A spinal deformity analysis system based on key point detection

Through the spinal deformity analysis system based on key point detection, the cumbersome and inconsistency of traditional analysis methods are solved, real-time monitoring of spinal health status and personalized rehabilitation suggestions are achieved, and the efficiency and accuracy of spinal health management are improved.

CN117838097BActive Publication Date: 2024-11-12FORETHOUGHT (SHANGHAI) MEDICAL TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410205610.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-23
Publication Date
2024-11-12
Estimated Expiration
2044-02-23

AI Technical Summary

Technical Problem

Traditional spinal deformity analysis methods require manual measurement and analysis by doctors, which are cumbersome and time-consuming. Due to human factors, insufficient accuracy and consistency are insufficient, making it difficult to monitor spinal changes in real time and cannot adjust the treatment plan in time.

Method used

A spinal deformity analysis system based on key point detection is designed, including spinal surface morphology acquisition subsystem, key point analysis subsystem, analysis and calculation subsystem, evaluation index computing subsystem and data analysis subsystem, through deep learning model Identify key points in the spinal vertebral area, calculate Cobb angle, ATR, rotation angle, chest curve and waist curve information, and generate evaluation indicators and early warning prompts.

Benefits of technology

It improves intuitive understanding of spinal health, provides timely rehabilitation advice and alert information, helps users adjust their posture and lifestyle, improves spinal health, promptly deal with emergency spinal problems, and reduces health risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117838097B_ABST
    Figure CN117838097B_ABST
Patent Text Reader

Abstract

The present invention discloses a spinal deformity analysis system based on key point detection, and relates to the medical field. The spinal deformity analysis system based on key point detection includes: a spinal surface morphology acquisition subsystem, a key point analysis subsystem, an analysis and calculation subsystem, an evaluation index calculation subsystem, a data analysis subsystem, and a central control platform; the spinal surface morphology acquisition subsystem is used to collect the morphological data of the spine on the human body surface; the key point analysis subsystem is used to process and feature analyze the information acquired by the spinal surface morphology acquisition subsystem, and obtain the coordinates and posture information of each vertebral area of ​​the spine to be analyzed. The present invention can more intuitively understand the health status of the spine through the data analysis subsystem, and the evaluation indicators and early warning prompts generated by the system are helpful to improve the user's cognition of spinal health, and encourage people to pay attention to and adopt a positive lifestyle and posture to maintain the normal physiological state of the spine.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the medical field, and in particular to a spinal deformity analysis system based on key point detection. Background Art

[0002] Spinal deformity is a common skeletal disease, including scoliosis, spinal rotation, etc., which usually develops gradually during childhood and adolescence. Early diagnosis and monitoring of spinal deformity are crucial for timely intervention and treatment. Traditional spinal deformity analysis usually relies on the doctor's experience and routine imaging examinations, such as X-rays. However, these methods have some limitations, such as radiation exposure and insufficient two-dimensional information to fully evaluate the three-dimensional spinal structure.

[0003] In recent years, technologies based on computer vision and deep learning have gradually made significant progress in the field of medical imaging. Key point detection is one of the important technologies. It can automatically locate key points from images, thereby achieving accurate analysis of human structure. For spinal deformities, key point detection can be used to calibrate specific positions of the spine and then quantify the morphological characteristics of the spine.

[0004] In the existing technology, traditional analysis methods usually require doctors to manually measure and analyze various indicators of the spine. This process is cumbersome and time-consuming. At the same time, due to human factors and subjective judgment, different doctors may give different assessment results for the spinal deformity of the same patient, resulting in insufficient accuracy and consistency. In addition, traditional analysis methods are usually one-time static analysis, which makes it difficult to monitor changes in the spine in real time and unable to adjust the treatment plan in time. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a spinal deformity analysis system based on key point detection, which solves the problem that traditional analysis methods usually require doctors to manually measure and analyze various indicators of the spine. This process is cumbersome and time-consuming. At the same time, due to human factors and subjective judgment, different doctors may give different evaluation results for the spinal deformity of the same patient, resulting in insufficient accuracy and consistency. In addition, traditional analysis methods are usually one-time static analysis, which makes it difficult to monitor changes in the spine in real time and cannot adjust the treatment plan in time.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a spinal deformity analysis system based on key point detection, comprising: a spinal surface morphology acquisition subsystem, a key point analysis subsystem, an analysis and calculation subsystem, an evaluation index calculation subsystem, a data analysis subsystem, and a central control platform; the spinal surface morphology acquisition subsystem is used to collect morphological data of the spine on the human body surface; the key point analysis subsystem is used to process and feature analyze the information obtained by the spinal surface morphology acquisition subsystem, obtain the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, and calculate the vertices of the upper vertebral area and the lower vertebral area. The vertex and the lateral edge points of each vertebral area are used as key points; the analysis and calculation subsystem is used to perform posture analysis and calculation on the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, to obtain the projection information of the spine in the sagittal plane, coronal plane and cross-section, and to analyze and calculate the Cobb angle, ATR, rotation angle, thoracic curve and lumbar curve information; the evaluation index calculation subsystem is used to output data based on the analysis and calculation subsystem, and calculate the evaluation index; the data analysis subsystem is used to analyze the spine based on the calculated evaluation index; the central control platform is used to control the operation of the entire system and coordinate the operation between the subsystems.

[0007] Furthermore, the specific steps for obtaining the coordinates and posture information of each vertebral region of the spine to be analyzed are as follows: obtain trajectory data of several spines to be analyzed, and construct a trajectory data set; divide the trajectory data set into a training data set and a verification data set; read the training data set to train the constructed spinal vertebrae detection learning network model, and calculate the feature loss function; use the verification data set to evaluate the model performance, and adjust the model parameters according to the verification results until the model predicts the feature data consistent with the actual feature data value; the input and input data of the constructed spinal vertebrae detection learning network model are the trajectory data of the spine to be analyzed, the coordinates and posture information of each vertebral region of the spine to be analyzed, and the constructed spinal vertebrae detection learning network model is a deep learning model with a densely connected convolutional network as the architecture and multi-scale feature extraction as a variant, wherein the deep learning model includes: input layer, densely connected convolutional layer, activation function, batch normalization layer, global average pooling layer, transition layer, fully connected layer, Dro pout layer, Softmax layer, loss function, optimizer Among them, the deep learning model includes: an input layer for receiving spinal image data, a densely connected convolution layer using DenseNet, which includes a convolution layer, a batch normalization layer and a ReLU activation function. The densely connected feature enables each convolution layer to receive input from all previous layers, which helps to better transfer gradients and promote feature reuse, a global average pooling layer for converting feature maps into global averages, summarizing multi-dimensional data into one-dimensional vectors, and providing input to the fully connected layer, a transition layer for reducing the size of feature maps, a fully connected layer for mapping high-level features to output categories, a Dropout layer for adding a Dropout layer after the fully connected layer, randomly discarding some neurons, which helps prevent overfitting, a Softmax layer for converting the output of the network into a probability distribution, a loss function for measuring the difference between the model output and the true label, and an optimizer for adjusting model parameters to minimize the loss function.

[0008] Furthermore, the formula for calculating the feature loss function is as follows: Among them, Ss is the feature loss function, Z gut is the t-th predicted coordinate information of the u-th vertebral region in the g-th spine trajectory data to be analyzed predicted by the model, is the tth actual coordinate information of the uth vertebral region in the gth trajectory data of the spine to be analyzed, g=1,2,3,…,G, G is the number of obtained trajectory data of the spine to be analyzed, u=1,2,3,…,U, U is the number of vertebral regions in the obtained trajectory data of the spine to be analyzed, t=1,2,3,…,T, T is the number of coordinate points in the vertebral region in the obtained trajectory data of the spine to be analyzed.

[0009] Furthermore, the analysis and calculation subsystem includes: a Cobb angle measurement and calculation module, an ATR measurement and calculation module, a rotation angle analysis module, and a thoracic curve and lumbar curve analysis module; the Cobb angle measurement and calculation module is used to calculate the Cobb angle based on the coordinate information of the vertex of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed; the ATR measurement and calculation module is used to calculate the ATR based on the coordinate information of the vertex of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed, wherein the ATR represents the ratio of the lateral translation of the highest point to the entire scoliosis height; the rotation angle analysis module is used to calculate the rotation angle of the spine based on the trajectory data of the spine to be analyzed; the thoracic curve and lumbar curve analysis module is used to analyze the physiological curvature of the spine in the sagittal plane, namely the thoracic curve and lumbar curve, based on the trajectory data of the spine to be analyzed.

[0010] Furthermore, the specific steps for calculating the Cobb angle are as follows: read the coordinate information of each coordinate point of the vertex of the upper vertebral area and the lower end base of the lower vertebral area in the image data of the spine to be analyzed, and record them as point A and point B respectively; based on computer image processing technology, use points A and B to draw vertical lines perpendicular to the upper edge of the upper vertebral area and the lower edge of the lower vertebral area respectively, and record the intersection of the two vertical lines as point C; take the midpoints of line segments AC and BC respectively, and draw vertical lines perpendicular to the midpoints of line segments AC and BC respectively; the intersection angle of the two vertical lines is the Cobb angle.

[0011] Furthermore, the specific steps of calculating ATR are as follows: read the coordinate information of point A and point B; calculate the length and height difference on the vertebral axis based on the coordinate information of point A and point B; calculate ATR based on the length and height difference of the spinal column axis, and the calculation formula is as follows: in, is the length of the vertebral axis, |y B -y A | is the height difference.

[0012] Furthermore, the specific steps for calculating the rotation angle of the spine are as follows: respectively read the coordinate information of each coordinate point in each vertebral area, respectively regard the coordinate midpoint of each cone as the center of each cone, and regard it as the respective local coordinate system; based on the trajectory data of the spine to be analyzed, obtain the posture information of each cone, and the posture information is recorded as a direction vector; based on the rotation matrix, the coordinate information and posture information of each coordinate point in the local coordinate system of each vertebral area are converted to the global coordinate system, and the coordinate information and posture information of each coordinate point of each vertebra in the global coordinate system are obtained; and the rotation angle of the spine is extracted based on the rotation matrix.

[0013] Furthermore, the specific steps of analyzing the thoracic curve and lumbar curve of the spine are as follows: reading the coordinate information of each lateral edge point of each vertebral region; and bringing the coordinate information of each lateral edge point of each vertebral region into the curvature calculation model to obtain the curvature value of the coordinate information of each lateral edge point of each vertebral region, and comparing them with the set thoracic curve threshold and lumbar curve threshold respectively; recording the part of each lateral edge point whose curvature value of the coordinate information is greater than the thoracic curve threshold as the thoracic curve, and recording the part of each lateral edge point whose curvature value of the coordinate information is less than the lumbar curve threshold as the lumbar curve; wherein the curvature calculation model is specifically: Among them, K rm is the curvature value of the coordinate information of the m lateral edge points in the rth vertebral area, y″ rm The coordinate information of the m lateral edge points in the rth vertebral region is brought into the second-order derivative of y with respect to x in the curve model y=f(x), y′ rm Substitute the coordinate information of the m lateral edge points in the rth vertebral region into the first-order derivative of y with respect to x in the curve model y=f(x), r=1,2,3,…,C, C is the number of vertebral regions, m=1,2,3,…,F, F is the number of coordinate points in the vertebral region.

[0014] Furthermore, the formula for calculating the evaluation index is as follows: PGz=α1*Cobb′+α2*ATR+α3*Xz; wherein PGz is the evaluation index, Cobb′ is the Cobb angle, Xz is the rotation angle, α1, α2, α3 are the proportional coefficients of Cobb′, ATR, and Xz, respectively, and α1+α2+α3=1.

[0015] Furthermore, the analysis of the spine based on the calculated evaluation index is specifically as follows: the evaluation index is subjected to standard processing, and the above-mentioned standard processing is: the evaluation index is converted into a percentage, and the percentage range is specifically as follows: 0≤PGz%≤100%; when the evaluation index is between 90% and 100%, the spine is regarded as normal; when the evaluation index is between 70% and 89%, the spine is regarded as slightly abnormal, and an early warning prompt is issued, and the early warning prompt is specifically as follows: change the sitting posture, maintain a correct standing posture, and perform appropriate physical exercise; when the evaluation index is between 50% and 69%, the spine is regarded as moderately abnormal, and at the same time, the Cobb′ and ATR of the spine are further analyzed, and an autonomous rehabilitation treatment plan is automatically generated; when the evaluation index is between 0% and 49%, the spine is regarded as severely abnormal, and an alarm message is issued at the same time.

[0016] The present invention has the following beneficial effects:

[0017] (1) The spinal deformity analysis system based on key point detection can provide a more intuitive understanding of the health status of the spine through the data analysis subsystem. The evaluation indicators and early warning prompts generated by the system help improve users' awareness of spinal health, prompting people to pay attention to and adopt a positive lifestyle and posture to maintain the normal physiological state of the spine.

[0018] (2) The spinal deformity analysis system based on key point detection provides timely and effective rehabilitation suggestions for individuals with mild and moderate spinal abnormalities through system warning prompts and automatically generated rehabilitation treatment plans. Users can adjust their sitting and standing postures and participate in specific physical exercises based on system feedback to improve spinal health. This real-time personalized rehabilitation intervention is expected to prevent or slow down the further development of spinal abnormalities at an early stage.

[0019] (3) The spinal deformity analysis system based on key point detection can alert users to seek medical help in a timely manner for serious spinal abnormalities. This helps speed up the treatment of emergency spinal problems and improve the efficiency of medical resource utilization. By providing detailed evaluation indicators, medical professionals can understand the patient's condition more quickly and accurately, provide emergency treatment to patients, and minimize the patient's spinal health risks.

[0020] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a block diagram of a spinal deformity analysis system based on key point detection according to the present invention.

[0022] Figure 2 The present invention is a flowchart of the specific steps of obtaining the coordinates and posture information of each vertebral area of ​​the spine to be analyzed in a spinal deformity analysis system based on key point detection.

[0023] Figure 3 This is a block diagram of the analysis and calculation subsystem in a spinal deformity analysis system based on key point detection of the present invention.

[0024] Figure 4 The present invention is a flowchart of the specific steps for calculating the Cobb angle in a spinal deformity analysis system based on key point detection.

[0025] Figure 5 The present invention is a structural schematic diagram of calculating the Cobb angle in a spinal deformity analysis system based on key point detection.

[0026] Figure 6 The present invention is a flowchart of the specific steps of calculating ATR in a spinal deformity analysis system based on key point detection.

[0027] Figure 7 This is a flowchart of the specific steps for calculating the rotation angle of the spine in a spinal deformity analysis system based on key point detection of the present invention.

[0028] Figure 8 The present invention is a flowchart of the specific steps of calculating and analyzing the thoracic curve and lumbar curve of the spine in a spinal deformity analysis system based on key point detection. DETAILED DESCRIPTION

[0029] The embodiment of the present application uses a spinal deformity analysis system based on key point detection to solve the problem that traditional analysis methods usually require doctors to manually measure and analyze various indicators of the spine. This process is cumbersome and time-consuming. At the same time, due to human factors and subjective judgment, different doctors may give different evaluation results for the spinal deformity of the same patient, resulting in insufficient accuracy and consistency. In addition, traditional analysis methods are usually one-time static analysis, which makes it difficult to monitor changes in the spine in real time and cannot adjust the treatment plan in time.

[0030] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0031] First, the morphological data of the spine on the human body surface is acquired through the spine surface morphology acquisition subsystem. Then, the key point analysis subsystem is used to input the morphological data of the human body surface into the pre-constructed spinal vertebrae detection learning network model. The model can identify the key points of each vertebral area of ​​the spine, including the apex of the upper vertebral area, the apex of the lower vertebral area, and the lateral edge points of each vertebral area. Secondly, the analysis and calculation subsystem is used to calculate the Cobb angle, ATR, rotation angle, thoracic curve, and lumbar curve information of the spine using the coordinate information of each coordinate point in each vertebral area. Then, the evaluation index calculation subsystem calculates the evaluation index based on the Cobb angle, ATR, and rotation angle of the spine to provide a quantitative evaluation of the health status of the spine. Finally, the data analysis subsystem is used to further analyze the calculated evaluation index to obtain more comprehensive information on the condition of the spine.

[0032] See also Figure 1The embodiment of the present invention provides a technical solution: a spinal deformity analysis system based on key point detection, comprising: a spinal surface morphology acquisition subsystem, a key point analysis subsystem, an analysis and calculation subsystem, an evaluation index calculation subsystem, a data analysis subsystem, and a central control platform; the spinal surface morphology acquisition subsystem is used to collect the morphological data of the spine on the human body surface; the key point analysis subsystem is used to process and feature analyze the information obtained by the spinal surface morphology acquisition subsystem, obtain the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, and calculate the vertices of the upper vertebral area and the lower vertebral area. The points and lateral edge points of each vertebral area are taken as key points; the analysis and calculation subsystem is used to perform posture analysis and calculation on the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, to obtain the projection information of the spine in the sagittal plane, coronal plane and cross-section, and to analyze and calculate the Cobb angle, ATR, rotation angle, thoracic curve and lumbar curve information; the evaluation index calculation subsystem is used to output data based on the analysis and calculation subsystem, and to calculate the evaluation index; the data analysis subsystem is used to analyze the spine based on the calculated evaluation index; the central control platform is used to control the operation of the entire system and coordinate the operation between the subsystems.

[0033] Specifically, Figure 2As shown, the specific steps for obtaining the coordinates and posture information of each vertebral area of ​​the spine to be analyzed are as follows: obtain the trajectory data of several spines to be analyzed, and construct a trajectory data set; divide the trajectory data set into a training data set and a verification data set; read the training data set to train the constructed spinal vertebra detection learning network model, and calculate the feature loss function; use the verification data set to evaluate the model performance, and adjust the model parameters according to the verification results until the model predicts the feature data consistent with the actual feature data value; the input and input data of the constructed spinal vertebra detection learning network model are the trajectory data of the spine to be analyzed, the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, and the constructed spinal vertebra detection learning network model is a deep learning model with a densely connected convolutional network as the architecture and multi-scale feature extraction as a variant, wherein the deep learning model includes: input layer, densely connected convolutional layer, activation function, batch normalization layer, global average pooling layer, transition layer, fully connected layer, Drop layer, etc. out layer, Softmax layer, loss function, optimizer Among them, the deep learning model includes: an input layer for receiving spinal image data, a densely connected convolution layer using DenseNet, which includes a convolution layer, a batch normalization layer and a ReLU activation function. The dense connection feature enables each convolution layer to receive input from all previous layers, which helps to better transfer gradients and promote feature reuse, a global average pooling layer for converting feature maps into global averages, summarizing multi-dimensional data into one-dimensional vectors, and providing input to the fully connected layer, a transition layer for reducing the size of the feature map, a fully connected layer for mapping high-level features to output categories, a Dropout layer for adding a Dropout layer after the fully connected layer, randomly discarding some neurons, which helps prevent overfitting, a Softmax layer for converting the output of the network into a probability distribution, a loss function for measuring the difference between the model output and the true label, and an optimizer for adjusting the model parameters to minimize the loss function.

[0034] In this embodiment, a rich data set is constructed by acquiring a large amount of spinal image data, which helps the training model to adapt to the spinal characteristics of different patients and improves the generalization ability of the model so that it can better adapt to diverse spinal deformities. By dividing the data set into a training set and a validation set, the performance of the model can be monitored in real time during the training process to prevent overfitting, which helps to improve the robustness of the model and make it perform better on unseen data. The densely connected convolutional layer helps to improve gradient propagation and feature reuse, and promotes the learning ability of the model. Through the global average pooling layer, multi-dimensional data is summarized into a one-dimensional vector, which helps to reduce the number of parameters and reduce the risk of overfitting. Randomly discarding neurons helps to prevent overfitting. , improve the generalization ability of the model. Through these designs, the feature extraction and learning capabilities of the model are enhanced, making it better adaptable to the complex characteristics of spinal image data. By using the training data set, the model can learn the characteristics of each vertebral area in the spinal image data, and the feature loss function can be used to measure the difference between the model output and the actual label. Through supervised learning, the model can adjust parameters to minimize the loss, thereby improving the accurate prediction of spinal features. Through the validation data set, the performance of the model can be evaluated on samples different from the training data to avoid overfitting of the model on a specific data set. By dynamically adjusting the model parameters, the generalization of the model can be further improved, making it more suitable for spinal images of different sources and types.

[0035] Specifically, the formula for calculating the feature loss function is as follows: Among them, Ss is the feature loss function, Z gut is the t-th predicted coordinate information of the u-th vertebral region in the g-th spine trajectory data to be analyzed predicted by the model, is the tth actual coordinate information of the uth vertebral region in the gth trajectory data of the spine to be analyzed, g=1,2,3,…,G, G is the number of obtained trajectory data of the spine to be analyzed, u=1,2,3,…,U, U is the number of vertebral regions in the obtained trajectory data of the spine to be analyzed, t=1,2,3,…,T, T is the number of coordinate points in the vertebral region in the obtained trajectory data of the spine to be analyzed.

[0036] In this embodiment, the calculation feature loss function logic provides an effective loss metric by comparing the difference between the pixel coordinates of the vertebral region predicted by the model and the actual coordinates, which helps the model adjust parameters to reduce prediction errors, helps train the model to better predict the key points of each vertebral region in the spinal image, thereby improving the accuracy of key point detection, and introduces quantitative statistical items, taking into account the number of trajectory data of the spine to be analyzed and the number of vertebral regions, which helps to balance the contributions of different data scales and vertebral regions, prevents the model from focusing too much on certain parts, makes the loss function more general, adapts to spinal image data of different scales and complexities, and improves the adaptability of the model.

[0037] Specifically, Figure 3 As shown, the analysis and calculation subsystem includes: a Cobb angle measurement and calculation module, an ATR measurement and calculation module, a rotation angle analysis module, and a thoracic curve and lumbar curve analysis module; the Cobb angle measurement and calculation module is used to calculate the Cobb angle based on the coordinate information of the vertex of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed; the ATR measurement and calculation module is used to calculate the ATR based on the coordinate information of the vertex of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed, wherein the ATR represents the ratio of the lateral translation of the highest point to the entire scoliosis height; the rotation angle analysis module is used to calculate the rotation angle of the spine based on the trajectory data of the spine to be analyzed; the thoracic curve and lumbar curve analysis module is used to analyze the physiological curvature of the spine in the sagittal plane, namely the thoracic curve and lumbar curve, based on the trajectory data of the spine to be analyzed.

[0038] In this implementation scheme, the Cobb angle is a key indicator for evaluating the degree of scoliosis. By accurately calculating the Cobb angle, the patient's scoliosis condition can be evaluated more accurately. The Cobb angle provides doctors with a quantitative assessment of the patient's spinal lesions, providing important support for medical decision-making. The coordinate information of the top point of the upper vertebral area and the bottom point of the lower vertebral area is obtained from the trajectory data, and the ATR is obtained by calculating the ratio of the lateral translation of the highest point to the entire scoliosis height. ATR provides more comprehensive information about the scoliosis morphology, taking into account the lateral translation, which helps to have a more comprehensive understanding of the patient's spinal condition. ATR can provide important information for doctors to formulate more personalized treatment plans. Because different ATRs may require different treatment strategies, the spinal trajectory data to be analyzed is used to calculate the rotation angle of the spine. The rotation angle provides information about the spinal posture, which helps to fully understand the three-dimensional morphology of the patient's spine. For patients who need surgical correction, the rotation angle provides a basis for surgical planning, enabling doctors to better predict the surgical effect. The trajectory data is used to analyze the physiological curvature of the spine in the sagittal plane, namely the thoracic curve and lumbar curve. The thoracic curve and lumbar curve provide detailed information about the curvature of the spine in the sagittal plane, which provides support for the comprehensive evaluation of the overall spinal morphology. For patients who need conservative treatment, the information on the thoracic curve and lumbar curve can provide a basis for formulating a spinal rehabilitation plan.

[0039] Specifically, Figure 4 , Figure 5 As shown in FIG. 1 , the specific steps for calculating the Cobb angle are as follows: read the coordinate information of each coordinate point of the vertex of the upper vertebral region and the lower end bottom point of the lower vertebral region in the image data of the spine to be analyzed, and record them as point A and point B respectively; based on computer image processing technology, use points A and B to draw vertical lines perpendicular to the upper edge of the upper vertebral region and the lower edge of the lower vertebral region respectively, and record the intersection of the two vertical lines as point C; take the midpoints of line segments AC and BC respectively, and draw vertical lines perpendicular to the midpoints of line segments AC and BC respectively; the intersection angle of the two vertical lines is the Cobb angle.

[0040] In this implementation, the coordinate information of the upper vertebral area and the lower vertebral area (point A and point B) is read, and the position information of the key points in the spinal image data to be analyzed is provided, which is the basis for subsequent calculations. Image processing technology is used to accurately find the intersection C of the vertical lines of the two vertebral areas, and establish a geometric reference for Cobb angle calculation. The midpoint of the two vertical lines is established, which provides an additional geometric structure for angle calculation and increases the stability of the calculation. By calculating the intersection angle of the two vertical lines, the specific value of the Cobb angle is obtained, which is a measure of the degree of scoliosis and provides an objective indicator for the evaluation of spinal diseases.

[0041] Specifically, Figure 6As shown, the specific steps for calculating ATR are as follows: read the coordinate information of point A and point B; calculate the length and height difference on the vertebral axis based on the coordinate information of point A and point B; calculate ATR based on the length and height difference of the spinal column axis, and the calculation formula is as follows: in, is the length of the vertebral axis, |y B -y A | is the height difference.

[0042] In this embodiment, the coordinate information of point A and point B on the image is obtained by computer image processing technology. This step provides the position information of the key points and provides a geometric basis for subsequent calculations. The length and height difference on the vertebral axis can be calculated using the coordinate information of point A and point B. These two values ​​are important parameters for evaluating spinal morphology and help to quantify the spatial relationship between vertebrae. The length and height difference on the vertebral axis are used to calculate ATR (vertebral rotation angle) using a predefined calculation formula. This angle reflects the degree of rotation of the vertebra and is an important indicator for evaluating the degree of spinal deformity. The calculation formula for ATR is: This formula uses the inverse tangent function to calculate the vertebral rotation angle, converting the length and height information measured in the image into a specific angle measurement. By calculating the length and height difference on the vertebral axis and calculating the ATR, the system can provide numerical and objective evaluation indicators to help doctors understand the morphological characteristics of the spine more accurately. ATR, as a measure of spinal rotation angle, helps to detect and diagnose spinal deformities early, providing earlier treatment and intervention opportunities, and providing detailed numerical information on the patient's spinal status, which helps doctors develop personalized treatment plans, including surgical decisions, rehabilitation plans, etc. Continuous monitoring of ATR can track changes in spinal deformities, evaluate treatment effects, and provide a basis for long-term follow-up.

[0043] Specifically, Figure 7 As shown, the specific steps for calculating the rotation angle of the spine are as follows: read the coordinate information of each coordinate point in each vertebral area respectively, regard the coordinate midpoint of each cone as the center of each cone, and regard it as their respective local coordinate systems; obtain the posture information of each cone based on the trajectory data of the spine to be analyzed, and the posture information is recorded as a direction vector; based on the rotation matrix, convert the coordinate information and posture information of each coordinate point in the local coordinate system of each vertebral area to the global coordinate system, and obtain the coordinate information and posture information of each coordinate point of each vertebra in the global coordinate system; and extract the rotation angle of the spine based on the rotation matrix.

[0044] In this implementation, the coordinate information of each coordinate point in each vertebral region is read one by one, which provides a basis for the spinal trajectory data and provides accurate spatial position information for subsequent analysis and calculation. The midpoint of each cone is used as the origin of its own local coordinate system, and the local coordinate system of the vertebra is aligned with the vertebra itself, which simplifies the subsequent processing of posture information and makes the coordinate conversion more intuitive. The posture information of each vertebra is obtained through the spinal trajectory data to be analyzed, which is usually expressed as a direction vector, and the posture information of each vertebra in the local coordinate system is provided, which is helpful for the subsequent coordinate transformation and calculation of the rotation angle. The rotation matrix is ​​used to convert the coordinate points and posture information in the local coordinate system of each vertebral region to the global coordinate system, realizing accurate conversion from the local coordinate system to the global coordinate system, providing a basis for describing the overall morphology of the spine in the global coordinate system, and extracting the rotation angle of the spine based on the information in the rotation matrix. The rotation angle provides information about the overall rotation posture of the spine, which is of great significance for understanding the three-dimensional morphology and posture of the spine.

[0045] Specifically, Figure 8 As shown in FIG. 1 , the specific steps of analyzing the thoracic and lumbar curves of the spine are as follows: read the coordinate information of each lateral edge point of each vertebral region; bring the coordinate information of each lateral edge point of each vertebral region into the curvature calculation model, obtain the curvature value of the coordinate information of each lateral edge point of each vertebral region, and compare it with the set thoracic curve threshold and lumbar curve threshold respectively; record the part of each lateral edge point whose curvature value of the coordinate information is greater than the thoracic curve threshold as the thoracic curve, and record the part of each lateral edge point whose curvature value of the coordinate information is less than the lumbar curve threshold as the lumbar curve; wherein the curvature calculation model is specifically: Among them, K rm is the curvature value of the coordinate information of the m lateral edge points in the rth vertebral area, y″ rm The coordinate information of the m lateral edge points in the rth vertebral region is brought into the second-order derivative of y with respect to x in the curve model y=f(x), y′ rm Substitute the coordinate information of the m lateral edge points in the rth vertebral region into the first-order derivative of y with respect to x in the curve model y=f(x), r=1,2,3,…,C, C is the number of vertebral regions, m=1,2,3,…,F, F is the number of coordinate points in the vertebral region.

[0046] In this implementation, each vertebral area is traversed, and the coordinate information of each lateral edge point is read, so as to provide accurate position information of the lateral edge points of the spine, provide basic data for curvature calculation and morphological analysis, and bring the coordinate information of each lateral edge point of each vertebral area into the curvature calculation model, and calculate the curvature value of each lateral edge point, which reflects the degree of curvature of the lateral edge point of the spine, and helps to quantify the physiological curvature of the spine in the sagittal plane, and compare the curvature value of each lateral edge point with the set thoracic curve threshold and lumbar curve threshold to determine whether it belongs to thoracic curve, lumbar curve or normal curvature, thereby realizing quantitative classification of scoliosis areas, enabling doctors to understand the thoracic curve and lumbar curve of patients more accurately, and determining the lateral edge points with curvature values ​​greater than the thoracic curve threshold as thoracic curve, and determining the lateral edge points with curvature values ​​less than the lumbar curve threshold as lumbar curve, so as to provide It provides detailed information on the curvature distribution of the spine in the sagittal plane, providing more comprehensive data support for disease analysis. The curvature calculation model uses the second-order derivative and first-order derivative in the curve model, and calculates the rate of change of the curve to obtain the curvature value of the lateral edge point. The curvature is calculated using a mathematical model, which makes the curvature value more mathematically meaningful and enhances the accurate description of the physiological curvature. Through the curvature calculation model, the system provides a comprehensive evaluation of the curvature of the lateral edge points of the spine, including the classification of specific values ​​and curvature types. Through the setting and comparison of thresholds, the system quantitatively analyzes the thoracic and lumbar curvatures of the spine, enabling doctors to understand the patient's scoliosis more clearly, and provides curvature-based data support to help doctors develop more personalized treatment plans, including surgery and rehabilitation plans.

[0047] Specifically, the formula for calculating the evaluation index is as follows: PGz=α1*Cobb′+α2*ATR+α3*Xz; wherein PGz is the evaluation index, Cobb′ is the Cobb angle, Xz is the rotation angle, α1, α2, α3 are the proportional coefficients of Cobb′, ATR, and Xz respectively, and α1+α2+α3=1.

[0048] In this implementation scheme, the logic for calculating the evaluation index is: summing up the Cobb angle, ATR, and rotation angle of the spine to obtain the sum value, then respectively performing ratio calculations on the Cobb angle, ATR, and rotation angle of the spine and the sum value to obtain the proportional coefficients of the Cobb angle, ATR, and rotation angle of the spine, and then calculating the evaluation index based on the Cobb angle, ATR, and rotation angle of the spine and their respective proportional coefficients. This logic combines multiple indicators in a sum value manner, so that the evaluation index considers all aspects of the spine more comprehensively and globally, rather than being limited to a single indicator. By calculating the ratio, the proportional coefficient is introduced, and the weight of each indicator in the evaluation can be adjusted according to actual needs, so that the evaluation is more in line with actual application scenarios and medical needs. The logic design is highly flexible and can be adjusted according to specific problems and needs. For example, if the accuracy of the Cobb angle is required to be higher, its weight in the evaluation index can be increased by increasing its proportional coefficient. Different indicators are associated with their proportional coefficients, which better reflects the overall characteristics of the spinal morphology and helps medical professionals to fully understand the patient's condition. Through the logical design, the calculation process of the evaluation index is more transparent and explainable, which helps medical professionals understand the formation principle of the evaluation results.

[0049] Specifically, the analysis of the spine based on the calculated evaluation indicators is as follows: the evaluation indicators are subjected to standard processing, and the above-mentioned standard processing is: the evaluation indicators are converted into percentages, and the percentage range is specifically: 0≤PGz%≤100%; when the evaluation indicator is between 90% and 100%, the spine is regarded as normal; when the evaluation indicator is between 70% and 89%, the spine is regarded as slightly abnormal, and an early warning prompt is issued, and the early warning prompt is specifically: change the sitting posture, maintain a correct standing posture, and perform appropriate physical exercise; when the evaluation indicator is between 50% and 69%, the spine is regarded as moderately abnormal, and the Cobb′ and ATR of the spine are further analyzed, and an autonomous rehabilitation treatment plan is automatically generated; when the evaluation indicator is between 0% and 49%, the spine is regarded as severely abnormal, and an alarm message is issued at the same time.

[0050] In this implementation scheme, the evaluation indicators are standardized and converted into percentages ranging from 0% to 100%. Such standardized processing is conducive to comparing the results of different indicators and improves the interpretability of the results. The spinal status is divided into different ranges, which is convenient for medical professionals and patients to understand the evaluation results. Each range corresponds to a certain spinal status, making the results more intuitive and easy to communicate. Specific suggestions for spinal status are provided: 90% to 100%: normal status, 70% to 89%: mild abnormality, and early warning prompts are issued. The early warning prompts include suggestions to change sitting posture, maintain correct standing posture and perform appropriate physical exercise, which helps to remind patients to take some simple lifestyle adjustments to prevent the problem from further deteriorating. 50 % to 69%: moderate abnormalities. For moderate abnormalities, further analysis is carried out with special attention to specific aspects of the spine. At the same time, the system automatically generates an autonomous rehabilitation treatment plan to provide patients with personalized rehabilitation suggestions and improve rehabilitation effects. 0% to 49%: severe abnormalities. For severe abnormalities, an alarm message is issued to prompt medical staff or patients to take emergency actions, which helps to intervene and treat in time to prevent the problem from further deteriorating. For abnormal conditions, timely warnings and suggestions provide opportunities for treatment, which is of great significance to the patient's health and rehabilitation and helps prevent further progression of the disease. In the face of moderate abnormalities, the system generates an autonomous rehabilitation treatment plan, which helps to provide more specific and personalized rehabilitation suggestions and provide patients with a better rehabilitation experience and effect.

[0051] In summary, this application has at least the following effects:

[0052] Through this spinal deformity analysis system based on key point detection, one can more intuitively understand the health status of the spine. The evaluation indicators and early warning prompts generated by the system help improve users' awareness of spinal health, prompting people to pay attention to and adopt a positive lifestyle and posture to maintain the normal physiological state of the spine.

[0053] For mild and moderate abnormal spinal conditions, the system's early warning prompts and automatically generated rehabilitation treatment plans provide individuals with timely and effective rehabilitation suggestions. Users can adjust their sitting and standing postures and participate in specific physical exercises based on system feedback to improve spinal health. This real-time personalized rehabilitation intervention is expected to prevent or slow the further development of spinal abnormalities in the early stages.

[0054] For serious abnormal spinal conditions, the system's alarm information reminds users to seek medical help in time. This helps speed up the treatment of emergency spinal problems and improve the efficiency of medical resource utilization. By providing detailed evaluation indicators, medical professionals can understand the patient's condition more quickly and accurately, provide emergency treatment for patients, and minimize the patient's spinal health risks.

[0055] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0056] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A spinal deformity analysis system based on key point detection, It is characterized in that include: Spinal surface morphology acquisition subsystem, key point analysis subsystem, analysis and calculation subsystem, evaluation index calculation subsystem, data analysis subsystem, and central control platform; The spine surface morphology acquisition subsystem is used to acquire the morphology data of the spine on the human body surface; The key point analysis subsystem is used to process and feature analyze the information acquired by the spinal surface morphology acquisition subsystem, obtain the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, and use the vertebral vertex of the upper end vertebral area and the vertebral vertex of the lower end vertebral area and the side edge points of each vertebral area as key points; The analysis and calculation subsystem is used to perform posture analysis and calculation on the coordinates and posture information of each vertebral area of ​​the spine to be analyzed to obtain the projection information of the spine in the sagittal plane, coronal plane, and cross-sectional plane, and analyze and calculate the Cobb angle, ATR, rotation angle, thoracic curve, and lumbar curve information; The evaluation index calculation subsystem is used to calculate the evaluation index based on the output data of the analysis and calculation subsystem. The formula for calculating the evaluation index is as follows: PGz=a 1 *Cobb′+a 2 *ATR+a 3 *Xz; Among them, PGz is the evaluation index, Cobb′ is the Cobb angle, Xz is the rotation angle, α 1 , α 2 , α 3 are the proportional coefficients of Cobb′, ATR, and Xz, respectively. The Cobb angle, ATR, and rotation angle are summed to obtain the sum value, and then the Cobb angle, ATR, and rotation angle are respectively calculated to calculate the ratio of the sum value to obtain the proportional coefficients of the Cobb angle, ATR, and rotation angle, and α 1 +α 2 +α 3 =1; The data analysis subsystem is used to analyze the spine based on the calculated evaluation index; The central control platform is used to control the operation of the entire system and coordinate the operation between the subsystems.

2. A spinal deformity analysis system based on key point detection according to claim 1, Features: The specific steps for obtaining the coordinates and posture information of each vertebral region of the spine to be analyzed are as follows: Acquire trajectory data of several spines to be analyzed and construct a trajectory data set; The trajectory dataset is divided into a training dataset and a validation dataset; Read the training data set to train the constructed spinal vertebra detection learning network model, and calculate the feature loss function; Use the validation data set to evaluate the model performance and adjust the model parameters based on the validation results until the model predicts feature data that matches the actual feature data values. The input and output data of the constructed spinal vertebrae detection learning network model are the trajectory data of the spine to be analyzed and the coordinates and posture information of each vertebral area of ​​the spine to be analyzed, respectively. The constructed spinal vertebrae detection learning network model is a deep learning model with a densely connected convolutional network as the architecture and multi-scale feature extraction as a variant. The deep learning model includes: input layer, densely connected convolutional layer, activation function, batch normalization layer, global average pooling layer, transition layer, fully connected layer, Dropout layer, Softmax layer, loss function, and optimizer.

3. A spinal deformity analysis system based on key point detection according to claim 2, Features: The formula for calculating the feature loss function is as follows: Among them, Ss is the feature loss function, Z gut is the t-th predicted coordinate information of the u-th vertebral region in the g-th spine trajectory data to be analyzed predicted by the model, is the tth actual coordinate information of the uth vertebral region in the gth trajectory data of the spine to be analyzed, g=1,2,3,…,G, G is the number of obtained trajectory data of the spine to be analyzed, u=1,2,3,…,U, U is the number of vertebral regions in the obtained trajectory data of the spine to be analyzed, t=1,2,3,…,T, T is the number of coordinate points in the vertebral region in the obtained trajectory data of the spine to be analyzed.

4. A spinal deformity analysis system based on key point detection according to claim 1, Features: The analysis and calculation subsystem includes: a Cobb angle measurement and calculation module, an ATR measurement and calculation module, a rotation angle analysis module, and a thoracic curve and lumbar curve analysis module; The Cobb angle measurement and calculation module is used to calculate the Cobb angle based on the coordinate information of the top point of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed; The ATR measurement and calculation module is used to calculate the ATR based on the coordinate information of the top point of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed, wherein the ATR represents the ratio of the lateral translation of the highest point to the entire scoliosis height; The rotation angle analysis module is used to calculate the rotation angle of the spine based on the trajectory data of the spine to be analyzed; The thoracic curve and lumbar curve analysis module is used to analyze the physiological curvature of the spine in the sagittal plane, namely the thoracic curve and the lumbar curve, based on the trajectory data of the spine to be analyzed.

5. A spinal deformity analysis system based on key point detection according to claim 4, Features: The specific steps to calculate the Cobb angle are as follows: Read the coordinate information of the top point of the upper vertebral area and the bottom point of the lower vertebral area in the trajectory data of the spine to be analyzed, and record them as point A and point B respectively; Based on computer image processing technology, vertical lines perpendicular to the upper edge of the upper vertebral area and the lower edge of the lower vertebral area are drawn from points A and B respectively, and the intersection of the two vertical lines is recorded as point C; Take the midpoints of line segments AC and BC respectively, and draw perpendicular lines to the midpoints of line segments AC and BC respectively; The angle between two perpendicular lines is the Cobb angle.

6. A spinal deformity analysis system based on key point detection according to claim 5, Features: The specific steps for calculating ATR are as follows: Read the coordinate information of point A and point B; Based on the coordinate information of point A and point B, the length and height difference on the vertebral axis are calculated respectively; The ATR is calculated based on the spinal axis length and height difference, and the calculation formula is as follows: in, is the length of the vertebral axis, |y B -y A | is the height difference.

7. A spinal deformity analysis system based on key point detection according to claim 6, Features: The specific steps for calculating the rotation angle of the spine are as follows: The coordinate information of each coordinate point of each vertebral region is read respectively, and the coordinate midpoint of each cone is regarded as the center of each cone, and regarded as the respective local coordinate system; Acquire posture information of each cone based on the trajectory data of the spine to be analyzed, wherein the posture information is recorded as a direction vector; Based on the rotation matrix, the coordinate information and posture information of each coordinate point in the local coordinate system of each vertebral area are converted to the global coordinate system to obtain the coordinate information and posture information of each coordinate point of each vertebral body in the global coordinate system; And extract the rotation angle of the spine based on the rotation matrix.

8. A spinal deformity analysis system based on key point detection according to claim 7, Features: The specific steps for analyzing the thoracic and lumbar curves of the spine are as follows: Read the coordinate information of each lateral edge point of each vertebral area; The coordinate information of each lateral edge point of each vertebral region is brought into the curvature calculation model to obtain the curvature value of the coordinate information of each lateral edge point of each vertebral region, and compared with the set thoracic curve threshold and lumbar curve threshold respectively; Each side edge point portion whose curvature value of coordinate information is greater than the thoracic curve threshold is recorded as the thoracic curve, and each side edge point portion whose curvature value of coordinate information is less than the lumbar curve threshold is recorded as the lumbar curve; The curvature calculation model is as follows: Among them, K rm is the curvature value of the coordinate information of the m lateral edge points in the rth vertebral area, y′ r ' m The coordinate information of the m lateral edge points in the rth vertebral region is brought into the second-order derivative of y with respect to x in the curve model y=f(x), y′ rm Substitute the coordinate information of the m lateral edge points in the rth vertebral region into the first-order derivative of y with respect to x in the curve model y=f(x), r=1,2,3,…,C, C is the number of vertebral regions, m=1,2,3,…,F, F is the number of coordinate points in the vertebral region.

9. The spinal deformity analysis system based on key point detection according to claim 1, Features: The spine is analyzed based on the calculated evaluation indicators as follows: The evaluation index is subjected to standard processing, and the above standard processing is: the evaluation index is converted into a percentage, and the specific range of the percentage is: 0≤PGz%≤100%; When the evaluation index is between 90% and 100%, the spine is considered normal; When the evaluation index is between 70% and 89%, the spine is considered to be slightly abnormal and an early warning prompt is issued, which specifically includes: changing the sitting posture, maintaining a correct standing posture, and performing appropriate physical exercises; When the evaluation index is between 50% and 69%, the spine is considered moderately abnormal, and the Cobb' and ATR of the spine are further analyzed, and an autonomous rehabilitation treatment plan is automatically generated; When the evaluation index is between 0% and 49%, the spine is considered to be severely abnormal and an alarm message is issued at the same time.

Citation Information

Cited By

  • A medical interactive system for vertebra

    CN122929066A