Two-step spinal x-ray image processing method based on single-vertebral-body segmentation

Abstract

The present invention belongs to the technical field of medical image processing. Disclosed in the present invention is a two-step spinal X-ray image processing method based on single-vertebral-body segmentation, the method comprising: step S1, using a YOLOv8 object detection model to identify the position of each single spinal vertebral body in an X-ray image, and generating single-vertebral-body bounding boxes; step S2, performing detailed segmentation within each detected single-vertebral-body bounding box, so as to precisely delineate a spinal contour; and step S3, performing feature point identification within the single-vertebral-body bounding boxes, and using a YOLOv8 POSE model to identify key feature points of vertebrae, so as to further improve the precision of feature point identification. In the two-step spinal X-ray image processing method based on single-vertebral-body segmentation disclosed in the present invention, bounding boxes obtained by means of a YOLOv8 POSE model are imported into a U-Net network for segmentation, such that object detection and semantic segmentation are combined to improve the precision of spinal segmentation and the accuracy of identification.

Description

A two-step spinal X-ray processing method based on single vertebral body segmentation

[0001] This patent application claims priority to the following Chinese patent application:

[0002] Submission Date: July 17, 2024; Application Number: 202410959702.6; Invention Title: A Two-Step Spinal X-ray Processing Method Based on Single Vertebra Segmentation;

[0003] The full text of the above application is incorporated herein by reference. Technical Field

[0004] This invention belongs to the technical field of medical image processing, specifically relating to a two-step spinal X-ray processing method based on single vertebral body segmentation. Background Technology

[0005] Accurate segmentation of vertebral masses in whole-spine X-ray images is crucial for the intelligent diagnosis of spinal diseases. Traditional image segmentation methods mainly include threshold-based segmentation, edge detection-based segmentation, and region-based segmentation. The vertebral masses in spinal X-ray images often resemble the background in appearance, and factors such as parts of the same structure being covered or blurred by other tissues increase the difficulty of segmentation and recognition, posing a significant challenge to segmentation performance.

[0006] For spinal X-ray images with complex structures, traditional image segmentation methods often suffer from problems such as low recognition accuracy and insufficient segmentation precision.

[0007] Summary of the Invention

[0008] To address the low accuracy and insufficient segmentation precision of traditional image segmentation methods for spinal X-ray images, this invention proposes a two-step spinal X-ray image processing method based on single vertebral segmentation. This invention primarily imports the bounding boxes obtained from the YOLOv8 POSE model into the U-Net network for segmentation, combining object detection and semantic segmentation to improve the accuracy of spinal segmentation and recognition.

[0009] To achieve the above objectives, this invention discloses a two-step spinal X-ray image processing method based on single vertebral body segmentation, comprising the following steps:

[0010] Step S1: Use the YOLOv8 object detection model to identify the location of each individual spinal vertebra in the X-ray image and generate a single vertebra object bounding box;

[0011] Step S2: Perform detailed segmentation within each detected single vertebral body target box to accurately delineate the vertebral contour;

[0012] Step S3: Perform feature point recognition within the target bounding box of a single vertebra. Use the YOLOv8 POSE model to identify key feature points of the vertebrae, further improving the accuracy of feature point recognition.

[0013] As a further preferred technical solution to the above technical solution, step S1 is specifically implemented as follows:

[0014] Step S1.1: Perform image preprocessing. Before inputting the X-ray image into the YOLOv8 target detection model, perform preprocessing, including normalization, rotation, and cropping, to adapt to the model's input requirements.

[0015] Step S1.2: Perform feature extraction. Use the backbone network (CSPDarknet53) of the YOLOv8 target detection model to extract multi-scale features from the X-ray image. Use residual blocks to deepen the network and improve the feature extraction effect. And solve the gradient vanishing and gradient exploding problems by introducing skip connections.

[0016] Step S1.3: Perform feature fusion by using FPN (Feature Pyramid Network) and PANet (Path Aggregation Network) to fuse features at different scales, thereby improving the accuracy and robustness of target detection.

[0017] Step S1.4: Predict the bounding box. Use the YOLO Head module of the YOLOv8 object detection model to predict the bounding box, class confidence, and position offset on the feature map.

[0018] Step S1.5: Perform non-maximum suppression (NMS) to remove redundant boxes and retain only high-confidence boxes, ensuring that each target box uniquely corresponds to a vertebra category.

[0019] As a further preferred technical solution to the above technical solution, for step S1.3:

[0020] A top-down feature pyramid structure is constructed using FPN to extract multi-scale semantic information from feature maps at different levels, thereby addressing the problem of insufficient multi-scale information in object detection.

[0021] PANet introduces lateral connections and adaptive feature pooling mechanisms to facilitate information transfer and fusion between feature maps of different resolutions, thereby making better use of the multi-scale features generated by FPN.

[0022] As a further preferred technical solution to the above technical solution, for step S1.4:

[0023] Classification Head: The classification head predicts the class label by applying convolutional layers and activation functions (such as Softmax) to the feature map. The convolutional layer converts the information in the feature map into a score for each class, and the class with the highest score is the class label of the object in the bounding box.

[0024] The Softmax activation function transforms the raw scores (logits) into numerical values ​​representing a probability distribution, such that the probability value for each class is between 0 and 1, and the sum of the probabilities of all classes equals 1, as shown in the following formula:

[0025] Where e represents the base of the natural logarithm (Euler's number), n represents the number of categories, and z i It is the original score of the i-th category;

[0026] Regression head: Used to predict the coordinate parameters and confidence score of each target box to determine the position and size of the target box. The coordinate parameters include the center coordinates, width, and height of the target box. The model predicts these coordinate parameters through a regression algorithm to accurately locate the target. The confidence score indicates the model's confidence in whether the target box contains the target and is used to filter the detection results.

[0027] By combining the above classification and regression heads, the target's category, location, and confidence information can be accurately predicted on the feature map.

[0028] As a further preferred technical solution to the above technical solution, step S1.5 is specifically implemented as follows:

[0029] Step S1.5.1: Perform confidence screening. In the target boxes predicted by the model, first sort all target boxes according to their confidence scores and retain the target boxes with the highest confidence scores.

[0030] Step S1.5.2: Calculate the overlap. For the remaining target boxes, calculate the overlap with the target box with the highest confidence using the Intersection over Union (IoU) ratio.

[0031] Step S1.5.3: Perform NMS operation to remove target boxes with IoU higher than the set threshold from the remaining target boxes, ensuring that no redundant target boxes with large overlap with the target box with the highest confidence are retained. Repeat this process until all target boxes have been processed.

[0032] Step S1.5.4: Ensure uniqueness. After the NMS operation is completed, each retained target box should correspond to a unique target. If multiple target boxes highly overlap with the same target, NMS will select the target box with the highest confidence, while other target boxes will be suppressed.

[0033] As a further preferred technical solution to the above technical solution, step S2 is specifically implemented as follows:

[0034] Step S2.1: Prepare data by preparing X-ray images of individual spinal vertebrae (3000 cases) that have been classified. Standardize the data into images suitable for model input through operations including rotation and cropping.

[0035] Step S2.2: Perform data preprocessing. Preprocess the extracted target boxes, including scaling, cropping, and normalization, to meet the input requirements of the U-Net network.

[0036] Step S2.3: Construct the U-Net network architecture. This U-Net network takes a 1024×512×1 image as input. Each encoder layer includes two 3×3 convolutional layers, followed by instance normalization, ReLU activation function and 2×2 max pooling. Dropout is applied to each encoder-decoder stage of the network to generate side outputs at 128, 256 and 512 resolutions of the decoder, and then generate the final output at the original resolution.

[0037] Step S2.4: Train the network using the extracted target boxes as training data. During training, optimize the network parameters through backpropagation algorithm so that the network can accurately learn the contour and structure of the vertebrae.

[0038] Step S2.5: Perform segmentation prediction, input the extracted target box into the U-Net network to obtain pixel-level labels of the cone contour;

[0039] Step S2.6: Post-process the segmentation results by applying a region growing algorithm to fill holes and applying Gaussian filtering to remove noise and smooth edges, thereby obtaining more accurate and smoother segmentation results.

[0040] As a further preferred technical solution to the above technical solution, step S3 is specifically implemented as follows:

[0041] Step S3.1: Prepare data by preparing X-ray images of individual vertebrae of the spine with feature point markings (1000 cases). Standardize the data into images suitable for model input through operations including rotation and cropping.

[0042] Step S3.2: Perform data augmentation by performing operations such as random rotation, scaling, and flipping on the training data to increase the diversity of the data and help the model better generalize to different vertebral poses and angles;

[0043] Step S3.3 calculates the loss function, using mean squared error (MSE) and Euclidean distance between key points to ensure that the loss function effectively guides the model to learn accurate feature point locations;

[0044] Mean squared error (MSE) is the sum of the squared distances between the target variable and the predicted value;

[0045] Where ρ is the Euclidean distance between point (x2,y2) and point (x1,y1);

[0046] Step S3.4: Perform model training, using the pre-trained YOLOv8 POSE model as the base model, perform machine learning to adapt to the vertebral key point detection task, optimize the model parameters through the backpropagation algorithm, and thus accurately identify the key feature points of the vertebrae.

[0047] Step S3.5: Apply the model by using the trained feature point recognition network optimized for a specific vertebral body type to the spine segmentation data that has been classified.

[0048] To achieve the above objectives, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the two-step spinal X-ray processing method based on single vertebral body segmentation.

[0049] To achieve the above objectives, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the two-step spinal X-ray processing method based on single vertebral body segmentation.

[0050] Deep learning technology has enabled the rapid development of image segmentation methods. Compared with traditional image segmentation methods, this invention improves both segmentation accuracy and precision, providing clinicians with more effective information. Attached Figure Description

[0051] Figure 1 is a flowchart of a two-step spinal X-ray processing method based on single vertebral body segmentation according to the present invention.

[0052] Figure 2 is a schematic diagram of the YOLOv8 target detection model for vertebral target bounding box detection in a two-step spinal X-ray processing method based on single vertebral segmentation according to the present invention.

[0053] Figure 3 is a schematic diagram of the vertebral contour recognition in the segmentation step of a two-step spinal X-ray processing method based on single vertebral body segmentation according to the present invention.

[0054] Figure 4 shows the actual operation results of feature point recognition in a two-step spinal X-ray processing method based on single vertebral body segmentation according to the present invention. Detailed Implementation

[0055] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0056] In the preferred embodiments of the present invention, those skilled in the art should note that X-ray images and the like involved in the present invention can be considered as prior art.

[0057] Preferred embodiment.

[0058] As shown in Figure 1, this invention discloses a two-step spinal X-ray image processing method based on single vertebral body segmentation, comprising the following steps:

[0059] Step S1: Use the YOLOv8 object detection model to identify the location of each individual spinal vertebra in the X-ray image and generate a single vertebra object bounding box;

[0060] Step S2: Perform detailed segmentation within each detected single vertebral body target box to accurately delineate the vertebral contour;

[0061] Step S3: Perform feature point recognition within the target bounding box of a single vertebra. Use the YOLOv8 POSE model to identify key feature points of the vertebrae, further improving the accuracy of feature point recognition (the YOLOv8 POSE model is a variant of the YOLOv8 model specifically designed for pose estimation. The YOLOv8 model detects the position and category of an object, while the YOLOv8 POSE model further detects the pose of the object or key points of the human body).

[0062] Specifically, step S1 is implemented as follows:

[0063] Step S1.1: Perform image preprocessing. Before inputting the X-ray image into the YOLOv8 target detection model, perform preprocessing, including normalization, rotation, and cropping, to adapt to the model's input requirements.

[0064] Step S1.2: Perform feature extraction. Use the backbone network (CSPDarknet53) of the YOLOv8 target detection model to extract multi-scale features from the X-ray image. Use residual blocks to deepen the network and improve the feature extraction effect. And solve the gradient vanishing and gradient exploding problems by introducing skip connections.

[0065] Step S1.3: Perform feature fusion by using FPN (Feature Pyramid Network) and PANet (Path Aggregation Network) to fuse features at different scales, thereby improving the accuracy and robustness of target detection.

[0066] Step S1.4: Predict the bounding box. Use the YOLO Head module of the YOLOv8 object detection model to predict the bounding box, class confidence, and position offset on the feature map.

[0067] Step S1.5: Perform non-maximum suppression (NMS) to remove redundant boxes and retain only high-confidence boxes, ensuring that each target box uniquely corresponds to a vertebra category.

[0068] More specifically, for step S1.3:

[0069] A top-down feature pyramid structure is constructed using FPN to extract multi-scale semantic information from feature maps at different levels, thereby addressing the problem of insufficient multi-scale information in object detection.

[0070] PANet introduces lateral connections and adaptive feature pooling mechanisms to facilitate information transfer and fusion between feature maps of different resolutions, thereby making better use of the multi-scale features generated by FPN.

[0071] Furthermore, regarding step S1.4:

[0072] Classification Head: The classification head predicts the class label by applying convolutional layers and activation functions (such as Softmax) to the feature map. The convolutional layer converts the information in the feature map into a score for each class, and the class with the highest score is the class label of the object in the bounding box.

[0073] The Softmax activation function transforms the raw scores (logits) into numerical values ​​representing a probability distribution, such that the probability value for each class is between 0 and 1, and the sum of the probabilities of all classes equals 1, as shown in the following formula:

[0074] Where e represents the base of the natural logarithm (Euler's number), n represents the number of categories, and z i It is the original score of the i-th category;

[0075] Regression head: Used to predict the coordinate parameters and confidence score of each target box to determine the position and size of the target box. The coordinate parameters include the center coordinates, width, and height of the target box. The model predicts these coordinate parameters through a regression algorithm to accurately locate the target. The confidence score indicates the model's confidence in whether the target box contains the target and is used to filter the detection results.

[0076] By combining the above classification and regression heads, the target's category, location, and confidence information can be accurately predicted on the feature map.

[0077] Furthermore, step S1.5 is specifically implemented as follows:

[0078] Step S1.5.1: Perform confidence screening. In the target boxes predicted by the model, first sort all target boxes according to their confidence scores and retain the target boxes with the highest confidence scores.

[0079] Step S1.5.2: Calculate the overlap. For the remaining target boxes, calculate the overlap with the target box with the highest confidence using the Intersection over Union (IoU) ratio.

[0080] Step S1.5.3: Perform NMS operation to remove target boxes with IoU higher than the set threshold from the remaining target boxes, ensuring that no redundant target boxes with large overlap with the target box with the highest confidence are retained. Repeat this process until all target boxes have been processed.

[0081] Step S1.5.4: Ensure uniqueness. After the NMS operation is completed, each retained target box should correspond to a unique target. If multiple target boxes highly overlap with the same target, NMS will select the target box with the highest confidence, while other target boxes will be suppressed.

[0082] Preferably, step S2 is specifically implemented as follows:

[0083] Step S2.1: Prepare data by preparing X-ray images of individual spinal vertebrae (3000 cases) that have been classified. Standardize the data into images suitable for model input through operations including rotation and cropping.

[0084] Step S2.2: Perform data preprocessing. Preprocess the extracted target boxes, including scaling, cropping, and normalization, to meet the input requirements of the U-Net network.

[0085] Step S2.3: Construct the U-Net network architecture. This U-Net network takes a 1024×512×1 image as input. Each encoder layer includes two 3×3 convolutional layers, followed by instance normalization, ReLU activation function and 2×2 max pooling. Dropout is applied to each encoder-decoder stage of the network to generate side outputs at 128, 256 and 512 resolutions of the decoder, and then generate the final output at the original resolution.

[0086] Step S2.4: Train the network using the extracted target boxes as training data. During training, optimize the network parameters through backpropagation algorithm so that the network can accurately learn the contour and structure of the vertebrae.

[0087] Step S2.5: Perform segmentation prediction, input the extracted target box into the U-Net network to obtain pixel-level labels of the cone contour;

[0088] Step S2.6: Post-process the segmentation results by applying a region growing algorithm to fill holes and applying Gaussian filtering to remove noise and smooth edges, thereby obtaining more accurate and smoother segmentation results.

[0089] Preferably, step S3 is specifically implemented as follows:

[0090] Step S3.1: Prepare data by preparing X-ray images of individual vertebrae of the spine with feature point markings (1000 cases). Standardize the data into images suitable for model input through operations including rotation and cropping.

[0091] Step S3.2: Perform data augmentation by performing operations such as random rotation, scaling, and flipping on the training data to increase the diversity of the data and help the model better generalize to different vertebral poses and angles;

[0092] Step S3.3 calculates the loss function, using mean squared error (MSE) and Euclidean distance between key points to ensure that the loss function effectively guides the model to learn accurate feature point locations;

[0093] Mean squared error (MSE) is the sum of the squared distances between the target variable and the predicted value;

[0094] Where ρ is the Euclidean distance between point (x2,y2) and point (x1,y1);

[0095] Step S3.4: Perform model training, using the pre-trained YOLOv8 POSE model as the base model, perform machine learning to adapt to the vertebral key point detection task, optimize the model parameters through the backpropagation algorithm, and thus accurately identify the key feature points of the vertebrae.

[0096] Step S3.5: Apply the model by using the trained feature point recognition network optimized for a specific vertebral body type to the spine segmentation data that has been classified.

[0097] Explanation of the algorithm in this invention:

[0098] U-Net: U-Net is a classic convolutional neural network architecture commonly used for medical image segmentation tasks. It features an encoder-decoder structure, which effectively captures contextual information from images. However, U-Net struggles with handling blurred boundaries and segmenting small objects.

[0099] FCN (Fully Convolutional Network): FCN is a type of fully convolutional network that can perform pixel-level classification of an entire image. It can generate dense prediction results, but it faces issues of memory consumption and computational complexity when processing large images.

[0100] DeepLab: DeepLab is a segmentation algorithm based on dilated convolution, which captures broader contextual information by increasing the receptive field of the convolution kernel. However, DeepLab struggles with fine structures and blurred boundaries.

[0101] Mask R-CNN: Mask R-CNN is an object detection and segmentation algorithm based on a Region Proposal Network. It can achieve object detection and segmentation simultaneously, but it faces the problem of slow training and inference speed when dealing with large-scale datasets.

[0102] YOLO (You Only Look Once) is a deep learning-based object detection algorithm that predicts the bounding boxes and categories of multiple objects in an image simultaneously through a single forward pass. Its main characteristics are speed and accuracy. However, because YOLO divides the image into a grid for prediction, it can lead to inaccuracies in detecting small objects.

[0103] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the two-step spinal X-ray processing method based on single vertebral body segmentation.

[0104] The present invention also discloses a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the two-step spinal X-ray processing method based on single vertebral body segmentation.

[0105] It is worth mentioning that the technical features such as X-ray images involved in this patent application should be regarded as prior art. The specific structure, working principle, and possible control methods and spatial arrangement of these technical features can be adopted using conventional choices in the field, and should not be regarded as the inventive point of this patent. This patent will not be further elaborated in detail.

[0106] For those skilled in the art, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A two-step spinal X-ray processing method based on single vertebra segmentation, characterized in that, Comprising the following steps: Step S1: using YOLOv8 target detection model to identify the position of each single spinal vertebra in the X-ray image, and generate single vertebra target frame; Step S2: detailed segmentation is performed within each detected single vertebra target frame to accurately outline the contour of the vertebra; Step S3: single vertebra target frame feature point recognition is performed, using YOLOv8 POSE model to identify the key feature points of the vertebra, further improving the accuracy of feature point recognition.

2. The two-step spinal X-ray processing method based on single vertebra segmentation of claim 1, wherein, Step S1 is implemented as follows: Step S1.1: image preprocessing, the X-ray image is preprocessed before being input into the YOLOv8 target detection model, including normalization, rotation and cutting operations to adapt to the input requirements of the model; Step S1.2: feature extraction, using the backbone network of YOLOv8 target detection model to extract multi-scale features from X-ray image, using residual block to deepen the network and improve the effect of feature extraction, and solving the problems of gradient vanishing and gradient explosion by introducing skip connection; Step S1.3: feature fusion, using FPN and PANet to fuse features of different scales to improve the accuracy and robustness of target detection; Step S1.4: target frame prediction, using YOLO Head module of YOLOv8 target detection model to predict target frame, class confidence and position offset on feature map; Step S1.5: non-maximum suppression, removing redundant frames by non-maximum suppression, only keeping high confidence frames to ensure that each target frame corresponds to a unique vertebra class.

3. The method of claim 2, wherein the method is a two-step method of processing a spine X-ray image based on single vertebra segmentation, the method comprising: a first step of segmenting a single vertebra from the spine X-ray image; and a second step of processing the segmented single vertebra. For step S1.3: FPN is used to build a top-down feature pyramid structure to extract multi-scale semantic information from feature maps of different levels to solve the problem of insufficient multi-scale information in target detection; PANet is used to introduce horizontal connection and adaptive feature pooling mechanism to transfer and fuse information between feature maps of different resolutions, thus utilizing the multi-scale features generated by FPN.

4. The method of claim 3, wherein the method further comprises: For step S1.4: Classification head: the classification label is predicted by applying convolution layer and activation function on feature map, the convolution layer converts the information in feature map into the score of each class, and the class with the highest score is the class label of the object in the target frame; Regression head: used to predict the coordinate parameters and confidence score of each target frame to determine the position and size of the target frame, the coordinate parameters include the center coordinates, width and height of the target frame, the model predicts the coordinate parameters by regression algorithm to accurately locate the position of the target; the confidence score represents the confidence degree of the model on whether the target is contained in the target frame, which is used to filter the detection results; The softmax activation function converts raw scores into values that represent a probability distribution, such that the probability value for each class is between 0 and 1, and the sum of the probabilities for all classes equals 1, as follows: where e denotes the base of the natural logarithm, n denotes the number of classes, z i is the raw score of the i-th class; The combination of the above classification head and regression head accurately predicts the class, position and confidence information of the target on the feature map. Step S1.5 is implemented as follows:

5. The method of claim 4, wherein the method is a two-step vertebral X-ray processing method based on single vertebral segmentation, characterized in that, Step S1.5.1: confidence screening, in the target frame predicted by the model, first sort all target frames according to the confidence score, and keep the target frame with the highest confidence; ​ Step S1.5.2: Perform overlap calculation, for the remaining target boxes, calculate the overlap with the highest confidence target box by intersection over union calculation; Step S1.5.3: Perform NMS operation, remove target boxes with IoU higher than a set threshold from the remaining target boxes, ensure that redundant target boxes with large overlap with the highest confidence target box are not retained, repeat this process until all target boxes are processed; Step S1.5.4: Perform uniqueness guarantee, after the NMS operation is completed, each retained target box corresponds to a unique target, if multiple target boxes overlap with the same target to a high degree, NMS will select the target box with the highest confidence, and other target boxes will be suppressed.

6. The two-step spinal X-ray processing method based on single vertebra segmentation of claim 5, wherein, Step S2 is implemented as follows: Step S2.1: Perform data preparation, prepare single spine cone X-ray image data for classification, standardize the data to images suitable for model input through operations including rotation, cropping; Step S2.2: Perform data preprocessing, perform data preprocessing on the extracted target box, including scaling, cropping and normalization operations to meet the input requirements of the U-Net network; Step S2.3: Build U-Net network architecture, this U-Net network takes 1024x512x1 images as input, each encoder layer includes two 3x3 convolution layers followed by instance normalization, ReLU activation function and 2x2 max pooling, Dropout is applied to each encoder-decoder stage of the network, side outputs are generated at 128, 256 and 512 resolutions of the decoder, and then a final output is generated at the original resolution; Step S2.4: Train the network, use the extracted target box as training data to train the U-Net network, during training, optimize the network parameters through the backpropagation algorithm, so that the network can accurately learn the contour and structure of the vertebrae; Step S2.5: Perform segmentation prediction, input the extracted target box into the U-Net network to obtain pixel-level labeling of the vertebra contour; Step S2.6: Post-process the segmentation result, apply region growing algorithm to fill in the holes, and apply Gaussian filter to remove noise and smooth edges to obtain more accurate and smooth segmentation results.

7. The method of claim 6, wherein the method is a two-step method of processing a spine X-ray image based on single vertebra segmentation, the method comprising: a first step of segmenting a single vertebra from the spine X-ray image; and a second step of processing the single vertebra to determine a spinal curve. Step S3 is implemented as follows: Step S3.1: Perform data preparation, prepare single spine cone X-ray image data with feature point labeling, standardize the data to images suitable for model input through operations including rotation, cropping; Step S3.2: Perform data augmentation, perform operations including random rotation, scaling and flipping on the training data to increase the diversity of the data, which helps the model to better generalize to different vertebrae poses and angles; Step S3.3 performs loss function calculation, which ensures that the loss function effectively guides the model to learn accurate feature point positions through mean square error and Euclidean distance of key points; The mean square error is the sum of the square of the distance between the target variable and the predicted value; Wherein, p is the Euclidean distance between point (x2, y2) and point (x1, y1); Step S3.4: Model training, use the pre-trained YOLOv8 POSE model as the base model, perform machine learning to adapt to the vertebra key point detection task, optimize the model parameters through the backpropagation algorithm, so as to accurately identify the key feature points of the vertebrae; Step S3.5: Model application, applying the trained feature point recognition network optimized for the specific vertebra type to the segmented spinal column data of which the classification is completed.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the two-step spinal X-ray processing method based on single vertebra segmentation of any one of claims 1 to 7 when executing the program. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the steps of the two-step spinal X-ray processing method based on single vertebra segmentation of any one of claims 1 to 7 when executed by the processor.

Images