A cervical vertebrae bone age analysis method based on deep learning

Abstract

The present invention discloses a cervical vertebrae bone age analysis method based on deep learning, which relates to the technical field of cervical vertebrae bone age prediction. The method constructs a data set with real clinical cephalometric lateral radiograph data, integrates multiple image processing models, uses a key point detection model to detect bone joint key points, extracts bone joint features through segmentation results and key point detection results, and inputs the features into a binary tree support vector machine for bone age staging. The method extracts deeper bone joint features by combining an instance segmentation model with a key point detection model, thereby improving the accuracy and stability of cervical vertebrae bone age staging. The process is fully automated and highly efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of cervical vertebrae bone age prediction, and in particular, to a cervical vertebrae bone age analysis method based on deep learning. Background Art

[0002] In the field of stomatology, virtual orthodontic systems developed by combining computer graphics, digital media, and graphics and image processing technologies are assisting dentists in diagnosis and treatment. Cervical vertebrae bone age analysis is an important evaluation method in orthodontics. By analyzing the patient's bone age, doctors can better understand the patient's bone growth and development, and thus design a more accurate orthodontic treatment plan. In oral and maxillofacial surgery, cervical vertebrae bone age analysis can also be used to evaluate the patient's bone development. Doctors can better understand the patient's bone development and thus choose a more appropriate surgical method.

[0003] However, existing methods for analyzing cervical vertebrae bone age rely on the doctor's personal experience, and the bone age estimation results are highly dependent on the doctor's subjective judgment and may be affected by image texture information.

[0004] At present, some methods have been proposed to realize automated bone age analysis, but these methods have not yet been able to solve the following problems:

[0005] 1) Most bone age analysis methods rely on standardized laboratory data, and the model generalization performance is poor, so they cannot be applied to real clinical data;

[0006] 2) The bone age analysis results of a single model are unstable and cannot meet the reliability requirements of computer-assisted diagnosis and treatment functions. Summary of the invention

[0007] In order to solve the above problems and accurately obtain the cervical vertebrae bone age, the present invention provides a cervical vertebrae bone age analysis method based on deep learning, the method comprising:

[0008] S1: Construct a cervical vertebrae bone age analysis system;

[0009] S2: obtaining a lateral cephalogram image to be processed, inputting the lateral cephalogram image to be processed into a cervical vertebra bone age analysis system for analysis and processing, and obtaining a cervical vertebra bone age analysis result corresponding to the lateral cephalogram image to be processed;

[0010] The cervical vertebrae bone age analysis system includes: a preprocessing unit, a region of interest extraction model, a joint segmentation integrated model, a key point detection integrated model, a processing unit and a cervical vertebrae bone age staging model;

[0011] The S1 includes:

[0012] S11: constructing a preprocessing unit, acquiring a cephalometric lateral radiograph image set, and preprocessing the cephalometric lateral radiograph image set based on the preprocessing unit;

[0013] S12: training the first type of neural network using the lateral cephalogram image set to obtain a region of interest extraction model; the region of interest extraction model is used to extract a region of interest image including a cervical spine region from the lateral cephalogram image;

[0014] S13: using the region of interest image to train N second-class neural networks respectively to obtain a joint segmentation integrated model, wherein the joint segmentation integrated model is used to segment a cervical vertebra image containing edge information from the region of interest image, where N is an integer greater than 1;

[0015] S14: using the region of interest image to train M third-class neural networks respectively to obtain a key point detection integrated model; the key point detection integrated model is used to extract measurement landmarks of cervical vertebrae from the region of interest image, where M is an integer greater than 1;

[0016] S15: based on the processing unit, based on the cervical vertebra image containing edge information, using a sliding window to intercept the lower edge area of ​​the cervical vertebra and calculate the curvature vector of the cervical vertebra, and calculating the curvature score of the cervical vertebra through the curvature vector of the cervical vertebra;

[0017] S16: generating measurement features of the cervical region based on the measurement landmarks of the cervical vertebrae by the processing unit, and forming a comprehensive feature of the cervical region based on the measurement features of the cervical region and the curvature scores of the cervical vertebrae;

[0018] S17: Using the comprehensive characteristics of the cervical vertebrae region to train a binary tree support vector machine to obtain a cervical vertebrae bone age staging model, wherein the cervical vertebrae bone age staging model is used to stage the cervical vertebrae according to the comprehensive characteristics of the cervical vertebrae region.

[0019] Among them, most of the current bone age analysis methods rely on standardized laboratory data, have poor model generalization performance, and cannot be applied to real clinical data. However, this method uses lateral cephalogram images and does not require standardized laboratory data. The model has strong generalization performance and can be applied to real clinical data. The bone age analysis results of traditional methods with a single model are unstable and cannot meet the reliability requirements of computer-assisted diagnosis and treatment functions. This method uses a cervical vertebra bone age analysis system, and the cervical vertebra bone age analysis system includes a focus area extraction model, a bone segmentation integrated model, a key point detection integrated model, a processing unit and a cervical vertebra bone age staging model. The combination and collaborative processing of multiple models can accurately and stably obtain cervical vertebra bone age analysis results.

[0020] Preferably, the preprocessing of the lateral cephalogram image set includes: unifying the format of the lateral cephalogram images and converting the images into RGB format; and normalizing the lateral cephalogram images to reduce the impact of differences between image domains.

[0021] Preferably, the joint segmentation integrated model segments the cervical vertebra image containing edge information from the region of interest image in the following manner:

[0022] The images of the region of interest are respectively input into N second-class neural networks of the joint segmentation integrated model, and N segmentation map results are output;

[0023] According to the N segmentation results, a pixel-level voting mechanism is used to select a homogeneous image area as the final segmentation result to obtain a cervical vertebra image containing edge information.

[0024] Preferably, in order to accurately obtain the semantic category of the final segmentation, the method adopts the following method for calculation. The second type of neural network model refers to a variety of different segmentation network models. The average value of the segmentation results of all segmentation network models can be taken by the following formula to make the semantic category of the final segmentation more accurate. The calculation method of the final segmentation result is:

[0025]

[0026] Where c∈D represents the semantic category of the final segmentation; D represents the set of pixels in the image of the region of interest; argmax(·) represents the index operation of taking the maximum value; softmax(·) represents the normalized exponential function; p i Represents the prediction result of the i-th second-class neural network model, i=1...N, i is the index variable.

[0027] Preferably, the measurement landmarks of the cervical vertebrae include: the front end point and the rear end point of the lower edge of the second vertebrae, the connection point of the upper edge and the front edge of the third and fourth vertebrae, the midpoint of the upper edge of the third and fourth vertebrae, the connection point of the upper edge and the rear edge of the third and fourth vertebrae, the connection point of the lower edge and the front edge of the third and fourth vertebrae, the midpoint of the lower edge of the third and fourth vertebrae, the connection point of the lower edge and the rear edge of the third and fourth vertebrae, the midpoint of the front edge and the midpoint of the rear edge of the third and fourth vertebrae. Wherein, by selecting the measurement landmarks of the above-mentioned cervical vertebrae, data such as various horizontal and vertical measurements and segment length-width ratios can be accurately calculated.

[0028] Preferably, the key point detection integrated model extracts the measurement landmarks of the cervical vertebrae from the region of interest image in the following manner:

[0029] The images of the region of interest are respectively input into M third-class neural networks of the key point detection integrated model, and M groups of initial measurement landmarks are output;

[0030] The measurement landmarks of the cervical vertebrae were calculated based on the initial measurement landmarks of group M using the average coordinate value method.

[0031] Preferably, the measurement landmarks of the cervical vertebrae are calculated as follows:

[0032]

[0033]

[0034]

[0035] Among them, x m is the coordinate value of the mth measurement landmark point of the cervical vertebra; is the initial coordinate value of the mth measurement mark point output by the jth third-class neural network, j = 1…M, j is the index variable, for The average value of indicates that the output of M neural networks is averaged to obtain an average coordinate value; h(·) is the outlier screening function; Δ is the outlier screening threshold; ||·|| is the Euclidean distance operation.

[0036] Among them, this method combines the model for segmentation and key point detection, there are multiple segmentation networks and key point networks, and the calculation method of the measurement landmark points of the cervical vertebrae takes the average of the output results, so that the final measurement result is accurate and contains a lot of measurement information.

[0037] Preferably, the S15 includes:

[0038] S151: Based on the cervical vertebra image containing edge information, the processing unit uses a sliding window to cut off the lower edge of the cervical vertebra according to a preset step length to obtain N vertebra lower edge segments;

[0039] S152: Calculate the curvature vector of each vertebral lower edge segment using the following calculation method:

[0040]

[0041] in, represents the curvature vector of the lower edge segment of the kth vertebra, w represents the size of the sliding window, The vector representing the midpoint of the lower edge segment of the k-th vertebra to the s-th point of the lower edge segment of the k-th vertebra, s=1…w, s is an index variable;

[0042] S153: Calculate the curvature score of the cervical vertebrae according to the curvature vectors of the lower edge segments of the N vertebrae, and the calculation method is:

[0043]

[0044] Where score represents the curvature score of the cervical vertebrae; n k represents the normal vector of the lower edge segment of the kth vertebra at the midpoint; sign represents the mathematical symbol function, and max represents the maximum value function.

[0045] Preferably, the measurement characteristics of the cervical region are calculated as follows:

[0046]

[0047]

[0048]

[0049]

[0050] Where T represents the measurement characteristics of the cervical spine area, represents the length-to-width ratio of the vertebra Co, o∈{3,4}, o is an index variable; represents the midpoint of the upper edge of vertebra Co; represents the midpoint of the lower edge of vertebra Co; represents the midpoint of the anterior edge of vertebra Co; represents the midpoint of the posterior edge of vertebra Co; represents the ratio of the anterior and posterior margins of vertebra Co; represents the junction of the superior and anterior margins of vertebra Co; represents the junction of the lower and anterior margins of vertebra Co; represents the junction of the superior and posterior margins of vertebra Co; represents the connection point between the lower and posterior edges of vertebra Co; μ represents the condyle matching coefficient; Indicates the anterior end point of the lower edge of vertebra C2; Indicates the posterior endpoint of the lower edge of vertebra C2.

[0051] Preferably, the cervical vertebrae include: a second vertebra, a third vertebra and a fourth vertebra, which are arranged from top to bottom along the cervical spine.

[0052] One or more technical solutions provided by the present invention have at least the following technical effects or advantages:

[0053] The present invention proposes a cervical vertebrae bone age analysis method based on deep learning. The method constructs a data set with real clinical cephalometric lateral radiograph data and integrates multiple image processing models, including: using Yolov5 to extract the region of interest, using Mask R-CNN, Solov2, BlendMask, multi-stage U-Net and other segmentation model instances to segment C2, C3, C4 joints, using HRNet, FCOS, Face-Net and other key point detection models to detect joint key points, extracting joint features such as edge morphology and curvature through the segmentation results and the key point detection results, and inputting the features into a binary tree support vector machine for bone age staging. The method extracts deeper joint features by combining the instance segmentation model with the key point detection model, thereby improving the accuracy and stability of cervical vertebrae bone age staging, and the process is fully automated with high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the present invention, and do not constitute a limitation on the embodiments of the present invention;

[0055] Figure 1 It is a schematic diagram of the construction process of the cervical vertebrae bone age analysis system in the present invention;

[0056] Figure 2 It is a schematic diagram of the measurement landmarks of the cervical vertebrae of the present invention;

[0057] Figure 3 It is a schematic diagram of the structure of the cervical vertebrae;

[0058] Among them, 1-the intersection point of the upper edge and the front edge, 2-the midpoint of the upper edge, 3-the intersection point of the upper edge and the rear edge, 4-the midpoint of the front edge, 5-the midpoint of the rear edge, 6-the intersection point of the upper edge and the front edge, 7-the midpoint of the lower edge, 8-the intersection point of the lower edge and the rear edge, 9-the second vertebra, 10-the third vertebra, 11-the fourth vertebra. DETAILED DESCRIPTION

[0059] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those within the scope of this description. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0061] Embodiment 1;

[0062] The present invention provides a cervical vertebrae bone age analysis method based on deep learning, the method comprising:

[0063] S1: Construct a cervical vertebrae bone age analysis system;

[0064] S2: obtaining a lateral cephalogram image to be processed, inputting the lateral cephalogram image to be processed into a cervical vertebra bone age analysis system for analysis and processing, and obtaining a cervical vertebra bone age analysis result corresponding to the lateral cephalogram image to be processed;

[0065] The cervical vertebrae bone age analysis system includes: a preprocessing unit, a region of interest extraction model, a joint segmentation integrated model, a key point detection integrated model, a processing unit and a cervical vertebrae bone age staging model;

[0066] The S1 includes:

[0067] S11: obtaining a lateral cephalogram image set, and preprocessing the lateral cephalogram image set;

[0068] S12: training the first type of neural network using the lateral cephalogram image set to obtain a region of interest extraction model; the region of interest extraction model is used to extract a region of interest image including a cervical spine region from the lateral cephalogram image;

[0069] In a preferred embodiment, the first type of neural network uses Yolov5, a target detection algorithm in deep learning, to process the preprocessed cephalometric lateral radiograph data and extract the region of interest image therein.

[0070] S13: using the region of interest image to train N second-class neural networks respectively to obtain a joint segmentation integrated model, wherein the joint segmentation integrated model is used to segment a cervical vertebra image containing edge information from the region of interest image;

[0071] In a preferred embodiment, the second type of neural network uses integrated bone segmentation models such as Mask R-CNN, Solov2, BlendMask and multi-stage U-Net neural network to segment the cervical spine area image containing edge information from the area of ​​interest image.

[0072] S14: using the region of interest image to train M third-class neural networks respectively, to obtain a key point detection integrated model; the key point detection integrated model is used to extract measurement landmarks of cervical vertebrae from the region of interest image;

[0073] In a preferred embodiment, the third type of neural network uses an integrated key point detection integrated model such as HRNet, FCOS, Face-Net neural network, etc. to extract measurement landmark points from the image of the region of interest.

[0074] S15: intercepting the lower edge area of ​​the cervical vertebrae using a sliding window according to the cervical vertebrae image containing edge information and calculating the curvature vector of the cervical vertebrae, and calculating the curvature score of the cervical vertebrae through the curvature vector of the cervical vertebrae;

[0075] S16: generating measurement features of the cervical region according to the measurement landmarks of the cervical vertebrae, and forming a comprehensive feature of the cervical region according to the measurement features of the cervical region and the curvature score of the cervical vertebrae;

[0076] S17: Using the comprehensive characteristics of the cervical vertebrae region to train a binary tree support vector machine to obtain a cervical vertebrae bone age staging model, wherein the cervical vertebrae bone age staging model is used to stage the cervical vertebrae according to the comprehensive characteristics of the cervical vertebrae region.

[0077] In this embodiment, the comprehensive characteristics of the cervical spine area are:

[0078]

[0079] Among them, the cervical spine can be divided into many age groups. By staging the cervical spine, the cervical spine can be divided into corresponding age groups to obtain the cervical spine bone age analysis results.

[0080] Preferably, the preprocessing of the lateral cephalogram image set includes: unifying the format of the lateral cephalogram images and converting the images into RGB format; and normalizing the lateral cephalogram images to reduce the impact of differences between image domains.

[0081] Preferably, the integrated bone segmentation model segments a cervical vertebra image containing edge information from the region of interest image, including:

[0082] The images of the region of interest are respectively input into N second-class neural networks of the joint segmentation integrated model, and N segmentation map results are output;

[0083] According to the N segmentation results, a pixel-level voting mechanism is used to select a homogeneous image area as the final segmentation result to obtain a cervical vertebra image containing edge information.

[0084] Among them, the pixel-level voting mechanism can be implemented by estimating 6DoF posture using PVNet-pixel-level voting network. The pixel-level voting mechanism is an existing election mechanism, and the embodiment of the present invention will not elaborate on the mechanism accordingly.

[0085] Preferably, the final segmentation result is calculated as follows:

[0086]

[0087] Where c∈D represents the semantic category of the final segmentation; D represents the set of pixels in the image of the region of interest; argmax(·) represents the index operation of taking the maximum value; softmax(·) represents the normalized exponential function; p i Represents the prediction result of the i-th second-class neural network model, i=1...N, i is the index variable.

[0088] See also Figure 2 Preferably, the measurement landmarks of the cervical vertebrae include: the front end point and the rear end point of the lower edge of the second vertebra C2, the connection point of the upper edge and the front edge, the midpoint of the upper edge, the connection point of the upper edge and the rear edge, the connection point of the lower edge and the front edge, the midpoint of the lower edge, the connection point of the lower edge and the rear edge, the midpoint of the front edge and the midpoint of the rear edge of the third vertebra C3 and the fourth vertebra C4.

[0089] Preferably, the key point detection integrated model is used to extract measurement landmarks of cervical vertebrae from the region of interest image, including:

[0090] The images of the region of interest are respectively input into M third-class neural networks of the key point detection integrated model, and M groups of initial measurement landmarks are output;

[0091] The measurement landmarks of the cervical vertebrae were calculated based on the initial measurement landmarks of group M using the average coordinate value method.

[0092] The method of averaging coordinate values ​​refers to the method of calculating the measurement landmarks of the cervical vertebrae using a corresponding average value calculation method or formula.

[0093] Preferably, the measurement landmarks of the cervical vertebrae are calculated as follows:

[0094]

[0095]

[0096]

[0097] Among them, x m is the coordinate value of the mth measurement landmark point of the cervical vertebra; is the initial coordinate value of the mth measurement mark point output by the jth third-class neural network, j = 1…M, j is the index variable, for The average value of indicates that the output of M neural networks is averaged to obtain an average coordinate value; h(·) is the outlier screening function; Δ is the outlier screening threshold; ||·|| is the Euclidean distance operation.

[0098] Preferably, the curvature score of the cervical vertebra is calculated as follows:

[0099] S151: using a sliding window to intercept the lower edge of the cervical vertebra according to a preset step length to obtain N lower edge segments of the vertebra;

[0100] S152: Calculate the curvature vector of each vertebral lower edge segment:

[0101]

[0102] in, represents the curvature vector of the lower edge segment of the kth vertebra, w represents the size of the sliding window, The vector representing the midpoint of the lower edge segment of the k-th vertebra to the s-th point of the lower edge segment of the k-th vertebra, s=1…w, s is an index variable;

[0103] S153: Calculate the curvature score of the cervical vertebrae according to the curvature vectors of the lower edge segments of the N vertebrae:

[0104]

[0105] Where score represents the curvature score of the cervical vertebrae; n k represents the normal vector of the lower edge segment of the kth vertebra at the midpoint; sign represents a mathematical symbol function, and max represents a maximum value function. The cervical vertebrae include: a second vertebra C2, a third vertebra C3, and a fourth vertebra C4.

[0106] Preferably, the measurement characteristics of the cervical spine region are calculated as follows:

[0107]

[0108]

[0109]

[0110]

[0111] Where T represents the measurement characteristics of the cervical spine area, represents the length-to-width ratio of the vertebra Co, o∈{3,4}, o is an index variable; represents the midpoint of the upper edge of vertebra Co; represents the midpoint of the lower edge of vertebra Co; represents the midpoint of the anterior edge of vertebra Co; represents the midpoint of the posterior edge of vertebra Co; represents the ratio of the anterior and posterior margins of vertebra Co; represents the junction of the superior and anterior margins of vertebra Co; represents the junction of the lower and anterior margins of vertebra Co; represents the junction of the superior and posterior margins of vertebra Co; represents the connection point between the lower and posterior edges of vertebra Co; μ represents the condyle matching coefficient; Indicates the anterior end point of the lower edge of vertebra C2; Indicates the posterior endpoint of the lower edge of vertebra C2.

[0112] Please refer to Figure 3 , Figure 3 It is a schematic diagram of the structure of cervical vertebrae, wherein the cervical vertebrae include: a second vertebra, a third vertebra and a fourth vertebra, and the second vertebra, the third vertebra and the fourth vertebra are arranged from top to bottom along the cervical spine.

[0113] 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.

[0114] 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 cervical vertebrae bone age analysis method based on deep learning, characterized in that: The method comprises: S1: Construct a cervical vertebrae bone age analysis system; S2: obtaining a lateral cephalogram image to be processed, inputting the lateral cephalogram image to be processed into a cervical vertebra bone age analysis system for analysis and processing, and obtaining a cervical vertebra bone age analysis result corresponding to the lateral cephalogram image to be processed; The cervical vertebrae bone age analysis system includes: a preprocessing unit, a region of interest extraction model, a joint segmentation integration model, a key point detection integration model, a processing unit and a cervical vertebrae bone age staging model; The S1 includes: S11: constructing a preprocessing unit, acquiring a cephalometric lateral radiograph image set, and preprocessing the cephalometric lateral radiograph image set based on the preprocessing unit; S12: training the first type of neural network using the lateral cephalogram image set to obtain a region of interest extraction model; the region of interest extraction model is used to extract a region of interest image including a cervical spine region from the lateral cephalogram image; S13: using the region of interest image to train N second-class neural networks respectively to obtain a joint segmentation integrated model, wherein the joint segmentation integrated model is used to segment a cervical vertebra image containing edge information from the region of interest image, where N is an integer greater than 1; S14: using the region of interest image to train M third-class neural networks respectively, to obtain a key point detection integrated model; the key point detection integrated model is used to extract measurement landmarks of cervical vertebrae from the region of interest image, where M is an integer greater than 1; S15: based on the processing unit, based on the cervical vertebra image containing edge information, using a sliding window to intercept the lower edge area of ​​the cervical vertebra and calculate the curvature vector of the cervical vertebra, and calculating the curvature score of the cervical vertebra through the curvature vector of the cervical vertebra; S16: generating measurement features of the cervical region based on the measurement landmarks of the cervical vertebrae by the processing unit, and forming a comprehensive feature of the cervical region based on the measurement features of the cervical region and the curvature scores of the cervical vertebrae; S17: training a binary tree support vector machine using the comprehensive characteristics of the cervical vertebrae region to obtain a cervical vertebrae bone age staging model, wherein the cervical vertebrae bone age staging model is used to stage the cervical vertebrae according to the comprehensive characteristics of the cervical vertebrae region; The S15 includes: S151: Based on the cervical vertebra image containing edge information, the processing unit uses a sliding window to cut the lower edge of the cervical vertebra according to a preset step length to obtain R vertebra lower edge segments; S152: Calculate the curvature vector of each vertebral lower edge segment using the following calculation method: in, represents the curvature vector of the lower edge segment of the kth vertebra, w represents the size of the sliding window, The vector representing the midpoint of the lower edge segment of the k-th vertebra to the s-th point of the lower edge segment of the k-th vertebra, s=1…w, s is an index variable; S153: Calculate the curvature score of the cervical vertebrae according to the curvature vectors of the lower edge segments of the R vertebrae, and the calculation method is: Where score represents the curvature score of the cervical vertebrae; n k represents the normal vector of the lower edge segment of the kth vertebra at the midpoint; sign represents the mathematical symbol function, and max represents the maximum value function.

2. According to a method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, it is characterized in that: The preprocessing of the lateral cephalogram image set includes: unifying the format of the lateral cephalogram images and converting the images into RGB format; and normalizing the lateral cephalogram images.

3. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, characterized in that: The bone segmentation integrated model segments the cervical vertebra image containing edge information from the region of interest image in the following manner: The images of the region of interest are respectively input into N second-class neural networks of the joint segmentation integrated model, and N segmentation map results are output; According to the N segmentation map results, a pixel-level voting mechanism is used to select a homogeneous image area as the final segmentation result to obtain a cervical spine image containing edge information.

4. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 3, characterized in that: The final segmentation result is calculated as follows: Where c∈D represents the semantic category of the final segmentation; D represents the set of pixels in the image of the region of interest; argmax(·) represents the index operation of taking the maximum value; softmax(·) represents the normalized exponential function; p i Represents the prediction result of the i-th second-class neural network model, i=1...N, i is the index variable.

5. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, characterized in that: The measurement landmarks of the cervical vertebrae include: the front end point and the rear end point of the lower edge of the second vertebra, the connection point of the upper edge and the front edge of the third vertebra and the fourth vertebra, the midpoint of the upper edge of the third vertebra and the fourth vertebra, the connection point of the upper edge and the rear edge of the third vertebra and the fourth vertebra, the connection point of the lower edge and the front edge of the third vertebra and the fourth vertebra, the midpoint of the lower edge of the third vertebra and the fourth vertebra, the connection point of the lower edge and the rear edge of the third vertebra and the fourth vertebra, and the midpoint of the front edge and the midpoint of the rear edge of the third vertebra and the fourth vertebra.

6. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, characterized in that: The key point detection integrated model extracts the measurement landmarks of the cervical vertebrae from the region of interest image in the following manner: The images of the region of interest are respectively input into M third-class neural networks of the key point detection integrated model, and M groups of initial measurement landmarks are output; The measurement landmarks of the cervical vertebrae were calculated based on the initial measurement landmarks of group M using the average coordinate value method.

7. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 6, characterized in that: The calculation method of the measurement landmark points of the cervical vertebrae is: Among them, x m is the coordinate value of the mth measurement landmark point of the cervical vertebra; is the initial coordinate value of the mth measurement mark point output by the jth third-class neural network, j = 1…M, j is the index variable, for is the average value; h(·) is the outlier screening function; Δ is the outlier screening threshold; ||·|| is the Euclidean distance operation.

8. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, characterized in that: The measurement characteristics of the cervical spine region are calculated as follows: Where T represents the measurement characteristics of the cervical spine area, represents the length-to-width ratio of the vertebra Co, o∈{3,4}, o is an index variable; represents the midpoint of the upper edge of vertebra Co; represents the midpoint of the lower edge of vertebra Co; represents the midpoint of the anterior edge of vertebra Co; represents the midpoint of the posterior edge of vertebra Co; represents the ratio of the anterior and posterior margins of vertebra Co; represents the junction of the superior and anterior margins of vertebra Co; represents the junction of the lower and anterior margins of vertebra Co; represents the junction of the superior and posterior margins of vertebra Co; represents the connection point between the lower and posterior edges of vertebra Co; μ represents the condyle matching coefficient; Indicates the anterior end point of the lower edge of vertebra C2; Indicates the posterior endpoint of the lower edge of vertebra C2.

9. The method for analyzing cervical vertebrae bone age based on deep learning according to claim 1, characterized in that: The cervical vertebrae include: a second vertebra, a third vertebra and a fourth vertebra.

Images