size is set to 8; during the training process of the knee joint detection model, in each iteration, the prediction result is calculated through forward propagation, and the weights of the knee joint detection model are updated through backpropagation.
[0018] Optionally, in the above scheme, the plurality of regions include the lateral femoral region, the medial femoral region, the lateral tibia region, and the medial tibia region.
[0019] Optionally, in the above scheme, the step of using images of different regions in the knee joint X-ray image to train a corresponding quantitative analysis model for each region includes:
[0020] The images of the lateral femoral region, medial femoral region, lateral tibia region, and medial tibia region in the X-ray images of the knee joint were standardized respectively.
[0021] Training datasets were divided into images of the lateral femoral region, medial femoral region, lateral tibia region, and medial tibia region, respectively.
[0022] Classifier models were trained using the corresponding training datasets of the images of the lateral femoral region, medial femoral region, lateral tibial region, and medial tibial region, respectively, to obtain quantitative analysis models of osteophytes in the lateral femoral region, medial femoral region, lateral tibial region, and medial tibial region. The osteophyte quantitative analysis models can output the severity category of osteophytes based on the input images of the corresponding regions.
[0023] In the above scheme, optionally, the classifier model adopts the EfficientNet model, and the loss function of the classifier model is the negative log-likelihood loss function; when training the classifier model, the optimizer uses RMSPropoptimizer, and the optimizer's decay is set to 0.9, momentum is set to 0.9, and weight_decay is set to 1e-5.
[0024] In a second aspect, a computer device loads a quantitative analysis model of osteophytes constructed using the method described in the first aspect, inputs the target knee joint X-ray image of the osteophytes to be analyzed into the quantitative analysis model, and obtains the quantitative analysis results of osteophytes in each region of the target knee joint X-ray image.
[0025] Optionally, in the above scheme, the step of inputting the acquired target knee joint X-ray image of the osteophyte to be analyzed into the osteophyte quantitative analysis model to obtain the osteophyte quantitative analysis results for each region in the target knee joint X-ray image includes:
[0026] The target knee joint X-ray image is preprocessed and divided into regions to obtain images of different regions in the target knee joint X-ray image.
[0027] The image of each region in the target knee joint X-ray is input into the corresponding osteophyte quantitative analysis model to obtain the probability prediction result of each category for each region. The category with the highest probability for each region is selected as the osteophyte severity category for each region and output.
[0028] In the above scheme, optionally, for each region in the X-ray image of the target knee joint, if the output is 0, it means that there are no osteophytes in the region; if the output is 1, it means that there are mild osteophytes in the region; if the output is 2, it means that there are severe osteophytes in the region.
[0029] Compared with the prior art, this application has at least the following beneficial effects:
[0030] This application provides a method for constructing a quantitative analysis model for knee osteophytes, including acquiring knee X-ray images, preprocessing the knee X-ray images, dividing the knee X-ray images into multiple regions based on a pre-trained knee joint detection model, acquiring images of different regions in the knee X-ray images, and training a corresponding quantitative analysis model for each region using the images of different regions in the knee X-ray images. After obtaining the quantitative analysis model for knee osteophytes using this method, the model can be used for quantitative analysis of knee osteophytes. This achieves automated quantitative analysis of knee osteophytes by using artificial intelligence and deep learning technologies to locate and identify different regions in knee X-ray images through model training, eliminating reliance on manual labor, reducing human error, lowering the risk of human error, improving the accuracy and consistency of analysis, and reducing the workload of manual measurement and analysis. Furthermore, automated quantitative assessment of osteophytes can reduce patient waiting time, improve the efficiency of medical services, and allow doctors to obtain key diagnostic information more quickly, enabling more timely and accurate treatment decisions. Attached Figure Description
[0031] Figure 1 A flowchart illustrating a method for constructing a quantitative analysis model of knee osteophytes according to an embodiment of this application;
[0032] Figure 2 This is a schematic diagram of the region division result of a knee joint X-ray image in one embodiment of this application;
[0033] Figure 3 This is a schematic diagram of the quantitative analysis results of the knee joint in one embodiment of this application;
[0034] Figure 4 This is a schematic diagram of the overall operation flow of a computer device provided in one embodiment of this application. Detailed Implementation
[0035] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0036] In the description of this application, unless otherwise stated, expressions such as "comprising," "including," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).
[0037] In this application embodiment, a method for constructing a quantitative analysis model of knee joint osteophytes is provided, such as... Figure 1 As shown, it includes:
[0038] S1, acquire knee joint X-ray images and preprocess the knee joint X-ray images.
[0039] Further, the knee joint X-ray image is preprocessed, including:
[0040] Fine-tuning of knee X-ray images makes the knee joint structure more prominent in the images;
[0041] Adjust the knee joint X-ray image according to the preset window width and preset window level;
[0042] Convert the knee joint X-ray image to JPEG format to obtain a knee joint JPEG image;
[0043] Downsampling of JPEG images of the knee joint;
[0044] Improve the contrast of JPEG images of the knee joint and remove noise from them;
[0045] Adjust the size of the knee joint JPEG image according to the preset size.
[0046] Specifically, the knee X-ray image was converted to JPEG format using Pydicom; the knee JPEG image was downsampled using bilinear interpolation; gamma correction was applied to improve the contrast of the knee JPEG image; and Gaussian blur was used to remove noise from the knee X-ray image.
[0047] In other words, step S1 acquires and preprocesses knee joint X-ray images; the process in step S1 includes:
[0048] S11. Preprocessing of knee X-ray images: The original DICOM image of the knee joint is finely adjusted to better highlight the knee joint structure. The adjustment is made according to the set window width and window level. Commonly used parameters are 2048 and 4096, which helps to highlight the details of the joint structure. The adjusted image is converted to JPEG format via Pydicom for subsequent processing and analysis. This gives you a JPEG image of the knee joint, which is convenient for subsequent processing and analysis.
[0049] S12. Knee Joint JPEG Image Preprocessing: Since X-ray images are typically large, bilinear interpolation can be used to downsample the image to reduce computational complexity, thereby reducing image size and noise. In bilinear interpolation, the output pixel value is calculated based on the weighted average of its surrounding 2×2 neighborhood pixels.
[0050] In addition, gamma correction can be applied to improve image contrast, Gaussian blur can be used to remove noise, and the image size can be adjusted to ensure that they are the same size for subsequent model training and analysis.
[0051] S2, based on a pre-trained knee joint detection model, divides the knee joint X-ray image into multiple regions and obtains images of different regions in the knee joint X-ray image.
[0052] Furthermore, the knee joint detection model adopts the YOLO detection model, and the Dice loss function is selected as the training objective of the model. When training the knee joint detection model, the Adam optimizer is selected, the learning rate parameter is set to 0.001, and the batch size is set to 8. During the training process of the knee joint detection model, in each iteration, the prediction result is calculated through forward propagation, and the weights of the knee joint detection model are updated through backpropagation.
[0053] These areas specifically include the lateral femoral region, the medial femoral region, the lateral tibia region, and the medial tibia region.
[0054] In other words, step S2 implements a YOLO-based detection model to segment the knee joint image into regions, obtaining the medial and lateral femoral and tibial regions of the knee joint. The region segmentation results of the knee joint X-ray image can be found in [reference needed]. Figure 2 The process in step S2 includes:
[0055] S21. Training the knee joint detection model: The YOLO detection model is used, and the Dice loss function is selected as the training objective. To train this model, the Adam optimizer can be selected, and the learning rate parameter is set to 0.001, while the batch size is set to 8 to improve training efficiency and speed.
[0056] S22. Train the YOLO detection model using images from the training set and labeled images as input. In each iteration, calculate the prediction result through forward propagation, and then update the model's weights through backpropagation. Make predictions using images from the validation set and labeled images, and calculate the Dice coefficient to evaluate the model's performance. Through multiple iterations, the model's architecture, hyperparameters, or training strategy can be adjusted to obtain the final YOLO detection model.
[0057] For example, the training process of the knee joint detection model can be as follows:
[0058] Prepare the training dataset: Collect a sufficient number of knee X-ray images and label the medial and lateral femoral and tibial regions of the knee joint in each image. Define the model architecture: Use the YOLO model as the knee joint detection model. This model includes components such as convolutional layers, pooling layers, and fully connected layers. The model can be optimized by adjusting hyperparameters such as the number of network layers and nodes. Configure training parameters: Select appropriate parameters such as loss function, optimizer, and learning rate to train the model; commonly used loss functions include cross-entropy loss and Dice loss. Train the model: Input the prepared dataset into the model for training; in each iteration, calculate the loss value between the model output and the labels, and update the model parameters through backpropagation; a suitable number of iterations can be set or early stopping can be used to avoid overfitting. Model evaluation and tuning: Evaluate the model using validation set data, and calculate metrics such as model precision and recall; if the model performs poorly, try adjusting the model architecture or training parameters and retrain. Model application: The trained knee joint detection model is used for region segmentation to obtain the medial and lateral femoral regions and the medial and lateral tibial regions of the knee joint for subsequent osteophyte analysis.
[0059] S3 utilizes images of different regions in knee X-rays to train a corresponding quantitative analysis model for each region.
[0060] Furthermore, using images of different regions in knee X-rays, a corresponding quantitative analysis model for osteophytes was trained for each region, including:
[0061] The images of the lateral femoral region, medial femoral region, lateral tibia region, and medial tibia region in the X-ray images of the knee joint were standardized respectively.
[0062] The training dataset was divided into images of the lateral femur region, the medial femur region, the lateral tibia region, and the medial tibia region, respectively.
[0063] Classifier models were trained using corresponding training datasets of images of the lateral femur, medial femur, lateral tibia, and medial tibia, respectively, to obtain quantitative analysis models of osteophytes in the lateral femur, medial femur, lateral tibia, and medial tibia regions. The osteophyte quantitative analysis models can output the severity category of osteophytes based on the input images of the corresponding regions.
[0064] The classifier model uses the EfficientNet model, and the loss function of the classifier model is the negative log-likelihood loss function. When training the classifier model, the optimizer uses the RMSProp optimizer, and the optimizer's decay is set to 0.9, momentum is set to 0.9, and weight_decay is set to 1e-5.
[0065] In other words, step S3 trains models separately for the medial and lateral femoral regions and the medial and lateral tibial regions of the knee joint, obtaining osteophyte analysis models for each region; the process in step S3 includes:
[0066] S31. Standardize the image based on the obtained lateral femoral region to ensure that the image size and pixel value range match the input requirements of the EfficientNet model. Iteratively train the EfficientNet model using the training set data. The optimizer used is RMSProp optimizer, where decay = 0.9, momentum = 0.9, weight_decay is 1e-5, and the loss function is negative log-likelihood loss function. Finally, a classification model that can quantify the lateral femoral region is obtained.
[0067] S32. Using the same method as above, train classifier models for the medial femur and medial and lateral tibia images respectively. At this time, four classifier models for the medial and lateral femur and the medial and lateral tibia are obtained.
[0068] The above method is used to construct a quantitative analysis model for osteophytes. After obtaining the target knee joint X-ray image of the osteophyte to be analyzed, the target knee joint X-ray image is input into the quantitative analysis model for osteophytes, and the quantitative analysis results of osteophytes in each region of the target knee joint X-ray image can be obtained.
[0069] The process involves inputting the target knee joint X-ray image of the osteophytes to be analyzed into the osteophyte quantitative analysis model to obtain the quantitative analysis results of osteophytes in each region of the target knee joint X-ray image, including:
[0070] The target knee joint X-ray image is preprocessed and divided into regions. The target knee joint X-ray image is divided into multiple regions, and images of different regions in the target knee joint X-ray image are obtained.
[0071] The image of each region in the X-ray image of the target knee joint is input into the corresponding osteophyte quantitative analysis model to obtain the probability prediction result of each category for each region. The category with the highest probability for each region is selected as the osteophyte severity category for each region and output.
[0072] In other words, after obtaining the knee osteophyte quantitative analysis model using the construction method provided in this application embodiment, the model can be used for quantitative analysis of knee osteophytes. This enables automated quantitative analysis of knee osteophytes by utilizing artificial intelligence and deep learning technologies to train the model to locate and identify the medial and lateral sides of the femur and tibia. Using this knee osteophyte quantitative analysis model can reduce human error and improve the accuracy and consistency of the analysis.
[0073] This quantitative analysis model for knee osteophytes automates the calculation process for evaluating knee osteophytes using a high-precision classifier, reducing the workload of manual measurement and analysis. This innovation significantly improves computational efficiency and reduces the risk of human error. Through the classification model, composite scaling techniques balance the network's width, depth, and resolution, resulting in improved network performance while reducing computational complexity, leading to faster training and inference speeds and greater hardware resource efficiency.
[0074] In one embodiment, a computer device is provided, which loads the osteophyte quantitative analysis model constructed by the method for constructing the knee osteophyte quantitative analysis model provided in the above embodiment, inputs the target knee joint X-ray image of the osteophyte to be analyzed into the osteophyte quantitative analysis model, and obtains the osteophyte quantitative analysis results of each region in the target knee joint X-ray image.
[0075] Furthermore, the acquired X-ray image of the target knee joint to be analyzed is input into the osteophyte quantitative analysis model to obtain the quantitative analysis results of osteophytes in each region of the target knee joint X-ray image, including:
[0076] The target knee joint X-ray image is preprocessed and divided into regions. The target knee joint X-ray image is divided into multiple regions, and images of different regions in the target knee joint X-ray image are obtained.
[0077] The image of each region in the X-ray image of the target knee joint is input into the corresponding osteophyte quantitative analysis model to obtain the probability prediction result of each category for each region. The category with the highest probability for each region is selected as the osteophyte severity category for each region and output.
[0078] For each region in the X-ray image of the target knee joint, an output of 0 indicates that there are no osteophytes in that region; an output of 1 indicates that there are mild osteophytes in that region; and an output of 2 indicates that there are severe osteophytes in that region.
[0079] In other words, the above steps are used for quantitative analysis and description of osteophytes in four regions, characterizing the severity of osteophytes in the knee joint. The results of the quantitative analysis of the knee joint can be found in [reference needed]. Figure 3The above steps include the following processes:
[0080] S41. Input the extracted images of the medial and lateral femur and medial and lateral tibia into a separately trained classifier, output the probability prediction result of each category, select the category with the highest probability, such as output 0 with the highest probability, which means there is no osteophyte; output 1 with the highest probability, which means there is mild osteophyte on this side; output 2 with the highest probability, which means there is severe osteophyte on this side, and use this as the diagnosis result.
[0081] S42. Repeat step S41 to complete the quantitative assessment of osteophytes on the medial side of the femur and the medial and lateral sides of the tibia. This will allow you to obtain the severity of osteophytes in each part of the entire knee joint.
[0082] In other words, such as Figure 4 As shown, the overall operation flow of the computer device includes: S1: acquiring X-ray images of the knee joint and preprocessing the images; S2: dividing the knee joint images into regions based on the YOLO detection model to obtain the medial and lateral femoral and tibial regions of the knee joint; S3: training models separately for the medial and lateral femoral and tibial regions of the knee joint to obtain osteophyte analysis models for each region; S4: quantitatively analyzing and describing the osteophytes in the four regions to characterize the severity of osteophytes in the knee joint.
[0083] Specifically, the computer device can be a laptop, a computer, or a dedicated quantitative analysis device for osteophytes. This device may include a high-performance processing unit to accelerate image processing; a large-capacity storage device to store large amounts of medical image data and model parameters; and an interface for input / output and user interaction, such as a display screen, keyboard, mouse, or touchscreen. The device also has network connectivity for data exchange and remote diagnosis with medical information systems. Furthermore, the device may include auxiliary functions such as automated data processing workflows and result visualization to improve the efficiency and accuracy of medical image analysis.
[0084] The computer device provided in this application embodiment can utilize artificial intelligence and deep learning technologies to train models to locate and identify the medial and lateral sides of the femur and tibia, thereby achieving automated quantitative analysis of knee joint osteophytes. This system can reduce human error and improve the accuracy and consistency of the analysis.
[0085] The computer device provided in this application aims to automate the calculation process of quantitative assessment of knee osteophytes by using a high-precision classifier, thereby reducing the workload of manual measurement and analysis. This innovation greatly improves computational efficiency and reduces the risk of human error. Through the classification model, the composite scaling technique balances the three dimensions of network width, depth, and resolution, resulting in better network performance, reduced computational complexity, faster training and inference speeds, and greater savings in hardware resources.
[0086] Furthermore, automated quantitative assessment of osteophytes reduces patient wait times and improves the efficiency of healthcare services. Doctors can obtain crucial diagnostic information more quickly, enabling them to make more timely and accurate treatment decisions.
[0087] Therefore, the application of the computer device provided in this application embodiment has significant clinical implications, offering a rapid, accurate, and reliable method for the diagnosis and treatment of knee joint diseases. It is expected to improve patient treatment outcomes and provide doctors with more comprehensive image analysis tools, thereby enabling precise clinical decision-making.
[0088] The computer equipment provided in this application embodiment enables quantitative analysis of osteophytes on knee X-ray images. This equipment can effectively screen and assess the severity of knee joint lesions, improving the diagnostic accuracy of affected areas. This provides doctors with better guidance and allows for more accurate patient treatment plans. Furthermore, the system can also be used for quantitative analysis of knee joint pathological changes, providing a valuable tool for clinical research and medical scientific research.
[0089] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A method of constructing a model for quantifying osteophytes of a knee joint, characterized by, include: Acquire knee joint X-ray images and preprocess the knee joint X-ray images; Based on a pre-trained knee joint detection model, the knee joint X-ray image is divided into regions, and images of different regions in the knee joint X-ray image are obtained. Using images of different regions in the knee joint X-ray, a corresponding quantitative analysis model for osteophytes is trained for each region.
2. The method of constructing a model for quantifying osteophytes of a knee joint according to claim 1, wherein, The preprocessing of the knee joint X-ray image includes: The knee joint X-ray image is finely adjusted to make the knee joint structure more prominent in the knee joint X-ray image; The knee joint X-ray image is adjusted according to the preset window width and preset window level; The knee joint X-ray image was converted into JPEG format to obtain a knee joint JPEG image; The knee joint JPEG image was downsampled; Improve the contrast of the knee joint JPEG image and remove noise from the knee joint JPEG image; The size of the knee joint JPEG image is adjusted according to a preset size.
3. The method for constructing a quantitative analysis model for knee osteophytes according to claim 2, characterized in that, Specifically, the knee joint X-ray image is converted to JPEG format using Pydicom; the knee joint JPEG image is downsampled using bilinear interpolation; gamma correction is applied to improve the contrast of the knee joint JPEG image; and Gaussian blur is used to remove noise from the knee joint X-ray image.
4. The method for constructing a quantitative analysis model for knee osteophytes according to claim 1, characterized in that, The knee joint detection model adopts the YOLO detection model, and the Dice loss function is selected as the training objective of the model. When training the knee joint detection model, the Adam optimizer is selected, the learning rate parameter is set to 0.001, and the batch size is set to 8. During the training process of the knee joint detection model, in each iteration, the prediction result is calculated through forward propagation, and the weights of the knee joint detection model are updated through backpropagation.
5. The method for constructing a quantitative analysis model for knee osteophytes according to claim 1, characterized in that, The multiple regions include the lateral femoral region, the medial femoral region, the lateral tibia region, and the medial tibia region.
6. The method for constructing a quantitative analysis model for knee osteophytes according to claim 5, characterized in that, The step of using images of different regions in the knee joint X-ray to train a corresponding quantitative analysis model for each region includes: The images of the lateral femoral region, medial femoral region, lateral tibia region, and medial tibia region in the X-ray images of the knee joint were standardized respectively. Training datasets were divided into images of the lateral femoral region, medial femoral region, lateral tibia region, and medial tibia region, respectively. Classifier models were trained using the corresponding training datasets of the images of the lateral femoral region, medial femoral region, lateral tibial region, and medial tibial region, respectively, to obtain quantitative analysis models of osteophytes in the lateral femoral region, medial femoral region, lateral tibial region, and medial tibial region. The osteophyte quantitative analysis models can output the severity category of osteophytes based on the input images of the corresponding regions.
7. The method for constructing a quantitative analysis model for knee osteophytes according to claim 6, characterized in that, The classifier model uses the EfficientNet model, and the loss function of the classifier model is the negative log-likelihood loss function. When training the classifier model, the optimizer uses the RMSProp optimizer, and the optimizer's decay is set to 0.9, momentum is set to 0.9, and weight_decay is set to 1e-5.
8. A computer device, characterized in that, The osteophyte quantitative analysis model is constructed using the method described in claim 1. The target knee joint X-ray image of the osteophyte to be analyzed is input into the osteophyte quantitative analysis model to obtain the osteophyte quantitative analysis results for each region in the target knee joint X-ray image.
9. The computer device according to claim 8, characterized in that, The step of inputting the target knee joint X-ray image of the osteophyte to be analyzed into the osteophyte quantitative analysis model to obtain the osteophyte quantitative analysis results for each region in the target knee joint X-ray image includes: The target knee joint X-ray image is preprocessed and divided into regions to obtain images of different regions in the target knee joint X-ray image. The image of each region in the X-ray image of the target knee joint is input into the corresponding osteophyte quantitative analysis model to obtain the probability prediction result of each category for each region. The category with the highest probability for each region is selected as the osteophyte severity category for each region and output.
10. The computer device according to claim 9, characterized in that, For each region in the X-ray image of the target knee joint, an output of 0 indicates that there are no osteophytes in that region; an output of 1 indicates that there are mild osteophytes in that region; and an output of 2 indicates that there are severe osteophytes in that region.