A proximal humerus osteoporosis automatic evaluation system and evaluation method based on shoulder joint X-ray images
Patent Information
- Application Number
- CN202411806685.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2026-10-20
AI Technical Summary
然而，这种方法成本高、设备要求严格，许多医院尤其是基层医疗机构难以配备专门的设备进行对应的定量检测
[0045]1.本发明通过深度学习技术自动从肩关节X线图像中提取骨质疏松相关特征，可以有效减少人工操作的主观性和误差，提高评估的精确性和一致性。本发明的系统不仅加快了评估速度，还允许大规模的骨质疏松筛查，提高了临床工作效率。
Smart Images
Figure BDA0005179421060000021 
Figure BDA0005179421060000022 
Figure BDA0005179421060000051
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, specifically to an automatic assessment system and method for proximal humeral osteoporosis based on shoulder joint X-ray images. Background Technology
[0002] Osteoporosis is a systemic bone disease characterized by decreased bone mass, damage to bone microstructure, increased bone fragility, and susceptibility to fractures. It is particularly prevalent in the elderly, where aging causes an imbalance in bone remodeling, leading to an increased bone resorption / bone formation ratio and progressive bone loss. Furthermore, age-related osteoporosis is associated with estrogen deficiency, reactive oxygen species accumulation, vitamin D deficiency, and chronic negative calcium balance; these factors work together to reduce bone formation and increase bone resorption.
[0003] Traditional osteoporosis diagnosis relies on quantitative bone mineral density (BMD) measurements, particularly dual-energy X-ray absorptiometry (DXA). However, this method is costly and requires sophisticated equipment, making it difficult for many hospitals, especially primary care facilities, to equip themselves with the necessary equipment for quantitative testing. Furthermore, DXA-measured BMD typically needs to be converted to T-scores for diagnosis, a complex process demanding a high level of expertise from the operators. In many clinical settings, quantitative BMD measurement is difficult to implement due to equipment limitations, cost issues, and the demands of specialized procedures.
[0004] Therefore, developing an automated osteoporosis assessment system and method based on shoulder joint X-ray images is of great clinical significance and has broad application prospects, enabling osteoporosis assessment using conventional X-ray images without relying on expensive equipment. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide an automatic assessment system and method for proximal humeral osteoporosis based on shoulder joint X-ray images.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0007] An automatic assessment system for proximal humeral osteoporosis based on shoulder joint X-ray images includes a data acquisition module, an image preprocessing module, a feature extraction module, an osteoporosis assessment module, and a result output module.
[0008] The data acquisition module is used to acquire X-ray images of the patient's shoulder joint;
[0009] The image preprocessing module performs preprocessing operations on the acquired X-ray images to improve the accuracy of subsequent analysis;
[0010] The feature extraction module extracts feature data from the preprocessed image for subsequent evaluation calculations;
[0011] The osteoporosis assessment module calculates the mean cortical bone thickness (CBTavg) and cortical bone thickness index (CBTg) of the proximal humerus based on the extracted feature data, thereby assessing the risk of osteoporosis.
[0012] The results output module outputs the evaluation results in graphical or numerical form for doctors' reference.
[0013] As an improvement, the system uses the LeNet model for data preprocessing, model training, testing, and visualization output.
[0014] As an improvement, the system uses code trained and tested on the MNIST dataset using the PyTorch framework; the MNIST training includes the following steps:
[0015] a. Load the MNIST dataset using the torchvision.datasets.MNIST class of the PyTorch framework and apply preprocessing operations, including conversion to Tensors and normalization.
[0016] b. Use the DataLoader class to create a data loader to load data in batches, and set up multi-threading to speed up the data loading process;
[0017] c. Define the LeNet model structure, including convolutional layers, pooling layers, fully connected layers, etc., for feature extraction and image classification;
[0018] d. Set the loss function, train the model, and optimize the parameters;
[0019] e. Use a learning rate scheduler to adjust the learning rate to optimize the training process and improve model performance;
[0020] f. During training, record and visualize key metrics such as training and testing loss and accuracy to monitor the model's learning progress and performance.
[0021] As an improvement, the feature extraction module employs deep learning technology, which automatically identifies and extracts feature data related to osteoporosis from X-ray images through a trained neural network model.
[0022] As an improvement, the osteoporosis assessment module includes the following steps:
[0023] a. Calculate the mean cortical bone thickness (CBTavg) of the proximal humerus by measuring the cortical bone thickness at two different levels and taking the average value;
[0024] b. Calculate the cortical bone thickness index (CBTg) by measuring CBTavg and dividing it by the outer bone diameter to correct for the effects of X-ray image magnification differences and total bone diameter measured by DXA on BMD measurement;
[0025] c. Assess the patient's osteoporosis risk level based on CBTavg and CBTg values, combined with clinical criteria.
[0026] As an improvement, the method for calculating the average cortical bone thickness (CBTavg) is as follows:
[0027]
[0028] Where M1 and M2 are the lateral and medial widths of the cortical bone at the first level detection line, respectively; M3 and M4 are the lateral and medial widths of the cortical bone at the second level detection line, respectively.
[0029] The first horizontal detection line is located at the proximal end of the humerus, and the second horizontal detection line is located 15-25 mm distal to the first horizontal detection line.
[0030] As an improvement, the method for calculating the cortical bone thickness index (CBTg) is as follows:
[0031]
[0032] Wherein, CBTavg is the average cortical bone thickness of the proximal humerus, and M1 is the lateral width of the cortical bone at the first level detection line.
[0033] The evaluation method of the present invention includes the following steps:
[0034] S1. Data Acquisition
[0035] Obtain X-ray images of the patient's shoulder joint, ensuring that the images are clear and meet medical imaging standards to facilitate subsequent processing and analysis;
[0036] S2. Image Preprocessing
[0037] Preprocessing operations are performed on the acquired X-ray images;
[0038] S3. Feature Extraction
[0039] Using a trained deep learning model, osteoporosis-related feature data can be automatically extracted from preprocessed images;
[0040] S4. Osteoporosis Assessment
[0041] Calculate the mean cortical bone thickness (CBTavg) and cortical bone thickness index (CBTg) of the proximal humerus to assess the risk of osteoporosis;
[0042] S5. Output Results
[0043] The assessment results will be output in graphical or numerical form for doctors to refer to and make decisions.
[0044] The advantages of this invention are:
[0045] 1. This invention automatically extracts osteoporosis-related features from shoulder joint X-ray images using deep learning technology, effectively reducing the subjectivity and error of manual operation and improving the accuracy and consistency of assessment. The system of this invention not only speeds up the assessment process but also allows for large-scale osteoporosis screening, improving clinical efficiency.
[0046] 2. Compared to traditional BMD measurements, this system does not require expensive specialized equipment and can utilize existing X-ray images for osteoporosis assessment, significantly reducing diagnostic costs. Furthermore, because the system is based on common imaging data, osteoporosis assessment becomes feasible even in resource-constrained medical settings.
[0047] 3. This system can provide intuitive evaluation results, including specific data files and graphic files, and generate evaluation reports based on the trained model, which can provide intuitive and reliable assistance to doctors in diagnosis. Attached Figure Description
[0048] Figure 1 This is a schematic diagram of an automatic assessment method for proximal humeral osteoporosis based on shoulder joint X-ray images, as described in Example 1. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0050] Example 1
[0051] This embodiment discloses an automatic assessment system for proximal humeral osteoporosis based on shoulder joint X-ray images, including a data acquisition module, an image preprocessing module, a feature extraction module, an osteoporosis assessment module, and a result output module;
[0052] The data acquisition module is used to acquire X-ray images of the patient's shoulder joint;
[0053] The image preprocessing module performs preprocessing operations on the acquired X-ray images to improve the accuracy of subsequent analysis;
[0054] The feature extraction module extracts feature data from the preprocessed image for subsequent evaluation calculations;
[0055] The osteoporosis assessment module calculates the mean cortical bone thickness (CBTavg) and cortical bone thickness index (CBTg) of the proximal humerus based on the extracted feature data, thereby assessing the risk of osteoporosis.
[0056] The results output module outputs the evaluation results in graphical or numerical form for doctors' reference.
[0057] This system uses the LeNet model for data preprocessing, model training, testing, and visualization output.
[0058] As an improvement, the system uses code trained and tested on the MNIST dataset using the PyTorch framework; the MNIST training includes the following steps:
[0059] a. Load the MNIST dataset using the torchvision.datasets.MNIST class of the PyTorch framework and apply preprocessing operations, including conversion to Tensors and normalization.
[0060] b. Use the DataLoader class to create a data loader to load data in batches, and set up multi-threading to speed up the data loading process;
[0061] c. Define the LeNet model structure, including convolutional layers, pooling layers, fully connected layers, etc., for feature extraction and image classification;
[0062] d. Set the loss function, train the model, and optimize the parameters;
[0063] e. Use a learning rate scheduler to adjust the learning rate to optimize the training process and improve model performance;
[0064] f. During training, record and visualize key metrics such as training and testing loss and accuracy to monitor the model's learning progress and performance.
[0065] The following section explains the system using code examples:
[0066] Create a SummaryWriter instance to record various metrics during the training process for visualization in TensorBoard. Define two sets of data preprocessing operations: one for training data (including random horizontal flipping, conversion to Tensors, and normalization) and the other for test data (including conversion to Tensors and normalization).
[0067] writer=SummaryWriter('LeNet')
[0068] data_transforms = {
[0069] 'train':transforms.Compose([
[0070] transforms.RandomHorizontalFlip(), # Randomly flips a given PIL image horizontally with a given probability.
[0071] `transforms.ToTensor()` converts the image to a Tensor, normalizing it to [0,1].
[0072] transforms.Normalize([0.406],[0.225]) # Normalize the tensor image using the mean and standard deviation.
[0073] ]),
[0074] 'test':transforms.Compose([
[0075]
[0076] Load the training and test sets using torchvision.datasets.MNIST and apply preprocessing operations.
[0077]
[0078] Define the LeNet neural network model, which contains two sequence modules, block_1 and block_2, for image recognition tasks.
[0079]
[0080]
[0081]
[0082]
[0083] The model training process includes forward propagation, loss calculation, backpropagation, and parameter updates, and records the training and testing losses and accuracy. Furthermore, the model sets up the loss function, optimizer, and learning rate scheduler, then calls the `train_model` function for training and plots the learning rate changes.
[0084] if __name__ == '__main__':
[0085] net = LeNet()
[0086] net = net.to(device)
[0087] lr_list = []
[0088] criterion=nn.CrossEntropyLoss()
[0089] #optimizer=optim.SGD(net.parameters(),lr=1e-3,momentum=0.9)
[0090] optimizer=optim.Adam(net.parameters(),lr=1e-3,weight_decay=1e-3)
[0091] lr_scheduler=lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.9)
[0092] net=train_model(net,criterion,optimizer,lr_scheduler,num_epochs=200)
[0093] plt.plot(range(100),lr_list,color='r')
[0094] This invention covers several key steps in deep learning projects, from data preprocessing to model definition, training, testing, and performance evaluation. It effectively reduces the subjectivity and error of manual operations, improving the accuracy and consistency of evaluations.
[0095] The feature extraction module employs deep learning technology to automatically identify and extract osteoporosis-related feature data from X-ray images using a trained neural network model. The osteoporosis assessment module includes the following steps:
[0096] a. Calculate the mean cortical bone thickness (CBTavg) of the proximal humerus by measuring the cortical bone thickness at two different levels and taking the average value;
[0097] The method for calculating the mean cortical bone thickness (CBTavg) is as follows:
[0098]
[0099] Where M1 and M2 are the lateral and medial widths of the cortical bone at the first level detection line, respectively; M3 and M4 are the lateral and medial widths of the cortical bone at the second level detection line, respectively.
[0100] The first horizontal detection line is located at the proximal end of the humerus, and the second horizontal detection line is located 20 mm distal to the first horizontal detection line.
[0101] b. Calculate the cortical bone thickness index (CBTg) by measuring CBTavg and dividing it by the outer bone diameter to correct for the effects of X-ray image magnification differences and total bone diameter measured by DXA on BMD measurement;
[0102] The method for calculating the cortical bone thickness index (CBTg) is as follows:
[0103] CBTg = CBTavg / M1
[0104] Wherein, CBTavg is the average cortical bone thickness of the proximal humerus, and M1 is the lateral width of the cortical bone at the first level detection line.
[0105] c. Assess the patient's osteoporosis risk level based on CBTavg and CBTg values, combined with clinical criteria.
[0106] This invention provides a highly reliable method for measuring cortical bone thickness of the humeral shaft based on standard X-ray images. Furthermore, it offers excellent prediction of local skeletal status and can be used for assessment in clinical settings where quantitative BMD measurement is not feasible, making it a simple, convenient, and inexpensive assessment tool.
[0107] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An automatic assessment system for proximal humeral osteoporosis based on shoulder joint X-ray images, characterized in that, It includes a data acquisition module, an image preprocessing module, a feature extraction module, an osteoporosis assessment module, and a result output module; The data acquisition module is used to acquire X-ray images of the patient's shoulder joint; The image preprocessing module performs preprocessing operations on the acquired X-ray images to improve the accuracy of subsequent analysis; The feature extraction module extracts feature data from the preprocessed image for subsequent evaluation calculations; The osteoporosis assessment module calculates the mean cortical bone thickness (CBTavg) and cortical bone thickness index (CBTg) of the proximal humerus based on the extracted feature data, thereby assessing the risk of osteoporosis. The results output module outputs the evaluation results in graphical or numerical form for doctors' reference.
2. The automatic evaluation system according to claim 1, characterized in that, The system uses the LeNet model for data preprocessing, model training, testing, and visualization output.
3. The automatic evaluation system according to claim 2, characterized in that, The system uses code trained and tested on the MNIST dataset using the PyTorch framework; the MNIST training includes the following steps: a. Load the MNIST dataset using the torchvision.datasets.MNIST class of the PyTorch framework and apply preprocessing operations, including conversion to Tensors and normalization. b. Use the DataLoader class to create a data loader to load data in batches, and set up multi-threading to speed up the data loading process; c. Define the LeNet model structure, including convolutional layers, pooling layers, fully connected layers, etc., for feature extraction and image classification; d. Set the loss function, train the model, and optimize the parameters; e. Use a learning rate scheduler to adjust the learning rate to optimize the training process and improve model performance; f. During training, record and visualize key metrics such as training and testing loss and accuracy to monitor the model's learning progress and performance.
4. The automatic evaluation system according to claim 1, characterized in that, The feature extraction module uses deep learning technology to automatically identify and extract feature data related to osteoporosis from X-ray images through a trained neural network model.
5. The automatic evaluation system according to claim 1, characterized in that, The osteoporosis assessment module includes the following steps: a. Calculate the mean cortical bone thickness (CBTavg) of the proximal humerus by measuring the cortical bone thickness at two different levels and taking the average value; b. Calculate the cortical bone thickness index (CBTg) by measuring CBTavg and dividing it by the outer bone diameter to correct for the effects of X-ray image magnification differences and total bone diameter measured by DXA on BMD measurement; c. Assess the patient's osteoporosis risk level based on CBTavg and CBTg values, combined with clinical criteria.
6. The automatic evaluation system according to claim 5, characterized in that, The method for calculating the mean cortical bone thickness (CBTavg) is as follows: Where M1 and M2 are the lateral and medial widths of the cortical bone at the first level detection line, respectively; M3 and M4 are the lateral and medial widths of the cortical bone at the second level detection line, respectively. The first horizontal detection line is located at the proximal end of the humerus, and the second horizontal detection line is located 15-25 mm distal to the first horizontal detection line.
7. The automatic evaluation system according to claim 6, characterized in that, The method for calculating the cortical bone thickness index (CBTg) is as follows: Wherein, CBTavg is the average cortical bone thickness of the proximal humerus, and M1 is the lateral width of the cortical bone at the first level detection line.
8. A method for assessing proximal humeral osteoporosis based on shoulder joint X-ray images, characterized in that, The automatic evaluation system according to any one of claims 1-7 is executed by comprising the following steps: S1. Data Acquisition Obtain X-ray images of the patient's shoulder joint, ensuring that the images are clear and meet medical imaging standards to facilitate subsequent processing and analysis; S2. Image Preprocessing Preprocessing operations are performed on the acquired X-ray images; S3. Feature Extraction Using a trained deep learning model, osteoporosis-related feature data can be automatically extracted from preprocessed images; S4. Osteoporosis Assessment Calculate the mean cortical bone thickness (CBTavg) and cortical bone thickness index (CBTg) of the proximal humerus to assess the risk of osteoporosis; S5. Output Results The assessment results will be output in graphical or numerical form for doctors to refer to and make decisions.