Human cervical vertebra COBB angle automatic measurement method based on key point detection
Through the automatic measurement method of the human cervical spine COBB angle based on key point detection, the cervical spine area detection and key point coordinate calculation are used to solve the problem of traditional measurements being time-consuming and labor-intensive and susceptible to human subjective factors, and high-precision and high-efficiency COBB angle measurement is achieved.
Patent Information
- Application Number
- CN202411761471.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-03
- Publication Date
- 2025-03-14
AI Technical Summary
Traditional COBB angle measurements rely on radiographic images and manual measurements, which are time-consuming and labor-intensive and susceptible to human subjective factors, making it difficult to improve the accuracy of key point detection.
The automatic measurement method of the human cervical spine COBB angle based on key point detection is adopted, including data preprocessing module, model construction and angle measurement module, and model optimization and verification module. The RTMDet model and RTMPose model-CobbNet are used to detect the cervical spine region and calculate the key point coordinates, and the model is trained and optimized by a large amount of labeled data.
It improves the accuracy of COBB angle measurement, reduces artificial errors, improves work efficiency, realizes real-time monitoring and dynamic evaluation, reduces subjective deviations, and helps scientific research and teaching.
Smart Images
Figure CN119624931A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field, and in particular to a method for automatically measuring the COBB angle of a human cervical spine based on key point detection. Background Art
[0002] The COBB angle is an important parameter for evaluating the degree of scoliosis and is widely used in clinical diagnosis and treatment effect evaluation. Traditional COBB angle measurement usually relies on radiographic imaging and manual measurement, which is not only time-consuming and labor-intensive, but also easily affected by human subjective factors. With the development of deep learning and computer vision technology, automated measurement methods based on key point detection have gradually attracted attention. Key point detection technology can automatically identify and locate key anatomical points in the image, providing accurate data support for subsequent angle calculations.
[0003] The measurement of COBB angles is highly dependent on the accurate positioning of key points, and any slight deviation may lead to errors in angle calculation. Therefore, how to improve the accuracy of key point detection is a key issue.
[0004] Due to individual differences and different shooting angles, cervical spine images have diverse presentations, and complex backgrounds and different pathological conditions increase the difficulty of key point detection.
[0005] To this end, we propose an automatic measurement method for the human cervical COBB angle based on key point detection to solve the existing problem. Summary of the invention
[0006] The purpose of the present invention is to solve the problems existing in the background technology and to propose a method for automatically measuring the COBB angle of the human cervical spine based on key point detection.
[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a method for automatic measurement of human cervical COBB angle based on key point detection, comprising a data preprocessing module, a model construction and angle measurement module and a model optimization and verification module, wherein the model construction and angle measurement module comprises an RTMDet model and an RTMPose model-CobbNet, wherein the RTMDet model detects the cervical region, and the detected cervical region coordinates and the original image are output to the RTMPose model-CobbNet through the data preprocessing module to obtain the required four key point coordinates, and the Cobb angle of the C2-7 cervical vertebrae is calculated through the four key point coordinates according to the calculation function, and the model optimization and verification module is used to perform model training and optimization through a large amount of labeled data, thereby continuously improving the performance and reliability of the model.
[0008] Preferably, the data preprocessing module includes image enhancement, noise removal and size normalization. Common methods of image enhancement include: grayscale transformation enhancement, histogram equalization, and frequency domain filtering. The main purpose of image enhancement is to improve the visual effect of the image, improve the quality of the image, and make it more suitable for subsequent analysis and processing. For cervical spine images, image enhancement can highlight the structural features of the cervical spine, such as intervertebral spaces, intervertebral foramina, etc., while suppressing unnecessary noise and interference information.
[0009] Preferably, the commonly used methods for noise removal include: filtering technology: linear filtering: such as mean filtering, Gaussian filtering, etc., which smoothes the image through convolution operations to remove noise; nonlinear filtering: such as median filtering, bilateral filtering, etc., which uses comparison or weighted operations between pixels to remove noise while retaining the edge information of the image; wavelet denoising: based on the wavelet transform method, the image is decomposed into sub-bands of different frequencies, and then the high-frequency sub-bands are subjected to noise reduction processing, and finally the denoised image is obtained through inverse wavelet transform.
[0010] Preferably, the size normalization refers to adjusting the size of the image to a uniform standard size to facilitate subsequent processing and analysis. For cervical spine images, size normalization can ensure that images collected from different sources and at different times have the same size, thereby facilitating comparison and analysis. Common methods include: linear interpolation: such as nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, etc., which adjust the size of the image to the target size through an interpolation algorithm; rigid body transformation: a transformation method that keeps the pixel distance between two points in the image unchanged, which can be used to adjust the direction and scale of the image to meet specific size requirements.
[0011] Preferably, the RTMDet model introduces large-kernel deep convolutions in the basic building blocks of the trunk and neck of the model. In order to balance the depth and width of the model, RTMDet reduces the number of building blocks to reduce the model depth, and optimizes the model architecture by increasing the model width to compensate for the model capacity. At the same time, RTMDet introduces soft labels in label allocation. In summary, the RTMDet model achieves efficient real-time target detection by introducing large-kernel deep convolutions, optimizing the model architecture, and improving the label allocation strategy.
[0012] Preferably, the RTMPose model-CobbNet uses an algorithm based on SimCC to predict key points. The algorithm regards key point positioning as a classification task, and divides the horizontal and vertical coordinates into bins of equal width and discretizes the continuous coordinates into integer bin labels, thereby achieving accurate positioning of key points. The SimCC algorithm includes coordinate discretization, model training, prediction and optimization, Gaussian label smoothing strategy, and micro-design. The SimCC algorithm regards key point coordinate prediction as a classification task and uses optimization methods such as Gaussian label smoothing strategy to achieve accurate positioning of key points. The algorithm has shown excellent performance in applications such as human posture estimation, and has low computational complexity and high accuracy.
[0013] Preferably, the model optimization and verification module steps include data preparation before model training, model training and optimization, model verification, model deployment and monitoring. Through the refinement and implementation of the above steps, it can ensure that the model is trained and optimized with the support of a large amount of labeled data, and its performance can be verified on an independent test set, thereby demonstrating higher reliability and robustness in practical applications.
[0014] Preferably, the calculation process of the Cobb angle includes: if there are coordinates of four points A, B, C, and D, a straight line L1 is made through AB, a straight line L2 is made through CD, a perpendicular line L3 is made through B to L1, and a perpendicular line L4 is made through B to L2, and the content of L3 and L4 is calculated to be the Cobb angle.
[0015] Compared with the prior art, the present invention has the following beneficial effects:
[0016] The human cervical vertebra COBB angle automatic measurement method based on key point detection of the present invention can improve the measurement accuracy during use: the automatic method based on key point detection helps to improve the accuracy of COBB angle measurement by accurately locating specific anatomical points of the cervical vertebra, thereby reducing human errors;
[0017] Improve work efficiency: The automated measurement system can significantly reduce doctors’ workload, avoid the tedious process of manual measurement, and improve clinical work efficiency and diagnosis speed;
[0018] Real-time monitoring and dynamic evaluation: The automatic measurement system can quickly process image data, realize real-time monitoring and dynamic evaluation, and facilitate doctors to adjust treatment plans in time;
[0019] Reduce subjective bias: The automated system uses standardized algorithms to perform measurements, avoiding subjective differences between different doctors and improving the consistency and reliability of measurement results;
[0020] Assisting scientific research and teaching: The automatic measurement system can also be used for the analysis and processing of large-scale imaging data to assist scientific research. At the same time, the system can be used as a teaching tool to help medical students learn and master the measurement method of COBB angles. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 This is the C2-7 Cobb angle measurement flow chart of the present invention. DETAILED DESCRIPTION
[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] Embodiment 1
[0024] like Figure 1 As shown, the present invention proposes an automatic measurement method for the COBB angle of the human cervical spine based on key point detection, including a data preprocessing module, a model construction and angle measurement module and a model optimization and verification module. The model construction and angle measurement module includes an RTMDet model and an RTMPose model-CobbNet. The RTMDet model detects the cervical spine area, and the detected cervical spine area coordinates and the original image are output to the RTMPose model-CobbNet through the data preprocessing module to obtain the required four key point coordinates. The Cobb angle of the C2-7 cervical vertebrae is calculated through the four key point coordinates according to the calculation function, and the model optimization and verification module is used to train and optimize the model through a large amount of labeled data to continuously improve the performance and reliability of the model.
[0025] Embodiment 2
[0026] like Figure 1 As shown, the present invention proposes a method for automatically measuring the COBB angle of the human cervical spine based on key point detection. Compared with the first embodiment, this embodiment further includes: the data preprocessing module includes image enhancement, noise removal and size normalization, and the common methods of image enhancement include: grayscale transformation enhancement: by adjusting the grayscale range of the image, the contrast of the image is improved to make the image clearer;
[0027] Histogram equalization: It is a normalization method based on pixel value distribution. It adjusts the histogram of the image to make the pixel value distribution more uniform, thereby enhancing the contrast of the image.
[0028] Frequency domain filtering: convert the image from the spatial domain to the frequency domain, use filters to enhance or suppress specific components in the frequency domain, and then convert it back to the spatial domain to improve the image quality;
[0029] The main purpose of image enhancement is to improve the visual effect of the image, enhance the image quality, and make it more suitable for subsequent analysis and processing. For cervical spine images, image enhancement can highlight the structural features of the cervical spine, such as the intervertebral space and intervertebral foramen, while suppressing unnecessary noise and interference information.
[0030] Common methods for noise removal include: Filtering technology: Linear filtering: such as mean filtering and Gaussian filtering, which smooth the image through convolution operations to remove noise;
[0031] Nonlinear filtering: such as median filtering and bilateral filtering, which use comparison or weighted operations between pixels to remove noise while retaining the edge information of the image;
[0032] Wavelet denoising: Based on the wavelet transform method, the image is decomposed into sub-bands of different frequencies, and then the high-frequency sub-bands are subjected to denoising. Finally, the denoised image is obtained through inverse wavelet transform.
[0033] Size normalization refers to adjusting the size of an image to a uniform standard size for subsequent processing and analysis. For cervical spine images, size normalization can ensure that images collected from different sources and at different times have the same size, which is convenient for comparison and analysis. Common methods include: linear interpolation: such as nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, etc., which adjust the size of the image to the target size through interpolation algorithms; rigid body transformation: a transformation method that keeps the pixel distance between two points in the image unchanged, which can be used to adjust the direction and scale of the image to meet specific size requirements.
[0034] The RTMDet model introduces large-kernel deep convolutions in the basic building blocks of the backbone and neck of the model, improving the model's ability to capture global context.
[0035] The use of large kernel convolution allows the model to extract features within a larger receptive field, thereby identifying targets more accurately. In order to balance the depth and width of the model, RTMDet reduces the number of building blocks to reduce the model depth and compensates for the model capacity by increasing the model width. This design improves the accuracy of the model while maintaining the model inference speed.
[0036] The model architecture is optimized. At the same time, RTMDet introduces soft labels in label assignment. Soft labels can more finely describe the degree of match between the target and the predicted box. This improvement improves the discriminative ability of the loss matrix and reduces the noise of label assignment, thereby improving the accuracy of the model. The model architecture of RTMDet mainly includes RTMDet using CSPDarkNet as the backbone. The architecture contains multiple stages, and each stage stacks several basic building blocks. These modules enhance feature extraction capabilities through large-core deep convolution; the Neck part extracts a multi-scale feature pyramid from the backbone and uses the same basic building blocks as the backbone for feature fusion. This design enhances the expressiveness of the pyramid feature map and helps to more accurately detect targets of different scales; the Detection head predicts the bounding box, classification results, and regression results of the object based on the feature map of each scale.
[0037] The detection head of RTMDet adopts the design of shared convolution weights and separate batch normalization layers to improve the inference speed and accuracy. In summary, the RTMDet model achieves efficient real-time target detection by introducing large-kernel deep convolution, optimizing model architecture, and improving label assignment strategy.
[0038] The RTMPose model - CobbNet uses a convolutional neural network (CNN) as the basic architecture and combines multiple optimization strategies to achieve efficient real-time multi-person pose estimation. RTMPose regards coordinate classification as an ordered regression task and follows the soft label encoding proposed in SORD. This loss function design helps to reduce quantization error and improve the robustness of the model.
[0039] The RTMPose model - CobbNet uses an algorithm based on SimCC to predict key points. The algorithm regards key point positioning as a classification task. It divides the horizontal and vertical coordinates into bins of equal width and discretizes the continuous coordinates into integer bin labels to achieve accurate positioning of key points. The SimCC algorithm includes coordinate discretization, model training, prediction and optimization, Gaussian label smoothing strategy, and micro-design. The SimCC algorithm regards the prediction of key point coordinates as a classification task and uses optimization methods such as Gaussian label smoothing strategy to achieve accurate positioning of key points. The algorithm has shown excellent performance in applications such as human posture estimation, and has low computational complexity and high accuracy.
[0040] The steps of the model optimization and verification module include data preparation before model training, including data annotation:
[0041] Clarify requirements: Before the project starts, clarify the goals, scope and standards of labeling to ensure the accuracy and consistency of the labeling work; Form a team: According to the needs of the labeling work, form a professional labeling team or outsource it to ensure that the labeling personnel have the corresponding professional knowledge and skills; Sample collection: Collect representative sample data from multiple channels to ensure the diversity, balance and accuracy of the samples.
[0042] Data cleaning: Clean the collected data to remove noise, outliers and duplicate data to ensure the quality of training data, and divide the cleaned data into training set, validation set and test set. The training set is used for model training, the validation set is used for model tuning, and the test set is used for the final evaluation of model performance;
[0043] Model training and optimization selects appropriate model architectures according to task requirements and data characteristics, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Through grid search, random search, or Bayesian optimization, the model's hyperparameters are tuned to find the optimal parameter combination, and model validation, deployment, and monitoring are performed. By refining and implementing the above steps, it is possible to ensure that the model is trained and optimized with the support of a large amount of labeled data, and its performance is verified on an independent test set, thereby demonstrating higher reliability and robustness in practical applications.
[0044] The calculation process of Cobb angle includes the coordinates of four key points (key points A and B on C2 vertebra, key point CD on C7 vertebra), and the Cobb angle is calculated according to the coordinates of the four points. The calculation process is as follows: if there are four points A, B, C, and D, make a straight line L1 through AB, a straight line L2 through CD, a perpendicular line L3 through B to L1, and a perpendicular line L4 through B to L2, and the intersection of L3 and L4 is the Cobb angle. The calculation function of Cobb angle is:
[0045]
[0046] The above-mentioned specific embodiments are only several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above-mentioned embodiments, those skilled in the art can make various alternative improvements and combinations to the above-mentioned specific embodiments.
[0047] It will be apparent to those skilled in the art that the 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 the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A method for automatically measuring the COBB angle of the human cervical spine based on key point detection, comprising a data preprocessing module, a model building and angle measurement module, and a model optimization and verification module, characterized in that: The model construction and angle measurement module includes an RTMDet model and an RTMPose model—CobbNet. The RTMDet model detects the cervical spine area. The detected cervical spine area coordinates and the original image are output to the RTMPose model—CobbNet through a data preprocessing module to obtain the required four key point coordinates. The Cobb angles of the C2-7 cervical vertebrae are calculated through the four key point coordinates according to the calculation function. The model optimization and verification module is used to train and optimize the model through a large amount of labeled data, thereby continuously improving the performance and reliability of the model.
2. The method for automatically measuring the COBB angle of the human cervical vertebra based on key point detection according to claim 1, characterized in that: The data preprocessing module includes image enhancement, noise removal and size normalization. Common methods of image enhancement include: grayscale transformation enhancement, histogram equalization, and frequency domain filtering. The main purpose of image enhancement is to improve the visual effect of the image, improve the quality of the image, and make it more suitable for subsequent analysis and processing.
3. The method for automatically measuring the COBB angle of the human cervical vertebra based on key point detection according to claim 2 is characterized in that: The commonly used methods for noise removal include: filtering technology: linear filtering: such as mean filtering, Gaussian filtering, etc., which smoothes the image through convolution operations to remove noise; nonlinear filtering: such as median filtering, bilateral filtering, etc., which uses comparison or weighted operations between pixels to remove noise while retaining the edge information of the image; wavelet denoising: based on the wavelet transform method, the image is decomposed into sub-bands of different frequencies, and then the high-frequency sub-bands are subjected to noise reduction processing, and finally the denoised image is obtained through inverse wavelet transform.
4. The method for automatically measuring the COBB angle of the human cervical vertebra based on key point detection according to claim 2, characterized in that: The size normalization refers to adjusting the size of the image to a uniform standard size for subsequent processing and analysis. For cervical spine images, size normalization can ensure that images collected from different sources and at different times have the same size, which is convenient for comparison and analysis. Common methods include: linear interpolation: such as nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, etc., which adjust the size of the image to the target size through the interpolation algorithm; rigid body transformation: a transformation method that keeps the pixel distance between two points in the image unchanged, which can be used to adjust the direction and scale of the image to meet specific size requirements.
5. The method for automatically measuring the COBB angle of the human cervical vertebra based on key point detection according to claim 1, characterized in that: The RTMDet model introduces large-kernel deep convolutions in the basic building blocks of the backbone and neck of the model. In order to balance the depth and width of the model, RTMDet reduces the number of building blocks to reduce the model depth, and optimizes the model architecture by increasing the model width to compensate for the model capacity. At the same time, RTMDet introduces soft labels in label assignment.
6. The method for automatically measuring the COBB angle of the human cervical vertebra based on key point detection according to claim 1, characterized in that: The RTMPose model - CobbNet uses an algorithm based on SimCC to predict key points. The algorithm regards key point positioning as a classification task. It divides the horizontal and vertical coordinates into bins of equal width and discretizes the continuous coordinates into integer bin labels to achieve accurate positioning of key points. The SimCC algorithm includes coordinate discretization, model training, prediction and optimization, Gaussian label smoothing strategy, and micro-design. The SimCC algorithm regards the prediction of key point coordinates as a classification task and uses optimization methods such as Gaussian label smoothing strategy to achieve accurate positioning of key points. The algorithm has shown excellent performance in applications such as human posture estimation, and has low computational complexity and high accuracy.
7. The method for automatically measuring the COBB angle of the human cervical spine based on key point detection according to claim 1, characterized in that: The steps of the model optimization and verification module include data preparation before model training, model training and optimization, model verification, model deployment and monitoring. Through the refinement and implementation of the above steps, it can ensure that the model is trained and optimized with the support of a large amount of labeled data, and its performance can be verified on an independent test set, thereby demonstrating higher reliability and robustness in practical applications.
8. The method for automatically measuring the COBB angle of the human cervical spine based on key point detection according to claim 1, characterized in that: The calculation process of the Cobb angle includes: if there are coordinates of four points A, B, C, and D, a straight line L1 is made through AB, a straight line L2 is made through CD, a perpendicular line L3 of L1 is made through B, and a perpendicular line L4 of L2 is made through B, and the content of L3 and L4 is calculated as the Cobb angle, and a calculation function is calculated.