Local pelvic skeleton image rigid registration method, device, equipment and medium

Abstract

The invention provides a rigid registration method and device for a local pelvic bone image, equipment and a medium. The rigid registration method comprises the following steps: extracting a 3D pelvic bone image from a 3D medical image; aiming at the 3D pelvic skeleton image, generating a 2D integral pelvic skeleton image corresponding to each group of alternative parameter combination; performing feature point matching on the 2D medical image and the 2D overall pelvic skeleton image, and performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic skeleton image based on a feature point matching result to determine a target 2D local pelvic skeleton image matched with the 2D medical image; and performing rigid registration on the 3D medical image based on the target 2D local pelvic skeleton image and the corresponding alternative parameter combination, so that the 3D medical image after rigid registration is aligned with the 2D medical image. According to the method, the rigid registration problem from local 2D X-ray imaging to global 3D CT data can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular, to a method, device, equipment and medium for rigid registration of local pelvic bone images. Background Art

[0002] The pelvis is a very important bone structure in the human body, which is composed of the sacrum, coccyx and two hip bones on the left and right. As a relatively large bone group in the human body, in pelvic surgery, there are often problems of incomplete imaging. Limited by the incomplete imaging, the accuracy of common registration methods becomes worse in the clinical use of the pelvis.

[0003] The feature extraction and segmentation methods used in traditional methods require manual intervention, so such methods are difficult to automate. Registration is based on gray information, and the transformation parameters are determined by optimizing the gray similarity of the images, but the effect on images with deformation or obvious gray changes is poor.

[0004] Current deep learning-based methods are difficult to train and learn local bones. Limited by the imaging machines of medical images, the X-ray images obtained by intraoperative C-arm imaging can often only capture local information of pelvic bones and are difficult to capture comprehensively. Supervised networks often learn and train based on global information. Since the local pelvis has greater uncertainty compared to the imaging of the complete pelvic bones, it is impossible to fully construct a data set, resulting in difficulty in implementing common deep learning supervised methods for the local pelvis. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for rigid registration of local pelvic bone images, which can effectively handle the rigid registration problem of local 2D X-ray imaging to global 3D CT data.

[0006] In a first aspect, the present invention provides a method for rigid registration of local pelvic bone images, including:

[0007] Obtain a 3D medical image to be registered, the content shown in the 3D medical image includes overall pelvic information, and extract a 3D pelvic bone image from the 3D medical image;

[0008] Determine multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate 2D overall pelvic bone images corresponding to each group of alternative parameter combinations for the 3D pelvic bone image; wherein, a group of alternative parameter combinations includes alternative parameter values corresponding to multiple rigid registration parameter items;

[0009] Perform feature point matching on the 2D medical image and the 2D overall pelvic bone image, and perform dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image matched by the 2D medical image;

[0010] Based on the target 2D local pelvic bone image and its corresponding alternative parameter combination, perform rigid registration on the 3D medical image to align the rigidly registered 3D medical image with the 2D medical image.

[0011] In one implementation, performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image matched by the 2D medical image includes:

[0012] Based on the feature point matching result, dynamically and adaptively adjust to determine the target size and position information of the cropping window corresponding to each 2D overall pelvic bone image;

[0013] Crop the corresponding 2D overall pelvic bone image according to the target size and position information of the cropping window to obtain the cropped 2D local pelvic bone image of each 2D overall pelvic bone image;

[0014] Determine the target 2D local pelvic bone image matched by the 2D medical image according to the similarity between the 2D medical image and each 2D local pelvic bone image.

[0015] In one implementation, based on the feature point matching result, dynamically and adaptively adjust to determine the target size and position information of the cropping window corresponding to each of the 2D overall pelvic bone images, including:

[0016] Perform the following operations on any one of the 2D overall pelvic bone images:

[0017] Based on the feature point matching result, determine the initial size and position information of the cropping window corresponding to this 2D overall pelvic bone image, and set a buffer area based on the initial size and position information for this 2D overall pelvic bone image;

[0018] Set a set of adaptive windows within the buffer area;

[0019] For each adaptive window in the set of adaptive windows, extract a set of feature point pairs from the 2D medical image and this 2D overall pelvic bone image based on the feature point matching result to determine the confidence level corresponding to this adaptive window using the set of feature point pairs;

[0020] Determine the target size and position information of the cropping window according to the confidence level corresponding to each adaptive window.

[0021] In one implementation, determining the confidence corresponding to the adaptive window using the set of feature point pairs includes:

[0022] Using the set of feature point pairs to determine the local similarity of the 2D medical image and the 2D overall pelvic bone image with respect to the adaptive window; assigning a confidence to the adaptive window based on the local similarity; wherein the confidence is positively correlated with the local similarity;

[0023] Determining the target size and position information of the cropping window according to the confidence corresponding to each adaptive window, including:

[0024] According to the confidence corresponding to each adaptive window, screening out the target adaptive window, and using the size and position information of the target adaptive window to adjust the initial size and position information of the cropping window to obtain the target size and position information of the cropping window.

[0025] In one implementation, dynamically and adaptively determining the target size and position information of the cropping window corresponding to each 2D overall pelvic bone image further includes:

[0026] Performing the following operations on any 2D overall pelvic bone image:

[0027] Through a pre-trained cropping and positioning model, based on the feature point matching result, respectively extracting the complete image feature of the 2D medical image and the local image feature of the 2D overall pelvic bone image, determining the similarity between the complete image feature and the local image feature based on the cross-attention mechanism, and determining the target size and position information of the cropping window of the 2D overall pelvic bone image based on the similarity.

[0028] In one implementation, determining the HU value corresponding to each voxel in the 3D medical image;

[0029] Based on multiple window transformation parameters and the HU value corresponding to each voxel, dividing the 3D medical image into a background and non-bone part, and a main pelvic bone structure part;

[0030] Using the background and non-bone part to remove the background and non-bone tissues in the 3D medical image, and extracting the 3D pelvic bone image in combination with the main pelvic bone structure part.

[0031] In one implementation, performing feature point matching on the 2D medical image and the 2D overall pelvic bone image includes:

[0032] Through a feature point matching model, based on the self-attention and cross-attention mechanisms, performing feature point extraction and feature point matching on the 2D medical image and the 2D overall pelvic bone image to obtain a feature point matching result.

[0033] In a second aspect, the present invention further provides a local pelvic bone image rigid registration device, including:

[0034] A 3D image extraction module, configured to obtain a 3D medical image to be registered, where the content displayed in the 3D medical image includes overall pelvic information, and extract a 3D pelvic bone image from the 3D medical image;

[0035] A 2D overall image generation module, configured to determine multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate a 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image; wherein, a group of alternative parameter combinations includes alternative parameter values corresponding to multiple rigid registration parameter items;

[0036] A template matching module, configured to perform feature point matching on the 2D medical image and the 2D overall pelvic bone image, and perform dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image matched by the 2D medical image;

[0037] A rigid registration module, configured to perform rigid registration on the 3D medical image based on the target 2D local pelvic bone image and its corresponding alternative parameter combination, so that the rigidly registered 3D medical image is aligned with the 2D medical image.

[0038] In a third aspect, the present invention further provides an electronic device, including a processor and a memory, where the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of the first aspect.

[0039] In a fourth aspect, the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores computer executable instructions, and when the computer executable instructions are called and executed by the processor, the computer executable instructions cause the processor to implement the method according to any one of the first aspect.

[0040] A rigid registration method, device, equipment and medium for local pelvic bone images provided by the present invention first obtains a 3D medical image to be registered, the content shown in the 3D medical image includes overall pelvic information, and extracts a 3D pelvic bone image from the 3D medical image; then determines multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, generates a 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image, and a group of alternative parameter combinations includes alternative parameter values corresponding to multiple rigid registration parameter items; then performs feature point matching on the 2D medical image and the 2D overall pelvic bone image, and performs dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image matched by the 2D medical image; finally, based on the target 2D local pelvic bone image and its corresponding alternative parameter combination, performs rigid registration on the 3D medical image to align the rigidly registered 3D medical image with the 2D medical image. The above method is applicable to scenarios where it is difficult to obtain a complete 3D medical image. After generating a 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image, the method sequentially performs feature point matching and dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image to obtain the target 2D local pelvic bone image matched by the 2D medical image. On this basis, combined with the alternative parameter combination corresponding to the target 2D local pelvic bone image, rigid registration of the 3D medical image can be achieved. The present invention can effectively handle the rigid registration problem from local 2D X-ray imaging to global 3D CT data.

[0041] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims and drawings.

[0042] To make the above objectives, features and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, provides detailed descriptions as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 Schematic diagram of a rigid registration framework based on deep learning provided by an embodiment of the present invention;

[0045] Figure 2 A schematic flowchart of a method for rigid registration of local pelvic bone images provided by an embodiment of the present invention;

[0046] Figure 3 A technical framework diagram of a method for rigid registration of local pelvic bone images provided by an embodiment of the present invention;

[0047] Figure 4 A schematic diagram of the DRR imaging principle provided by an embodiment of the present invention;

[0048] Figure 5 A schematic diagram of the AspanFormer feature point matching result provided by an embodiment of the present invention;

[0049] Figure 6 A schematic diagram of the projection angle change provided by an embodiment of the present invention;

[0050] Figure 7 A schematic diagram of completing template matching based on the feature point matching relationship provided by an embodiment of the present invention;

[0051] Figure 8 A schematic structural diagram of a device for rigid registration of local pelvic bone images provided by an embodiment of the present invention;

[0052] Figure 9 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Image-Guided Surgery (IGS) is a method that uses medical imaging technology to guide surgeons in real-time during surgery. This method precisely compares preoperative imaging data (such as CT (Computed Tomography), MRI (Nuclear Magnetic Resonance Imaging), ultrasound, etc.) with the patient's actual anatomical structure, thereby providing surgeons with higher operating precision and lower risks. Among them, the three-dimensional data obtained through CT and other methods before surgery has richer information and higher precision compared to two-dimensional data. However, it is often limited by long acquisition times and difficult acquisitions, and cannot be obtained in real-time in the surgical environment. The two-dimensional XR fluoroscopy data obtained using X-ray projection is obtained quickly, and the technology and equipment are more complete. Obtaining XR images during surgery does not affect the progress of the surgery. For image-guided surgery, it is usually necessary to obtain three-dimensional information of the surgical site through CT or magnetic resonance scanning before surgery, which is used as a positioning standard during surgery and also provides doctors with information about the lesion before surgery to evaluate the patient's condition and plan the surgical plan. During surgery, two-dimensional images obtained in real-time through X-rays and other methods promptly reflect the current surgical situation and track the current position, so that doctors can understand the current specific information.

[0055] Registration is the process of aligning different digital images of the same lesion or tissue. Medical image registration technology is to find a spatial transformation that can make the corresponding points of two images from different acquisition times and different devices match in spatial relationship and structure. It is a key technology in image-guided surgery. If the two-dimensional X-ray images obtained in real-time during surgery and the high-precision three-dimensional CT images obtained before surgery can be aligned through registration technology, the advantages of both can be combined to help doctors obtain richer data, thereby achieving higher-precision navigation surgery.

[0056] The accuracy and timeliness of the matching between two-dimensional images and three-dimensional model information are crucial for surgery. The registration between digital information of different dimensions often requires converting the two types of image information into a unified dimension first. In the research and application of medical image registration, there are already relatively mature methods to project three-dimensional volume information into two-dimensional image information, and projecting three-dimensional information into two-dimensional images is a method with relatively high accuracy. The generation of Digitally Reconstructed Radiograph (DRR) is an important procedure in several medical imaging applications. This method can project and convert three-dimensional image information into two-dimensional images, thus completing the dimensionality reduction of three-dimensional digital information. By using digital reconstruction radiology technology, the registration problem between 2D images and 3D image information is converted into the registration problem between 2D images and 2D images.

[0057] Traditional image registration techniques can currently be mainly divided into two categories: registration based on features and registration based on grayscale. Among them, the method of registration based on feature information extracts a small amount of feature information from the images to be registered, and performs registration through the extracted feature information. Commonly used features include anatomical features of the human body and externally implanted marker points for marking. Commonly used anatomical features include anatomical landmark points, curve features, and surface features, etc. Registration based on grayscale completes registration by comparing similarity through grayscale value information.

[0058] With the maturity of neural network-related technologies, methods using deep learning have gradually emerged to complete registration. ShunMiao et al. first used the regression method of convolutional neural network to achieve the registration of 2D / 3D medical images, and used the neural network to regress the transformation parameters of the images. Existing rigid registration methods based on deep learning mostly use this structure as the basis, and focus on optimizing the network model to improve efficiency, specifically as Figure 1 shown in a schematic diagram of a rigid registration framework based on deep learning.

[0059] In this registration framework, first, the CT data is projected through DRR projection to obtain a two-dimensional DRR projection image and the corresponding deformation parameter ground-truth as labels. The two-dimensional image data and the corresponding ground-truth labels are input into the network to complete supervised network learning, thereby obtaining a registration network. The real X-ray image is input into the learned registration model, thereby predicting the spatial transformation parameters and completing the registration.

[0060] However, the feature extraction and segmentation methods used in traditional methods require manual intervention, so such methods are difficult to automate. Registration based on grayscale information determines the transformation parameters by optimizing the grayscale similarity of the images, but it has poor effects on images with obvious deformation or significant grayscale changes. Current methods based on deep learning are difficult to train and learn local bones. Based on this, the embodiments of the present invention provide a method, device, equipment, and medium for rigid registration of local pelvic bone images, which can effectively handle the rigid registration problem of local 2D X-ray imaging to global 3D CT data.

[0061] For the convenience of understanding this embodiment, first, a method for rigid registration of local pelvic bone images disclosed in the embodiments of the present invention will be introduced in detail. Refer to Figure 2 the schematic flowchart of a method for rigid registration of local pelvic bone images shown, and this method mainly includes the following steps S202 to step S208:

[0062] Step S202: Obtain the 3D medical image to be registered. The content shown in the 3D medical image includes the overall pelvic information, and extract the 3D pelvic bone image from the 3D medical image.

[0063] Among them, the 3D medical image can be a CT image. In one example, the HU (Hounsfiled Unit) value corresponding to each voxel in the 3D medical image can be determined, and the background, non-bone part, and main pelvic bone structure part in the 3D medical image can be identified by combining window transformation parameters, and then the 3D pelvic bone image can be extracted according to the identification result.

[0064] Step S204: Determine multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate 2D overall pelvic bone images corresponding to each group of alternative parameter combinations for the 3D pelvic bone image.

[0065] Among them, a group of alternative parameter combinations includes alternative parameter values corresponding to multiple rigid registration parameter items. For example, a group of alternative parameter combinations can include the alternative parameter value corresponding to the rotation parameter item around the X-axis (abbreviated as the rotation angle around the X-axis) and the alternative parameter value corresponding to the rotation parameter item around the Y-axis (abbreviated as the rotation angle around the Y-axis). The 2D medical image can be an X-ray image, and the 2D overall pelvic bone image can be a DRR image. In one example, after determining multiple groups of alternative parameter combinations, multiple 2D overall pelvic bone images based on the rotation of the X-axis and Y-axis can be generated for the 3D pelvic bone image.

[0066] Step S206: Perform feature point matching on the 2D medical image and the 2D overall pelvic bone image, and perform dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image that matches the 2D medical image.

[0067] Among them, dynamic adaptive template matching can be understood as a process of dynamically and adaptively adjusting the cropping size, and after cropping the 2D overall pelvic bone image according to the adjusted cropping size to obtain the corresponding 2D local pelvic bone image, screening out the target 2D local pelvic bone image that matches the 2D medical image based on template matching. In one example, the ASpanFormer feature point matching network can be used to perform feature point matching on the 2D medical image and the 2D overall pelvic bone image; then for each 2D overall pelvic bone image, the size and position information of the cropping window of the 2D overall pelvic bone image are dynamically and adaptively adjusted based on a sliding window or a neural network model, and then the 2D overall pelvic bone image is cropped according to the adjusted cropping window to obtain the corresponding 2D local pelvic bone image; finally, the 2D local pelvic bone image with the highest similarity to the 2D medical image is used as the target 2D local pelvic bone image for matching.

[0068] Step S208: Based on the target 2D local pelvic bone image and its corresponding alternative parameter combination, perform rigid registration on the 3D medical image so that the rigidly registered 3D medical image is aligned with the 2D medical image.

[0069] In one example, a complete rigid registration parameter combination can be solved according to the 2D medical image, the target 2D local pelvic bone image, and its corresponding alternative parameter combination. The rigid registration parameter combination includes the rotation angle around the X-axis, the rotation angle around the Y-axis, the rotation angle around the Z-axis, the translation amount along the X-axis, the translation amount along the Y-axis, and the translation amount along the Z-axis. Rigidly registering the 3D medical image according to the complete parameter combination can align the rigidly registered 3D medical image with the 2D medical image.

[0070] The rigid registration method for local pelvic bone images provided by the embodiments of the present invention is directed to scenarios where it is difficult to obtain a complete 3D medical image. After generating the 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image, perform feature point matching and dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image in sequence to obtain the target 2D local pelvic bone image that matches the 2D medical image. On this basis, combined with the alternative parameter combination corresponding to the target 2D local pelvic bone image, rigid registration of the 3D medical image can be achieved. The present invention can effectively handle the rigid registration problem from local 2D X-ray imaging to global 3D CT data.

[0071] For ease of understanding, the embodiments of the present invention provide a Figure 3 technical framework diagram of a rigid registration method for local pelvic bone images as shown. The registration framework is mainly divided into the following three major parts: First, in the first step of image preprocessing, it is necessary to project the 3D CT image into a 2D overall DRR image through the DRR projection technology according to the given alternative parameter combination. Second, perform feature point matching between the 2D overall DRR image and the 2D X-ray image to obtain the corresponding point pair relationship and the position information of the corresponding points. Third, calculate the relationship between the two images according to the specific information of the feature point matching, dynamically and adaptively crop the 2D overall DRR image, and at the same time use the cropped 2D local DRR image with the highest similarity to the 2D X-ray image as the target 2D local DRR image to solve the rigid registration parameter combination, and then perform rigid registration on the 3D CT image.

[0072] For ease of understanding, the embodiments of the present invention provide a specific implementation manner of a rigid registration method for local pelvic bone images.

[0073] For the foregoing step S202, an embodiment of the present invention provides a specific implementation manner of obtaining a 3D medical image to be registered, where the content shown in the 3D medical image at least includes local pelvic bones, and extracting a 3D pelvic bone image from the 3D medical image. The 3D medical image is a CT image.

[0074] It includes:

[0075] In the preprocessing part, mainly project the preoperatively obtained 3D CT image into the required 2D overall DRR image (i.e., 2D overall pelvic bone image) according to the requirements. See Figure 4 A schematic diagram of the DRR imaging principle shown. The DRR program simulates the imaging of X-rays. By using the method of simulating the emission of X-rays from a point light source, passing through the three-dimensional volume data, and projecting it onto a plane to generate a two-dimensional image, the two-dimensional planarization of the three-dimensional volume data is realized. In this process, different two-dimensional projection results can be determined by setting the rigid registration parameter combinations r x , r y , r z , t x , t y , t z .

[0076] For the DRR projection technology, the screening of bone information can be completed by performing window transformation on the CT image, so as to obtain a DRR image with cleaner data. During the process of taking the CT image, each voxel on the ray path calculates the attenuation and scattering of X-rays according to the HU value of the voxel and other physical properties. In medical imaging, the HU unit is used to distinguish different types of tissues according to their density, such as bones, muscles, and fats. The window transformation technology is a technology that can process CT data according to the HU value. It completes the screening of CT data by setting an appropriate HU value range, so as to focus the information on the bones and reduce the impurity information. Using the CT data screened by the window technology for DRR projection can obtain a cleaner DRR projection image.

[0077] In specific implementation, the following steps A1 to A3 can be referred to:

[0078] Step A1, determine the HU value corresponding to each voxel in the 3D medical image.

[0079] Step A2, based on multiple window transformation parameters and the HU value corresponding to each voxel, divide the 3D medical image into the background and non-bone part and the main structure part of the pelvic bones.

[0080] Step A3, use the background and non-bone part to remove the background and non-bone tissues in the 3D medical image, and combine with the main structure part of the pelvic bones to extract the 3D pelvic bone image.

[0081] Exemplarily, the HU values (Hounsfield units) of CT images can be divided into the following three ranges to better extract pelvic bones:

[0082] (1) Low HU value range: -1000 HU to -500 HU (or close to this range).

[0083] Characteristic: This range usually corresponds to low-density tissues such as air and fat. In pelvic CT images, although the pelvic bones themselves do not fall into this range, the surrounding soft tissues (such as muscles, fat) and the air in the pelvic cavity may be within this range. By identifying this range, non-bone tissues can be distinguished, providing background information for subsequent bone extraction.

[0084] (2) Medium HU value range: 200 HU to 500 HU (or adjusted according to specific circumstances).

[0085] Characteristic: This range usually corresponds to the transition area from soft tissues to harder tissues, including muscles, some cartilage, and the thinner or looser parts of the pelvic bones. Although the main parts of the pelvic bones may not be within this range, some edges or transition areas may have such HU values. By identifying this range, the boundaries of bone extraction can be further refined.

[0086] (3) High HU value range: 500 HU to +1000 HU (or higher, according to the density change of pelvic bones).

[0087] Characteristic: This range usually corresponds to high-density tissues such as bones and calcified tissues. The main parts of the pelvic bones (such as the hip bones, ilium, pubis, and ischium) usually have such HU values. By identifying this range, the main structures of the pelvic bones can be accurately extracted.

[0088] In practical applications, window transformation techniques or multi-window transformation techniques can be used, combined with the above HU value ranges, to extract pelvic bones. First, background and non-bone tissues are removed according to the low HU value range; then, the boundaries and main structures of the bones are determined according to the medium and high HU value ranges; finally, the extraction results are optimized through morphological operations (such as dilation, erosion, smoothing, etc.) to obtain a more accurate 3D pelvic bone image.

[0089] In another implementation, for the results of directly performing window transformation on CT images, a deep learning-based method can be used for segmentation instead, which is more flexible.

[0090] For the foregoing step S204, an embodiment of the present invention provides a specific implementation manner of determining multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generating a 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image, including: according to the fluctuation range of the X-ray image, generating a 2D overall pelvic bone image rotated based on the X-axis and the Y-axis for the CT image, for example, generating a 2D overall pelvic bone image every two degrees in the X-axis direction and the Y-axis direction respectively.

[0091] For the foregoing step S206, an embodiment of the present invention provides a specific implementation manner of performing feature point matching on the 2D medical image and the 2D overall pelvic bone image, including: through a feature point matching model, based on the self-attention and cross-attention mechanisms, extracting and matching feature points on the 2D medical image and the 2D overall pelvic bone image to obtain a feature point matching result.

[0092] Specifically, by performing feature point matching on the 2D overall pelvic bone image and the X-ray image, the relationship of the feature points can be obtained. In the experiment, the ASpanFormer feature point matching network was used as the feature point matching tool. This network model is a network structure based on Transformer, and uses the self-attention and cross-attention mechanisms to simultaneously complete the process of feature point extraction and matching, and obtains, for example, Figure 5 As shown in a schematic diagram of the AspanFormer feature point matching result. Compared with the traditional method, the ASpanFormer model based on deep learning has better matching results in the extraction and matching of feature points of medical images.

[0093] For the foregoing step S206, an embodiment of the present invention further provides a specific implementation manner of performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine the target 2D local pelvic bone image matched by the 2D medical image, including the following steps B1 to B3:

[0094] Step B1, based on the feature point matching result, dynamically and adaptively determining the target size and position information of the cropping window corresponding to each 2D overall pelvic bone image.

[0095] In one implementation manner, the size and position information of the cropping window can be dynamically and adaptively adjusted based on a sliding window, or the size and position information of the cropping window can be dynamically and adaptively adjusted based on a neural network. Specifically as follows:

[0096] Method 1, dynamically and adaptively adjusting the size and position information of the cropping window based on a sliding window, and performing the following operations on any 2D overall pelvic bone image:

[0097] (1.1) Based on the feature point matching results, determine the initial size and position information of the cropping window corresponding to the 2D overall pelvic bone image, and set a buffer area for the 2D overall pelvic bone image based on the initial size and position information. Exemplarily, based on the size and position information of the minimum bounding box of the feature point matching results in the 2D overall pelvic bone image, determine the initial size and position information of the cropping window corresponding to the 2D overall pelvic bone image, and then use the cropping window edge characterized by the initial size and position information of the cropping window. Taking the cropping window edge as the center, construct a buffer area based on a preset width.

[0098] (1.2) Set a set of adaptive windows in the buffer area. In one example, set multiple adaptive sliding windows (referred to as adaptive windows for short) within the buffer area range, and the set composed of the adaptive windows is called the adaptive window set.

[0099] (1.3) For each adaptive window in the adaptive window set, extract a set of feature point pairs from the 2D medical image and the 2D overall pelvic bone image based on the feature point matching results, so as to determine the confidence level corresponding to the adaptive window by using the set of feature point pairs.

[0100] In one implementation, the set of feature point pairs can be used to determine the local similarity of the 2D medical image and the 2D overall pelvic bone image relative to the adaptive window, and a confidence level is assigned to the adaptive window based on the local similarity. The confidence level is positively correlated with the local similarity.

[0101] Optionally, algorithms such as the triangle-based similarity evaluation algorithm, NCC (Normalized Cross-Correlation) algorithm, MSE (Mean-Square Error) algorithm, SSIM (Structural Similarity) algorithm, etc. can be used to determine the local similarity of the 2D medical image and the 2D overall pelvic bone image relative to the adaptive window.

[0102] Taking the triangle-based similarity evaluation algorithm as an example: for the feature point relationship constructed between two images, the corresponding point positions can be found. Considering that the angle does not change with the scaling and translation relationships in the plane and is only sensitive to the rotation relationship in three dimensions, as Figure 6 shown in a schematic diagram of a projection angle change, therefore, corresponding triangle pairs can be formed according to the feature point matching relationship, and the similarity of the triangle pairs can be measured to obtain the local similarity of the 2D medical image and the 2D overall pelvic bone image relative to the adaptive window.

[0103] Taking the SSIM algorithm as an example: Use the structural similarity index SSIM for 2D medical images, and the local similarity of this 2D overall pelvic bone image relative to the adaptive window. The structural similarity index SSIM is a commonly used image similarity evaluation index, which evaluates image similarity from three aspects: brightness, contrast, and structure.

[0104] (1.4) Determine the target size and position information of the cropping window according to the confidence corresponding to each adaptive window. Specifically, according to the confidence corresponding to each adaptive window, filter out the target adaptive window, and use the size and position information of the target adaptive window to adjust the initial size and position information of the cropping window to obtain the target size and position information of the cropping window. In one example, if the confidence of a certain adaptive window is relatively high, it can be used as the target adaptive window, and the cropping window can be enlarged or reduced according to the positional relationship between the target adaptive window and the cropping window; conversely, if the confidence of a certain adaptive window is relatively low, it is prohibited to use this adaptive window to enlarge or reduce the cropping window.

[0105] Method 2: Dynamically and adaptively adjust the size and position information of the cropping window based on a neural network. For any 2D overall pelvic bone image, perform the following operations: Through a pre-trained cropping and positioning model, based on the feature point matching results, extract the complete image features of the 2D medical image and the local image features of this 2D overall pelvic bone image respectively, determine the similarity between the complete image features and the local image features based on the cross-attention mechanism, and determine the target size and position information of the cropping window of this 2D overall pelvic bone image based on the similarity.

[0106] In specific implementation, first, the 2D medical image and the 2D overall pelvic bone image are respectively input into different convolutional neural networks (CNNs), so that the CNNs extract the complete image features of the 2D medical image and the local image features of the 2D overall pelvic bone image based on the feature point matching results. Then, the complete image features and the local image features are input into the cross-attention module. The cross-attention module is the core part of this network. It is used to calculate the similarity between the local image features and the complete image features and help the model "focus" on the most matching region in the image. Among them: Q (Query) represents the local image features; K (Key) and V (Value) represent the complete image features. The role of the cross-attention module is to calculate the correlation according to Q (Query) and K (Key), and then perform weighted summation on V (Value) based on the correlation to obtain a weighted feature representation. The specific steps are as follows: measure their correlation by calculating the dot product between Q (Query) and K (Key), such as using dot product attention; apply Softmax normalization to convert the calculated correlation into a probability distribution, representing the matching weights at each position; use these matching weights to perform weighted summation on V (Value), so as to obtain the final matching feature representation. Finally, further locate and crop the region according to the similarity. Among them, in the range with a higher similarity, pre-frame selection is performed, and the bounding box is further dynamically adjusted according to the similarity data to avoid truncating key structures.

[0107] In the embodiment of the present invention, the network is trained using multiple groups of local and global information. In particular, small-angle rotation of the overall image is added to increase the robustness of the network.

[0108] Step B2, crop the corresponding 2D overall pelvic bone image according to the target size and position information of the cropping window to obtain the 2D local pelvic bone image after cropping each 2D overall pelvic bone image. Exemplarily, refer to Figure 7 a schematic diagram of template matching completed based on the feature point matching relationship as shown. Among them, the left is the schematic diagram of feature point matching, and the right is the 2D local pelvic bone image T' after cropping.

[0109] Step B3, determine the target 2D local pelvic bone image matched by the 2D medical image according to the similarity between the 2D medical image and each 2D local pelvic bone image. In one example, by comparing the target X-ray image with the templates one by one, the 2D local pelvic bone image with the highest target similarity is used as the target 2D local pelvic bone image matched by the 2D medical image, and the X-axis and Y-axis rotation parameters used when generating the target 2D local pelvic bone image are recorded as the required r x , r y .

[0110] For the foregoing step S208, an embodiment of the present invention provides an implementation manner of performing rigid registration on a 3D medical image based on a target 2D local pelvic bone image and its corresponding alternative parameter combination, so that the rigidly registered 3D medical image is aligned with the 2D medical image, including: known parameter combination r x , r y , and the geometric relationship between the X-ray image and its matched target 2D local pelvic bone image, r z , t x , t y , t z can be gradually solved, that is, the complete rigid registration parameter combination r x , r y , r z , t x , t y , t z is obtained. Exemplarily, for any three feature points in the X-ray image, the feature points matching them are determined from the target 2D local pelvic bone image to form feature point pairs, and the rotation angle around the Z axis is determined based on the vector pairs constructed by the feature point pairs; and, the scaling coefficient is determined based on the feature points matching between the X-ray image and the target 2D local pelvic bone image, and the translation amount along the Z axis is determined based on the scaling coefficient and the distance from the target 2D local pelvic bone image to the radiation source; and, the pixel movement distance coefficient corresponding to the movement per unit physical distance is determined based on the template image with a known physical translation relationship, and the translation amounts along the X axis and the Y axis are determined according to the pixel movement distance coefficient and the scaling coefficient.

[0111] Project the CT image according to the rigid registration parameter combination to form a new 2D local pelvic bone image. If the similarity between the new 2D local pelvic bone image and the X-ray image meets the threshold, then perform rigid registration on the CT image using the rigid registration parameter combination, which can align the CT image with the X-ray image; otherwise, fine-tune the rigid registration parameter combination until the similarity between the new 2D local pelvic bone image and the X-ray image meets the threshold.

[0112] In summary, the embodiments of the present invention combine the technology of deep learning and use the feature point matching results for registration. This method is mainly used for registering the local X-ray images taken during the operation and the CT images taken before the operation for larger bone groups such as the pelvis during the operation. By using the technology of deep learning, the feature point matching can be made more accurate and rapid, and more sufficient feature point matching results can be obtained. Registering based on the feature point matching results can make the registration method more flexible. Performing template matching on the global image based on the feature point matching results can focus the information on the local area and make the registration more flexible. Moreover, since the deep learning feature point matching process does not require manual setting of feature points, the information can be processed faster and more automatically.

[0113] Based on the foregoing embodiments, the embodiments of the present invention provide a rigid registration device for local pelvic bone images. Refer to Figure 8 the structural schematic diagram of a rigid registration device for local pelvic bone images shown in

[0114] A 3D image extraction module 802, configured to obtain a 3D medical image to be registered, where the content displayed in the 3D medical image includes overall pelvic information, and extract a 3D pelvic bone image from the 3D medical image;

[0115] A 2D overall image generation module 804, configured to determine multiple groups of alternative parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate a 2D overall pelvic bone image corresponding to each group of alternative parameter combinations for the 3D pelvic bone image; wherein, a group of alternative parameter combinations includes alternative parameter values corresponding to multiple rigid registration parameter items;

[0116] A template matching module 806, configured to perform feature point matching on the 2D medical image and the 2D overall pelvic bone image, and perform dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching results to determine the target 2D local pelvic bone image matched by the 2D medical image;

[0117] A rigid registration module 808, configured to perform rigid registration on the 3D medical image based on the target 2D local pelvic bone image and its corresponding alternative parameter combination, so that the rigidly registered 3D medical image is aligned with the 2D medical image.

[0118] The rigid registration device for local pelvic bone images provided by the embodiments of the present invention is applicable to scenarios where it is difficult to obtain complete 3D medical images. After generating the 2D overall pelvic bone images corresponding to each set of alternative parameter combinations for the 3D pelvic bone images, feature point matching and dynamic adaptive template matching are successively performed on the 2D medical images and the 2D overall pelvic bone images to obtain the target 2D local pelvic bone images that match the 2D medical images. On this basis, combined with the alternative parameter combinations corresponding to the target 2D local pelvic bone images, rigid registration of the 3D medical images can be achieved. The present invention can effectively handle the rigid registration problem from local 2D X-ray imaging to global 3D CT data.

[0119] In one implementation, the template matching module 806 is specifically configured to:

[0120] Based on the feature point matching results, dynamically and adaptively determine the target size and position information of the cropping window corresponding to each 2D overall pelvic bone image;

[0121] Crop the corresponding 2D overall pelvic bone image according to the target size and position information of the cropping window to obtain the cropped 2D local pelvic bone image of each 2D overall pelvic bone image;

[0122] Determine the target 2D local pelvic bone image that matches the 2D medical image according to the similarity between the 2D medical image and each 2D local pelvic bone image.

[0123] In one implementation, the template matching module 806 is specifically configured to:

[0124] Perform the following operations on any one of the 2D overall pelvic bone images:

[0125] Based on the feature point matching results, determine the initial size and position information of the cropping window corresponding to the 2D overall pelvic bone image, and set a buffer area for the 2D overall pelvic bone image based on the initial size and position information;

[0126] Set a set of adaptive windows within the buffer area;

[0127] For each adaptive window within the set of adaptive windows, extract a set of feature point pairs from the 2D medical image and the 2D overall pelvic bone image based on the feature point matching results to determine the confidence level corresponding to the adaptive window using the set of feature point pairs;

[0128] Determine the target size and position information of the cropping window according to the confidence level corresponding to each adaptive window.

[0129] In one implementation, the template matching module 806 is specifically configured to:

[0130] Using the set of feature point pairs, determine the local similarity of the 2D medical image and the 2D overall pelvic bone image relative to the adaptive window; assign a confidence level to the adaptive window based on the local similarity; wherein, the confidence level is positively correlated with the local similarity.

[0131] In one implementation, the template matching module 806 is specifically configured to:

[0132] According to the confidence level corresponding to each adaptive window, filter out the target adaptive window, and use the size and position information of the target adaptive window to adjust the initial size and position information of the cropping window, so as to obtain the target size and position information of the cropping window.

[0133] In one implementation, the template matching module 806 is specifically configured to:

[0134] Perform the following operations on any 2D overall pelvic bone image:

[0135] Through a pre-trained cropping and positioning model, based on the feature point matching result, respectively extract the complete image features of the 2D medical image and the local image features of the 2D overall pelvic bone image, determine the similarity between the complete image features and the local image features based on the cross-attention mechanism, and determine the target size and position information of the cropping window of the 2D overall pelvic bone image based on the similarity.

[0136] In one implementation, the 3D image extraction module 802 is specifically configured to:

[0137] Determine the HU value corresponding to each voxel in the 3D medical image;

[0138] Based on multiple window transformation parameters and the HU value corresponding to each voxel, divide the 3D medical image into a background and non-bone part and a pelvic bone main structure part;

[0139] Use the background and non-bone part to remove the background and non-bone tissues in the 3D medical image, and combine the pelvic bone main structure part to extract the 3D pelvic bone image.

[0140] In one implementation, the multi-level template matching module 806 is specifically configured to:

[0141] Through a feature point matching model, based on the self-attention and cross-attention mechanisms, perform feature point extraction and feature point matching on the 2D medical image and the 2D overall pelvic bone image to obtain a feature point matching result.

[0142] The device provided by the embodiments of the present invention has the same implementation principle and the same technical effects as the foregoing method embodiments. For the sake of brief description, for the parts not mentioned in the device embodiments, reference may be made to the corresponding content in the foregoing method embodiments.

[0143] An embodiment of the present invention provides an electronic device. Specifically, the electronic device includes a processor and a storage device; a computer program is stored on the storage device, and when the computer program is run by the processor, it executes the method described in any one of the above-described embodiments.

[0144] Figure 9 FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 100 includes: a processor 90, a memory 91, a bus 92, and a communication interface 93. The processor 90, the communication interface 93, and the memory 91 are connected through the bus 92; the processor 90 is configured to execute an executable module stored in the memory 91, such as a computer program.

[0145] Among them, the memory 91 may include a high-speed random access memory (RAM, Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 93 (which may be wired or wireless), a communication connection is established between the system network element and at least one other network element, and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0146] The bus 92 may be an ISA bus, a PCI bus, an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 9 only a bidirectional arrow is used in FIG. 5, but it does not mean that there is only one bus or one type of bus.

[0147] Among them, the memory 91 is used to store a program. After receiving an execution instruction, the processor 90 executes the program. The method executed by the device defined by the flow process disclosed in any one of the foregoing embodiments of the present invention can be applied to the processor 90 or implemented by the processor 90.

[0148] The processor 90 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 90 or the instructions in the form of software. The above-mentioned processor 90 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 91, and the processor 90 reads the information in the memory 91 and combines its hardware to complete the steps of the above method.

[0149] The computer program product of the readable storage medium provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, reference can be made to the foregoing method embodiments and will not be elaborated herein.

[0150] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0151] Finally, it should be noted that the above-mentioned embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the technical field can still modify the technical solutions described in the foregoing embodiments, or easily conceive of changes, or perform equivalent replacements for some of the technical features within the technical scope disclosed by the present invention; and these modifications, changes, or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for rigid registration of local pelvic bone images, characterized in that: include: Acquire a 3D medical image to be registered, wherein the content displayed by the 3D medical image includes overall pelvic information, and extract a 3D pelvic bone image from the 3D medical image; Determine multiple groups of candidate parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate a 2D whole pelvic bone image corresponding to each group of the candidate parameter combinations for the 3D pelvic bone image; wherein a group of the candidate parameter combinations includes candidate parameter values ​​corresponding to multiple rigid registration parameter items; Performing feature point matching on the 2D medical image and the 2D overall pelvic bone image, and performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine a target 2D local pelvic bone image for matching the 2D medical image; Based on the target 2D local pelvic bone image and the corresponding candidate parameter combination, the 3D medical image is rigidly registered so that the rigidly registered 3D medical image is aligned with the 2D medical image.

2. The local pelvic bone image rigid registration method according to claim 1, characterized in that: Performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result to determine a target 2D local pelvic bone image for matching the 2D medical image includes: Based on the feature point matching results, dynamically and adaptively adjust and determine the target size and position information of the cropping window corresponding to each of the 2D whole pelvic bone images; According to the target size and position information of the cropping window, the corresponding 2D whole pelvic bone image is cropped to obtain a 2D local pelvic bone image after each of the 2D whole pelvic bone images is cropped; According to the similarity between the 2D medical image and each of the 2D local pelvic bone images, a target 2D local pelvic bone image for matching the 2D medical image is determined.

3. The local pelvic bone image rigid registration method according to claim 2, characterized in that: Based on the feature point matching results, dynamically and adaptively adjusting and determining the target size and position information of the cropping window corresponding to each of the 2D whole pelvic bone images, including: For any of the 2D whole pelvic bone images, perform the following operations: Based on the feature point matching result, determine the initial size and position information of the cropping window corresponding to the 2D whole pelvic bone image, and set a buffer area for the 2D whole pelvic bone image based on the initial size and position information; Setting an adaptive window set in the buffer area; For each adaptive window in the adaptive window set, extracting a feature point pair set from the 2D medical image and the 2D whole pelvic bone image based on the feature point matching result, so as to determine the confidence corresponding to the adaptive window by using the feature point pair set; According to the confidence level corresponding to each adaptive window, the target size and position information of the cropping window are determined.

4. The local pelvic bone image rigid registration method according to claim 3, characterized in that: Determining the confidence level corresponding to the adaptive window using the feature point pair set includes: Determine the local similarity of the 2D medical image and the 2D whole pelvic bone image relative to the adaptive window by using the feature point pair set; assign confidence to the adaptive window based on the local similarity; wherein the confidence is positively correlated with the local similarity; Determining target size and position information of the cropping window according to the confidence corresponding to each adaptive window includes: According to the confidence corresponding to each of the adaptive windows, a target adaptive window is screened out, and the initial size and position information of the cropping window are adjusted using the size and position information of the target adaptive window to obtain the target size and position information of the cropping window.

5. The local pelvic bone image rigid registration method according to claim 2, characterized in that: Dynamically and adaptively adjusting and determining the target size and position information of the cropping window corresponding to each of the 2D whole pelvic bone images, further comprising: For any of the 2D whole pelvic bone images, perform the following operations: Through the pre-trained cropping and positioning model, based on the feature point matching results, the complete image features of the 2D medical image and the local image features of the 2D whole pelvic bone image are extracted respectively, and the similarity between the complete image features and the local image features is determined based on the cross-attention mechanism, and the target size and position information of the cropping window of the 2D whole pelvic bone image is determined based on the similarity.

6. The local pelvic bone image rigid registration method according to claim 1, characterized in that: Extracting a 3D pelvic bone image from the 3D medical image includes: Determine a HU value corresponding to each voxel in the 3D medical image; Based on a plurality of window transformation parameters and the HU value corresponding to each voxel, the 3D medical image is divided into a background and non-bone part and a pelvic bone main structure part; The background and non-bone parts are used to remove the background and non-bone tissues in the 3D medical image, and the 3D pelvic bone image is extracted in combination with the main structure of the pelvic bone.

7. The local pelvic skeleton image rigid registration method according to claim 1, characterized in that: Performing feature point matching on the 2D medical image and the 2D whole pelvic bone image includes: Through the feature point matching model, based on the self-attention and cross-attention mechanisms, feature point extraction and feature point matching are performed on the 2D medical image and the 2D whole pelvic bone image to obtain a feature point matching result.

8. A local pelvic bone image rigid registration device, characterized in that: include: A 3D image extraction module, used to obtain a 3D medical image to be registered, wherein the content displayed by the 3D medical image includes overall pelvic information, and extract a 3D pelvic bone image from the 3D medical image; A 2D whole image generation module, used to determine multiple groups of candidate parameter combinations according to the fluctuation range of the 2D medical image corresponding to the 3D medical image, and generate a 2D whole pelvic bone image corresponding to each group of the candidate parameter combinations for the 3D pelvic bone image; wherein a group of the candidate parameter combinations includes candidate parameter values ​​corresponding to multiple rigid registration parameter items; A template matching module, used for performing feature point matching on the 2D medical image and the 2D overall pelvic bone image, and performing dynamic adaptive template matching on the 2D medical image and the 2D overall pelvic bone image based on the feature point matching result, so as to determine a target 2D local pelvic bone image for matching the 2D medical image; A rigid registration module is used to perform rigid registration on the 3D medical image based on the target 2D local pelvic bone image and the corresponding candidate parameter combination, so that the rigidly registered 3D medical image is aligned with the 2D medical image.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images