Pelvic bone key point positioning method and system and storage medium

Abstract

The invention discloses a pelvis key point positioning method and system and a storage medium. The pelvis key point positioning method comprises the following steps: S1, generating coordinate points of an original image and a distorted image of a skeleton image; s2, key point mapping calculation after bone distortion; s3, constructing four quadrants, and realizing image distortion augmentation through equal ratio transformation; and S4, training the key point positioning model through deep learning to realize clinical pelvic bone key point positioning. According to the method, a more representative data set is generated by introducing an imitated distortion data augmentation technology, and automatic and high-accuracy pelvic key point positioning is realized; according to the method, the positioning precision of the key points of the pelvic bone can be reduced from 1.3 mm to 1.1 mm, and the positioning precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a pelvic key point positioning method, system and storage medium. Background Art

[0002] In the field of artificial intelligence and computer vision, pelvic key point localization is an important technology in medical image analysis. Localization of deformed pelvic key points is an important part of diagnosis of developmental hip dislocation and pelvic trauma. Current algorithms mostly rely on traditional image processing methods, such as template matching and random forests, which are difficult to deal with complex backgrounds and different postures. There are currently some pelvic key point localization algorithms based on deep learning. These algorithms achieve direct prediction from input images to key point coordinates through end-to-end training. However, these algorithms rely on massive amounts of data for training, which places stringent requirements on data collection and annotation. In particular, the lack of deformed and pathological samples limits the model's ability to locate the key points of the deformed pelvis.

[0003] Current data augmentation techniques mainly expand training data sets by rotation, scaling, translation, mirroring, and color transformation to increase data diversity and improve the generalization ability of the model. However, these traditional data augmentation methods may not be able to effectively simulate various complex scenes in real clinical scenarios when faced with complex medical imaging data, especially for the common bone development distortions in pelvic medical imaging. Therefore, there is an urgent need for an augmentation method that can better simulate bone distortion to improve the robustness and accuracy of the pelvic key point positioning algorithm. Summary of the invention

[0004] Based on the technical problems existing in the background technology, the present invention proposes a pelvic key point positioning method, system and storage medium, and generates a more representative data set by introducing simulated distortion data augmentation technology to achieve automatic and high-accuracy pelvic key point positioning.

[0005] The pelvic key point positioning method proposed by the present invention has the following steps: S1: Generation of coordinate points of original and distorted images of bone images; S2: key point mapping calculation after skeleton distortion; S3: Construct four quadrants and realize image distortion and augmentation through geometric transformation; S4: Through deep learning, the key point positioning model is trained to achieve clinical pelvic key point positioning.

[0006] Preferably, the coordinates of the sampling points of the original image before distortion in S1 are In the image area Randomly generated inside. Coordinates of sampling points after distortion exist is centered and has a width and height of The image size is randomly generated in a rectangle of , Gao Wei .

[0007] Preferably, the calculation method of the key point mapping in S2 is: the key point coordinates in the image before distortion are , the coordinates after distortion are , define a distortion mapping function ; if ,but ;otherwise .

[0008] if , ;otherwise ; Thus, the complete distortion mapping function is obtained .

[0009] Preferably, the grayscale of each pixel of the distorted image defined in S3 is , by finding the pixel points of the distorted image corresponding to the pixel points of the original image Gray value , thus achieving image distortion.

[0010] Preferably, the steps of implementing image distortion are as follows: define an inverse mapping function: ; if ,but ;otherwise .

[0011] if , ;otherwise ; Thus, the complete distortion mapping function is obtained , thereby assigning the grayscale value of the original image coordinate point obtained by inverse mapping to the corresponding coordinate point of the distorted image, and the data generation of the distorted image can be completed.

[0012] Preferably, the deep learning method in S4 is: converting the key point positioning task into a heat map distribution estimation task; designing a U-type network, extracting features and generating a key point heat map through deconvolution; and realizing pelvic key point positioning by using the highest response point of the heat map as the key point position.

[0013] The pelvic key point positioning system proposed by the present invention comprises: An image processing unit, used for generating coordinate points of original image and distorted image of bone images; A key point mapping calculation unit, used for key point mapping calculation after bone distortion; Image distortion augmentation unit, used to construct four quadrants and realize image distortion augmentation through geometric transformation; The model training unit trains the key point positioning model through deep learning to achieve clinical pelvic key point positioning.

[0014] The computer-readable medium provided by the present invention stores a computer program, and when the computer program is executed by a processor, the above-mentioned pelvic key point positioning method is implemented.

[0015] Beneficial technical effects of the present invention: The present invention generates a more representative data set through the distortion-simulated data augmentation technology, which improves the positioning accuracy; enhances the robustness of the model under different conditions, and the positioning accuracy fluctuates less under different distortion conditions; reduces the reliance on a large amount of manually annotated data, and alleviates the problems of high cost and limited data volume of medical image data annotation; because it can handle various types of distortion and complex scenes, it is not only suitable for pelvic key point positioning, but can also be extended to other medical image analysis tasks that are prone to distortion. It has broad practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the simulated distortion data augmentation process proposed by the present invention; Figure 2 This is the augmented effect diagram of the simulated distortion data proposed by the present invention; Figure 3 This is a structural block diagram of the network deep learning method proposed in the present invention. DETAILED DESCRIPTION

[0017] The present invention will be further explained below in conjunction with specific embodiments.

[0018] Example 1 The pelvic key point positioning method proposed by the present invention has the following steps: (1) Generation of coordinate points of original and distorted images of bone images like Figure 1 In part (a) and part (b), let the image size be width and height respectively. , the sampling point coordinates of the original image before distortion In the image area Randomly generated inside. Coordinates of sampling points after distortion exist is centered and has a width and height of Randomly generated within the rectangle.

[0019] (2) Key point mapping calculation after skeleton distortion For the key points in the image before distortion ,in are their horizontal and vertical coordinates in the image respectively. In this step, the distorted coordinates are calculated. First, we define a distortion mapping function .

[0020] if ,but ;otherwise .

[0021] Similarly, if , ;otherwise .

[0022] Then the complete distortion mapping function can be obtained from the above formula .

[0023] (3) Construct four quadrants and realize image distortion and augmentation through geometric transformation like Figure 1 In part (c) and part (d), for the distorted image, the grayscale of each pixel is defined as , we can find the pixel points of the distorted image corresponding to the pixel points of the original image Gray value , thus achieving image distortion. Therefore we need to find the inverse mapping function: .

[0024] if ,but ;otherwise .

[0025] Similarly, if , ;otherwise .

[0026] The complete distortion mapping function can be obtained from the above formula: . Then, the grayscale value of the original image coordinate point obtained by inverse mapping is assigned to the coordinate point corresponding to the distorted image, and the data generation of the distorted image can be completed. The distortion effect diagram is shown in Figure 2 shown.

[0027] (4) Through deep learning, train the key point positioning model to achieve clinical pelvic key point positioning The present invention uses a deep learning method to locate pelvic key points in X-ray images. First, the key point location task is converted into a heat map distribution estimation task. Figure 3As shown in the figure, a U-shaped network is designed to extract features and generate key point heat maps through deconvolution. Six independent heat maps are generated for six key points in each image, and finally the highest response point of the heat map is used as the key point position, thereby realizing the positioning of the pelvic key points.

[0028] It has been determined that the present invention can reduce the positioning accuracy of key points of the pelvic bones from an average error of 1.3 mm to an average error of 1.1 mm, thereby improving the positioning accuracy.

[0029] Example 2 The pelvic key point positioning system proposed by the present invention comprises: An image processing unit, used for generating coordinate points of original image and distorted image of bone images; A key point mapping calculation unit, used for key point mapping calculation after bone distortion; Image distortion augmentation unit, used to construct four quadrants and realize image distortion augmentation through geometric transformation; The model training unit trains the key point positioning model through deep learning to achieve clinical pelvic key point positioning.

[0030] Example 3 The computer-readable medium provided by the present invention stores a computer program, and when the computer program is executed by a processor, the method for locating key points of the pelvis in Example 1 is implemented.

Claims

1. A method for locating key points of a pelvis, characterized in that: The steps are as follows: S1: Generation of coordinate points of original and distorted images of bone images; S2: key point mapping calculation after skeleton distortion; S3: Construct four quadrants and realize image distortion and augmentation through geometric transformation; S4: Through deep learning, the key point positioning model is trained to achieve clinical pelvic key point positioning.

2. The pelvic key point positioning method according to claim 1, characterized in that: Coordinates of sampling points of the original image before distortion in S1 In the image area Randomly generated inside. Coordinates of sampling points after distortion exist is centered and has a width and height of The image size is randomly generated in a rectangle of , Gao Wei .

3. The pelvic key point positioning method according to claim 2, characterized in that: The calculation method of key point mapping in S2 is: the coordinates of the key points in the image before distortion are , the coordinates after distortion are , define a distortion mapping function ; if ,but ;otherwise ; if ;otherwise ; Thus, the complete distortion mapping function is obtained .

4. The pelvic key point positioning method according to claim 1, characterized in that: The grayscale of each pixel of the distorted image is defined in S3 as , by finding the pixel points of the distorted image corresponding to the pixel points of the original image Gray value , thus achieving image distortion.

5. The pelvic key point positioning method according to claim 4, characterized in that: The steps to achieve image distortion are as follows: Define the inverse mapping function: ; if ,but ;otherwise ; if , ;otherwise ; Thus, the complete distortion mapping function is obtained , thereby assigning the grayscale value of the original image coordinate point obtained by inverse mapping to the corresponding coordinate point of the distorted image, and the data generation of the distorted image can be completed.

6. The pelvic key point positioning method according to claim 1, characterized in that: The deep learning method in S4 is: converting the key point positioning task into a heat map distribution estimation task; designing a U-shaped network to extract features and generate a key point heat map through deconvolution; using the highest response point of the heat map as the key point position to achieve pelvic key point positioning.

7. A pelvic key point positioning system, characterized in that: include: An image processing unit, used for generating coordinate points of original image and distorted image of bone image; A key point mapping calculation unit, used for key point mapping calculation after bone distortion; Image distortion augmentation unit, used to construct four quadrants and realize image distortion augmentation through geometric transformation; The model training unit trains the key point positioning model through deep learning to achieve clinical pelvic key point positioning.

8. A computer readable medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for locating key points of the pelvis as described in any one of claims 1 to 6 is implemented.

Images