Automatic slice positioning method, device, medium and product in 3D cardiac CT

Abstract

The application discloses an automatic section positioning method, device, medium and product in 3D heart CT, and relates to the field of image section positioning. The method uses a block random sampling strategy to perform image resampling on a 3D heart CT image aligned to a standard 3D heart atlas space to obtain training sample data; a positioning regression network is trained in a self-supervised manner using the training sample data; a query image is input into the trained positioning regression network to obtain a section position of the query image in the atlas space; a candidate position of the query image in a target 3D heart CT space is obtained through a rigid transformation based on the section position; image resampling is performed at the candidate position to obtain a candidate database; and an image similarity algorithm is used to find a candidate image section with the highest similarity in the candidate database, so that the position of the query image in the target 3D heart CT space is obtained, thereby realizing positioning and viewing of any section and improving the positioning efficiency of any section.

Description

Technical Field

[0001] This application relates to the field of image section localization, and in particular to an automatic section localization method, apparatus, device, medium and product in 3D cardiac CT. Background Technology

[0002] In clinical practice, the cross-sectional views of cardiac CT scans help doctors quickly locate and view images, providing data references for disease diagnosis, cross-sectional planning, and surgical planning. However, standard cross-sectional views of cardiac CT scans are mostly based on specific definitions of cardiac anatomy, and coupled with significant individual differences, cross-sectional positioning heavily relies on expert experience. Furthermore, due to the large search space of 3D cardiac CT, manual cross-sectional positioning consumes a significant amount of doctors' time, resulting in low efficiency.

[0003] Existing methods for sectional localization in cardiac CT scans mainly fall into two categories: The first is the traditional method, which relies on manually designing and selecting features to locate the target cardiac section. However, this method is highly sensitive to feature selection, leading to serious localization errors and poor generalization. Furthermore, significant individual differences make section localization heavily dependent on expert experience. Additionally, the large search space of 3D cardiac CT necessitates manual section localization, which is time-consuming and inefficient. The second method is based on deep learning. This method primarily uses a trained neural network model to acquire prior knowledge of the entire 3D cardiac CT image (e.g., anatomical segmentation, key point labeling, etc.) for section localization. However, this method requires manually labeling a large number of high-quality anatomical structure samples, resulting in significant time and labor costs.

[0004] Furthermore, existing technologies are all methods and models designed for specific cross-sections, and cannot locate and view arbitrary cross-sections. However, in actual clinical cardiac CT examinations and surgical planning, the cross-sections are not fixed, but different cross-sections are selected according to the differences of different individuals and different diseases for diagnosis and planning.

[0005] In summary, most existing cross-section localization methods are based on prior knowledge such as segmentation or key points, heavily relying on time-consuming and labor-intensive manual annotation or expert experience. Furthermore, the trained models are limited to the localization of specific cross-sections and cannot achieve the localization and viewing of arbitrary cross-sections. Summary of the Invention

[0006] The purpose of this application is to provide an automatic section positioning method, device, equipment, medium, and product in 3D cardiac CT, which can realize the positioning and viewing of any section, and at the same time, can improve the positioning efficiency of any section.

[0007] To achieve the above objectives, this application provides the following solution:

[0008] Firstly, this application provides an automatic section localization method in 3D cardiac CT, comprising:

[0009] Standard 3D cardiac atlases are generated from 3D cardiac CT images aligned to a uniform space.

[0010] Using a block random sampling strategy, CT images in the standard 3D cardiac atlas space are resampled to obtain 2D sections, the positions of the 2D sections are determined, and training sample data is generated.

[0011] The localization and regression network is trained using the training sample data in a self-supervised manner until the set conditions are met, and the trained localization and regression network is obtained.

[0012] The query image is input into the trained localization and regression network to obtain the cross-sectional position of the query image in the spectral space.

[0013] Based on the cross-sectional position of the query image, the candidate position of the query image in the target 3D cardiac CT space is obtained through rigid transformation between the standard 3D cardiac atlas and the target 3D cardiac CT space.

[0014] Image resampling is performed at candidate locations in the target 3D cardiac CT space to obtain candidate image sections, thereby generating a candidate database;

[0015] Using an image similarity algorithm, the candidate image section with the highest similarity to the query image is found in the candidate database, and the position of this candidate image section is used as the localization result of the query image to complete the automatic localization of the section in 3D cardiac CT.

[0016] Optionally, a standard 3D cardiac atlas is generated based on 3D cardiac CT images aligned to a uniform space, specifically including:

[0017] A CT space is randomly selected from all 3D cardiac CT spaces as the CT image reference space, and all remaining 3D cardiac CT images are aligned to the CT image reference space using a symmetric differential homeomorphic deformable nonlinear registration algorithm.

[0018] The standard 3D cardiac atlas is generated using an averaging algorithm based on all CT images aligned to the CT image reference space.

[0019] Optionally, the standard 3D cardiac atlas is generated using an averaging algorithm based on all CT images aligned to the CT image reference space, specifically including:

[0020] An initial template is obtained by averaging all CT images aligned to the CT image reference space.

[0021] The symmetric differential homeomorphic deformable nonlinear registration algorithm is used to register all CT images aligned to the CT image reference space to the initial template.

[0022] The initial template was adjusted using a sharpening filter to obtain the standard 3D cardiac atlas.

[0023] Optionally, the loss function of the localization regression network is:

[0024]

[0025] In the formula, λ represents the weight of the translation loss, r represents the rotation in the true rigid transformation, and t represents the translation in the true rigid transformation. Represents the rotation in the predicted rigid transformation. This represents the translation in the predicted rigid transformation. Represents rotation in the predicted rigid transformation The geodesic distance error between the rotation r in the true rigid transformation and the rotation r is Represents translation in the predicted rigid transformation The mean square error between the translation t in the true rigid transformation and the loss total This indicates the value used for loss.

[0026] Optionally, the candidate location of the query image in the target 3D cardiac CT space is represented as follows:

[0027]

[0028] In the formula, The candidate location is indicated by RT, which represents the rigid transformation between the standard 3D cardiac atlas and the target 3D cardiac CT space. This indicates the position of the cross section of the queried image in the spectral space.

[0029] Optionally, the candidate image slice with the highest similarity to the query image is represented as:

[0030]

[0031] In the formula, This represents the candidate image slice that has the highest similarity to the query image. Let represent the i-th candidate image slice in the candidate database, and m represent the number of candidate image slices. Represents the i-th candidate image slice The image similarity to the query image q.

[0032] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the automatic section positioning method in 3D cardiac CT as described above.

[0033] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the automatic section localization method in 3D cardiac CT as described above.

[0034] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the automatic section localization method in 3D cardiac CT as described above.

[0035] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0036] This application provides an automatic section localization method, device, medium, and product in 3D cardiac CT. By generating a standard 3D cardiac atlas, it solves the problem of inconsistent standard section views caused by large individual differences in cardiac structure. Guided by the standard 3D cardiac atlas, it achieves automatic localization and viewing of any section in 3D cardiac CT using a coarse-to-fine approach (i.e., determining the final query image location from candidate positions). Furthermore, using a self-supervised approach, the localization regression network is trained with the training sample data, eliminating the need for manual annotation and thus improving the localization efficiency of any section. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is an application environment diagram of an automatic section positioning method in 3D cardiac CT according to an embodiment of this application;

[0039] Figure 2 A flowchart illustrating an automatic section localization method in 3D cardiac CT provided in an embodiment of this application;

[0040] Figure 3 An implementation architecture diagram of an automatic section localization method in 3D cardiac CT provided in an embodiment of this application;

[0041] Figure 4 An array diagram of automatic section positioning results provided in an embodiment of this application;

[0042] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] The automatic section localization method in 3D cardiac CT provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send a query image to server 104. After receiving the query image, server 104 inputs it into a trained localization regression network to obtain the cross-sectional position of the query image in the atlas space. Based on the cross-sectional position of the query image, a rigid transformation between the standard 3D cardiac atlas and the target 3D cardiac CT space is used to obtain candidate positions of the query image in the target 3D cardiac CT space. Image resampling is performed at the candidate positions in the target 3D cardiac CT space to obtain candidate image cross-sections, generating a candidate database. Using an image similarity algorithm, the candidate image cross-section with the highest similarity to the query image is found in the candidate database, and the position of this candidate image cross-section is used as the localization result of the query image to complete the automatic localization of the cross-section in the 3D cardiac CT. Server 104 can feed back the obtained localization result of the query image to terminal 102. In addition, in some embodiments, the automatic section localization method in 3D cardiac CT can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly perform automatic localization on the query image, or the server 104 can obtain the query image from the data storage system and perform automatic localization processing on the query image.

[0046] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0047] In one exemplary embodiment, such as Figure 2 As shown, an automatic section localization method is provided in 3D cardiac CT. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes steps 200 to 206. Wherein:

[0048] Step 200: Generate a standard 3D cardiac atlas based on 3D cardiac CT images aligned to a uniform space.

[0049] Step 201: Using a block random sampling strategy, resample the CT images in the standard 3D cardiac atlas space to obtain 2D cross-sections, determine the location of the 2D cross-sections, and generate training sample data.

[0050] Step 202: Train the localization and regression network using training sample data in a self-supervised manner until the set conditions are met, and then obtain the trained localization and regression network.

[0051] Step 203: Input the query image into the trained localization and regression network to obtain the cross-sectional position of the query image in the graph space.

[0052] Step 204: Based on the cross-sectional position of the query image, obtain the candidate position of the query image in the target 3D cardiac CT space through rigid transformation of the standard 3D cardiac atlas and the target 3D cardiac CT space.

[0053] Step 205: Resample the image at the candidate location in the target 3D cardiac CT space to obtain candidate image sections, thereby generating a candidate database.

[0054] Step 206: Using an image similarity algorithm, find the candidate image section with the highest similarity to the query image in the candidate database, and use the position of this candidate image section as the localization result of the query image to complete the automatic localization of the section in 3D cardiac CT.

[0055] By implementing steps 201 to 206 above, this application enables the localization and viewing of arbitrary cross-sections. Furthermore, this application employs a self-supervised approach, using training sample data to train the localization regression network, which eliminates the need for manual annotation and thus improves the localization efficiency of arbitrary cross-sections.

[0056] In another exemplary embodiment of this application, to address the issue of inconsistent standard cross-sectional views due to significant individual differences in cardiac structure, and to improve the accuracy of slice positioning, a standard 3D cardiac atlas is generated by unifying the coordinate system and registering it to a 3D cardiac CT space under the unified coordinate system. Based on this, step 200 is replaced by steps 301-302:

[0057] Step 301: Randomly select one CT space from all 3D cardiac CT spaces as the CT image reference space, and align all remaining 3D cardiac CT images to the CT image reference space using a symmetric differential homeomorphic deformable nonlinear registration algorithm.

[0058] Step 302: Generate a standard 3D cardiac atlas based on all CT images aligned to the CT image reference space using an averaging algorithm.

[0059] The specific implementation process of step 302 can be described as follows:

[0060] (1) Average all CT images aligned to the CT image reference space to obtain the initial template.

[0061] (2) A symmetric differential homeomorphic deformable nonlinear registration algorithm is used to register all CT images aligned to the CT image reference space to the initial template.

[0062] (3) Use a sharpening filter to adjust the initial template to obtain a standard 3D heart atlas.

[0063] In practical applications, after obtaining the initial template in step (2), a new template can be created by averaging all the deformed CT images. Then, all transformations from step (2) are averaged to create a single transformation. Applying this single transformation will distort the atlas towards the true average shape. Then, the initial template is continuously adjusted using a sharpening filter to enhance the edges.

[0064] In another exemplary embodiment of this application, in order to train a localization regression network in a unified coordinate system to eliminate the requirement of manual annotation and achieve automatic cross-section localization based on arbitrary input cross-sections instead of fixed cross-sections, this application trains customized atlas cues using a self-supervised method to locate target cross-sections for different input query images. Based on this, step 202 is replaced by steps 401 to 402:

[0065] Step 401: Parameterize the position of the 2D section in the 3D cardiac CT space, denoted as p. i ={r i ,t i}, which is a rigid transformation represented by rotation and translation.

[0066] Step 402: To obtain n pairs of training data (2D cross-sections and their corresponding positions), this application utilizes a block random sampling strategy to resample a large number of 2D cross-sections in 3D cardiac CT under a unified coordinate system.

[0067] Step 403: Train the localization regression network PosNet using a large amount of sampled paired data. The localization regression network PosNet is denoted as γ(·μ), where μ is the parameter in the localization regression network PosNet. The localization regression network PosNet consists of two parts: the encoder module γ... enc (·;μ enc ) and the localization prediction module γ pos (·;μ pos μ enc It is the encoder module γ enc The parameter in, μ pos It is the localization prediction module γ pos The parameters in the training data. Each 2D slice in the training data. Input to encoder module γ enc (·;μ enc 0 to generate a feature vector f i ,for:

[0068]

[0069] Step 404: By using the feature vector Input to the localization prediction module γ pos (·;μ pos To predict standard 3D cardiac atlases The position of the cross section for:

[0070]

[0071] Step 405: The loss function is designed as the sum of the mean squared error losses between the geodesic distance and translation of 3D rotation and the true distance and translation in the rigid transformation, denoted as:

[0072]

[0073] In the formula, λ represents the weight of the translation loss, r represents the rotation in the true rigid transformation, and t represents the translation in the true rigid transformation. Represents the rotation in the predicted rigid transformation. This represents the translation in the predicted rigid transformation. Represents rotation in the predicted rigid transformation The geodesic distance error between the rotation r in the true rigid transformation and the rotation r is Represents translation in the predicted rigid transformation The mean square error between the translation t in the true rigid transformation and the loss total This indicates the value used for loss.

[0074] In another exemplary embodiment of this application, in order to achieve coarse localization of the query image, step 204 provided above can be replaced by steps 501 to 504:

[0075] Step 501: The query image q is input into the trained localization and regression network PosNet to obtain the query image q in a standard 3D cardiac atlas. The position in the middle is denoted as

[0076]

[0077] Step 502: Standard 3D Cardiac Atlas And query image q in standard 3D cardiac atlas The position in the middle Spectral hints that make up the query image q

[0078] Step 503: For coarse localization, standard 3D cardiac atlas The rigid transformation RT between the two is obtained by rigidly registering the target 3D cardiac CT space v to the target.

[0079] Step 504: The rigid transformation RT is applied to the position. This yields the coarse location (i.e., candidate location) in the target 3D cardiac CT space v, denoted as...

[0080] In another exemplary embodiment of this application, after coarse localization, in order to achieve fine localization of the query image, so as to achieve automatic localization of any section in 3D cardiac CT in a coarse-to-fine manner guided by atlas prompts, steps 205 and 206 provided above in this application can be replaced by steps 601 to 602:

[0081] Step 601: Resample candidate image sections at coarse locations (i.e., candidate locations) in the target 3D cardiac CT space v to form a candidate database.

[0082] Step 602: Use image similarity to find the most similar cross section Furthermore, the location of the cross-section in the target 3D cardiac CT scan is obtained. The image similarity method used here is structural similarity, which is the representation of structural similarity by the human visual system, evaluating the similarity of images from aspects such as brightness, contrast, and structure.

[0083] The candidate image cross-section with the highest similarity to the query image is represented as follows:

[0084]

[0085] In the formula, This represents the candidate image slice with the highest similarity to the query image. Let represent the i-th candidate image slice in the candidate database, and m represent the number of candidate image slices. Represents the i-th candidate image slice The image similarity to the query image q.

[0086] In summary, guided by the atlas hints of the generated corresponding query image q, and through standard 3D cardiac atlases... and target 3D cardiac CT space The rigid transformation yields the query image in the target 3D cardiac CT space. Candidate positions Then, the final target cross section is determined by using image similarity based on the candidate cross sections generated at the candidate locations. and its corresponding position

[0087] Based on the above description, the overall implementation architecture of the automatic section localization method in 3D cardiac CT provided in this application can be found in [reference needed]. Figure 3 The process can be summarized into two parts: atlas prompting and section localization. The atlas prompting part includes generating a standard 3D cardiac atlas and training a localization regression network in a self-supervised manner, i.e., steps 200-202 above. The section localization part includes coarse localization and fine localization, i.e., steps 203-206.

[0088] Furthermore, the advantages of the automatic section localization method in 3D cardiac CT provided in this application are illustrated by experimental methods.

[0089] Figure 4 The first column represents three query planes in arbitrary CT space: 2C (major axis 2-cavity), 4C (major axis 4-cavity), and X (arbitrary plane). For Figure 4 For Case 1 and Case 2, the first column is the positioning result of this application, the second column is the annotation result of the radiology expert, and the third column is a visual comparison result of the results of this application and the radiology expert. Figure 4 In the diagram, green represents the location results of this application, and magenta represents the results marked by radiology experts. Figure 4 Case 1 and Case 2 in the examples are query results corresponding to three query sections in different target CT spaces. Specifically, Case 1 shows the results in the first target 3D cardiac CT space. The first sub-image in the first row is the section corresponding to 2C located by the query in this application; the third sub-image in the first row is the section corresponding to 2C annotated by the doctor; and the fourth sub-image in the first row is a visualization of the section located by the query in this application and the doctor-annotated section. The second sub-image in the second row is the section corresponding to 4C located by the query in this application; the third sub-image in the second row is the section corresponding to 4C annotated by the doctor; and the fourth sub-image in the second row is a visualization of the section located by the query in this application and the doctor-annotated section. The second sub-image in the third row is the section corresponding to X located by the query in this application; the third sub-image in the third row is the section corresponding to X annotated by the doctor; and the fourth sub-image in the third row is a visualization of the section located by the query in this application and the doctor-annotated section. Figure 4 Case 2 is the result in the second target 3D cardiac CT space. Its detailed description is similar to that of Case 1, so it will not be repeated here.

[0090] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores 3D cardiac CT slice localization data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an automatic slice localization method in 3D cardiac CT.

[0091] Those skilled in the art will understand that Figure 5The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0092] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0093] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0095] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An automatic section positioning method in 3D cardiac CT, characterized in that, The automatic section localization method in 3D cardiac CT includes: Standard 3D cardiac atlases are generated from 3D cardiac CT images aligned to a uniform space. Using a block random sampling strategy, CT images in the standard 3D cardiac atlas space are resampled to obtain 2D sections, the positions of the 2D sections are determined, and training sample data is generated. The PosNet localization regression network is trained using the training sample data in a self-supervised manner until the set conditions are met, and then the trained localization regression network is obtained. The query image is input into the trained localization and regression network to obtain the cross-sectional position of the query image in the spectral space. Based on the cross-sectional position of the query image, the candidate position of the query image in the target 3D cardiac CT space is obtained through rigid transformation between the standard 3D cardiac atlas and the target 3D cardiac CT space. Image resampling is performed at candidate locations in the target 3D cardiac CT space to obtain candidate image sections, thereby generating a candidate database; Using an image similarity algorithm, the candidate image section with the highest similarity to the query image is found in the candidate database, and the position of this candidate image section is used as the localization result of the query image to complete the automatic localization of the section in 3D cardiac CT.

2. The automatic section positioning method in 3D cardiac CT according to claim 1, characterized in that, Generate standard 3D cardiac atlases based on 3D cardiac CT images aligned to a unified space, specifically including: A CT space is randomly selected from all 3D cardiac CT spaces as the CT image reference space, and all remaining 3D cardiac CT images are aligned to the CT image reference space using a symmetric differential homeomorphic deformable nonlinear registration algorithm. The standard 3D cardiac atlas is generated using an averaging algorithm based on all CT images aligned to the CT image reference space.

3. The automatic section positioning method in 3D cardiac CT according to claim 2, characterized in that, The standard 3D cardiac atlas is generated using an averaging algorithm based on all CT images aligned to the CT image reference space, specifically including: An initial template is obtained by averaging all CT images aligned to the CT image reference space. The symmetric differential homeomorphic deformable nonlinear registration algorithm is used to register all CT images aligned to the CT image reference space to the initial template. The initial template was adjusted using a sharpening filter to obtain the standard 3D cardiac atlas.

4. The automatic section positioning method in 3D cardiac CT according to claim 1, characterized in that, The loss function of the localization regression network PosNet is: In the formula, λ represents the weight of the translation loss, r represents the rotation in the true rigid transformation, and t represents the translation in the true rigid transformation. Represents the rotation in the predicted rigid transformation. This represents the translation in the predicted rigid transformation. Represents rotation in the predicted rigid transformation The geodesic distance error between the rotation r in the true rigid transformation and the rotation r is Represents translation in the predicted rigid transformation The mean square error between the translation t in the true rigid transformation and the loss total This represents the loss value.

5. The automatic section positioning method in 3D cardiac CT according to claim 1, characterized in that, The candidate positions of the query image in the target 3D cardiac CT space are represented as follows: In the formula, The candidate location is indicated by RT, which represents the rigid transformation between the standard 3D cardiac atlas and the target 3D cardiac CT space. This indicates the position of the cross section of the queried image in the spectral space.

6. The automatic section positioning method in 3D cardiac CT according to claim 1, characterized in that, The candidate image slice with the highest similarity to the query image is represented as: In the formula, This represents the candidate image slice that has the highest similarity to the query image. Let represent the i-th candidate image slice in the candidate database, and m represent the number of candidate image slices. Represents the i-th candidate image slice The image similarity to the query image q.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the automatic section localization method in 3D cardiac CT according to any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the automatic section localization method in 3D cardiac CT as described in any one of claims 1-6.

9. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the automatic section localization method in 3D cardiac CT as described in any one of claims 1-6.

Images