Heart CTA image group registration template construction method based on deep learning

Abstract

The invention discloses a heart CTA image group registration template construction method based on deep learning, and aims to solve the problems of fuzzy template, low construction efficiency and unstable alignment caused by significant deformation and intensity difference of the heart under different cardiac phases and scanning conditions. The method comprises the following steps: acquiring and preprocessing a heart CTA image; grouping is carried out based on feature similarity and topology persistent coherence; a multi-resolution three-dimensional registration network is adopted in each level to carry out coarse-to-fine registration estimation; differential deformation is realized through a space converter, and end-to-end and symmetric training is carried out by adopting combined loss of multi-scale reconstruction similarity and displacement field smooth regularization; clear sub-templates are obtained through iteration of a registration-averaging-updating strategy in the sub-groups, and finally, the sub-templates are converged step by step to obtain a global heart template, so that efficient and stable heart template construction is realized.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing and computer vision, specifically relating to a method for constructing a registration template for cardiac CTA image clusters based on deep learning. Background Technology

[0002] Currently, cardiac CTA offers high resolution and high contrast in depicting coronary artery and myocardial structures. However, due to differences in cardiac cycle, respiratory motion, contrast agents, and scanning protocols, samples within the same population exhibit significant non-rigid displacement and intensity variations. Traditional template construction often relies on single-scale or optimization-based registration strategies, which are easily affected by initial values, parameters, and sample differences, leading to template blurring and inefficient construction. Direct group registration across the entire population also presents optimization challenges due to significant population differences. Therefore, there is an urgent need for a template construction method that combines divide-and-conquer population-level organization with coarse-to-fine depth multi-resolution registration capabilities, capable of capturing large deformations while aligning detailed boundaries, thereby improving template clarity and construction efficiency. Summary of the Invention

[0003] In view of the problems existing in the construction of cardiac CTA image cluster registration templates based on deep learning, this invention is proposed.

[0004] Therefore, the purpose of this invention is to solve the problems that construction of registration strategies that rely on single scales or optimization are easily affected by initial values, parameters and sample differences, resulting in template ambiguity and inefficient construction; and that direct group registration on the entire population is also difficult to optimize due to large population differences.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide a method for constructing a cardiac CTA image group registration template based on deep learning. The method includes: acquiring cardiac CTA images for template construction, forming a set of fixed and moving images; preprocessing the cardiac CTA images, including intensity normalization and spatial normalization, to obtain uniform voxel sizes and cardiac regions of interest; dividing the images into subgroups based on sample feature similarity and topological persistent coherence, forming a bottom-up hierarchical subgroup sequence; using a multi-resolution 3D registration network within each level to perform coarse-to-fine non-rigid registration of the moving / fixed images, outputting a multi-scale displacement field and completing differentiable deformation through a spatial transformer; generating sub-templates within the subgroups using an iterative strategy of registration-aggregation-update, and gradually converging the lower-level sub-templates into a global template; and performing symmetric training and optimization using a combination loss of multi-scale reconstruction similarity and displacement field smoothing regularization within an end-to-end framework to evaluate template quality and deformation physicality.

[0007] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the cardiac CTA includes ECG-gated arterial phase scanning, injection of non-ionic contrast agent via peripheral vein, acquisition completed by intelligent triggering, and flushing with physiological saline after acquisition.

[0008] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the preprocessing includes window cropping and Z-score normalization of CT values; voxel resampling to an isotropic resolution of 1.0–1.25 mm; and cropping the cardiac region of interest within the thoracic cavity and unifying it to a standard size of 192×192×192 voxels.

[0009] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the subgroup division is obtained by calculating the feature similarity map between samples and combining it with topological persistent cohomology to obtain stable subgroups and hierarchical sequences. The features between samples include representation vectors and / or intensity-morphological statistics obtained by autoencoders or variational autoencoders, and the groups are expanded from bottom to top according to similarity thresholds and scale factors.

[0010] As a preferred embodiment of the deep learning-based cardiac CTA image cluster registration template construction method of the present invention, the multi-resolution 3D registration network includes a hierarchical window self-attention structure for input with both moving and fixed image channels, and a 3D convolutional upsampling decoding network with three-scale displacement branches. The key hyperparameters are:

[0011] The embedding dimension is 96, the layer depth is 2,2,4,2, the number of multi-heads is 4,4,8,8, the window size is 5,6,5, the drop path rate is 0.3, and the patch size includes a dual-path configuration of 2 and 4.

[0012] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method described in this invention, the following is provided: In the medium / high resolution stage, an adaptive weighting based on a normalized intensity difference map is introduced: The absolute value of the intensity difference between the coarsely registered moving image and the fixed image is taken and normalized to [0,1] according to the sample, and linearly mapped to a weight map of [0.5,1.0]. The two input channels are weighted voxel by voxel to highlight the myocardial-blood pool and coronary artery boundary regions.

[0013] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the spatial transformer includes trilinear interpolation deformation of three-dimensional volume data using unit coordinate grid and differentiable grid sampling, with a coordinate normalization range of [-1,1] and multi-resolution voxel sizes of 48×48×48, 96×96×96 and 192×192×192.

[0014] As a preferred embodiment of the deep learning-based cardiac CTA image cluster registration template construction method of the present invention, the update strategy of the cardiac CTA image cluster registration template is iterative: the sub-template is initialized with the mean or median image, the samples in the sub-group are registered to the coordinate system of the current sub-template, and the aggregated image is calculated as the new template. The iteration continues until the template change is lower than a preset threshold. Each sub-template is used as the input of the previous level and this process is repeated until the global template is obtained.

[0015] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the loss function of the combined loss consists of a weighted sum of the reconstruction similarity loss of the three scales (fine / medium / coarse) and the displacement field gradient smoothing regularization. The reconstruction similarity is at least one of normalized cross-correlation or mean square error, and the three-scale supervision weights decrease sequentially from fine to coarse.

[0016] As a preferred embodiment of the deep learning-based cardiac CTA image group registration template construction method of the present invention, the symmetric training includes adopting a symmetric training strategy, that is, simultaneously optimizing the registration in both x→y and y→x directions; and using a progressive refinement method of upsampling and residual fusion for the displacement field in the multi-resolution path.

[0017] In summary, the deep learning-based method for constructing cardiac CTA image cluster registration templates proposed in this invention offers significant advantages over traditional methods. First, by introducing a subgrouping mechanism based on sample feature similarity and topological persistent coherence, a bottom-up hierarchical organization of heterogeneous cardiac CTA data is achieved, effectively alleviating the difficulty of cluster registration optimization caused by excessive individual differences and improving registration stability and convergence efficiency. Second, a multi-resolution 3D registration network combined with a coarse-to-fine non-rigid registration strategy is employed. A hierarchical window self-attention mechanism is introduced at the encoding end to enhance the modeling ability for large-scale deformations. At the decoding end, a three-scale displacement branch is set, and a progressive refinement method combining upsampling and residual fusion is used to achieve a smooth transition from global alignment to precise matching of local details, significantly improving the spatial clarity and anatomical structural consistency of the template. Furthermore, an adaptive weighting mechanism based on intensity difference maps is introduced at medium to high resolution to highlight the registration accuracy of the myocardium-blood pool and coronary artery boundary regions, enhancing the alignment effect of key structures. Furthermore, an iterative registration-aggregation-update template generation strategy is adopted, with sub-templates gradually converging into a global template, balancing local representativeness and overall consistency. A combined loss function, integrating multi-scale reconstruction similarity loss and displacement field gradient smoothing regularization, and implementing a symmetric training strategy, not only ensures bidirectional consistency in registration but also guarantees the physical rationality and smoothness of the deformation field. The overall solution is implemented within an end-to-end framework, exhibiting good learnability and generalization ability. It effectively solves the problems of template ambiguity and low construction efficiency caused by initial value sensitivity and strong parameter dependence in traditional methods. It provides high-quality, high-fidelity population reference templates for subsequent cardiac structure analysis, disease modeling, and personalized diagnosis and treatment, possessing significant clinical application value and promising prospects for wider application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0019] Figure 1 This is a flowchart illustrating a method for constructing a cardiac CTA image cluster registration template based on deep learning, as provided in one embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of a medium for a method of constructing a cardiac CTA image group registration template based on deep learning, as provided in an embodiment of the present invention.

[0021] Figure 3This is a schematic diagram of a computing device for constructing a cardiac CTA image group registration template based on deep learning, as provided in one embodiment of the present invention. Detailed Implementation

[0022] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0027] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0028] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0029] Example

[0030] Reference Figures 1-3 This is the first embodiment of the present invention, which provides a method for constructing a cardiac CTA image cluster registration template based on deep learning, including:

[0031] S1: Acquire cardiac CTA images for template construction, forming a set of fixed and moving images.

[0032] Among them, cardiac CTA includes ECG-gated arterial phase scanning, injection of non-ionic contrast agent via peripheral vein, acquisition completed by intelligent triggering, and flushing with physiological saline after acquisition.

[0033] S2: Preprocessing of cardiac CTA images, including intensity normalization and spatial normalization, to obtain uniform voxel size and cardiac regions of interest.

[0034] The preprocessing included windowing and Z-score normalization of CT values; voxel resampling to an isotropic resolution of 1.0–1.25 mm; and cropping of the cardiac region of interest within the thoracic cavity and standardizing it to a size of 192×192×192 voxels.

[0035] S3: Subgrouping is performed based on sample feature similarity and topological persistent cohomology to form a bottom-up hierarchical subgroup sequence.

[0036] The subgroups are divided by calculating the feature similarity map between samples and combining it with topological persistent cohomology to obtain stable subgroups and hierarchical sequences. The features between samples include the representation vectors and / or intensity-morphological statistics obtained by autoencoders or variational autoencoders, and the subgroups are expanded from bottom to top according to the similarity threshold and scale factor.

[0037] S4: A multi-resolution 3D registration network is used at each level to perform coarse-to-fine non-rigid registration of the moving / fixed map, outputting a multi-scale displacement field and completing differentiable deformation through a spatial transformer.

[0038] The multi-resolution 3D registration network includes a dual-channel input consisting of a moving or fixed image. The encoder employs a hierarchical window self-attention structure, while the decoder uses 3D convolutional upsampling with a three-scale translation branch. Key hyperparameters are:

[0039] The embedding dimension is 96, the layer depth is 2,2,4,2, the number of multi-heads is 4,4,8,8, the window size is 5,6,5, the drop path rate is 0.3, and the patch size includes a dual-path configuration of 2 and 4.

[0040] Furthermore, in the medium / high resolution stage, an adaptive weighting based on a normalized intensity difference map is introduced: the absolute value of the intensity difference between the coarsely registered moving map and the fixed map is taken and normalized to [0,1] according to the sample, and linearly mapped to a weight map of [0.5,1.0]. The two input channels are weighted voxel by voxel to highlight the myocardial-blood pool and coronary artery boundary regions.

[0041] The spatial transformer includes trilinear interpolation deformation of three-dimensional volume data using unit coordinate grid and differentiable grid sampling, with coordinate normalization range of [-1,1] and multi-resolution voxel sizes of 48×48×48, 96×96×96 and 192×192×192.

[0042] S5: Within a subgroup, an iterative strategy of registration-aggregation-update is used to generate sub-templates, which are then aggregated from lower-level sub-templates to the global template.

[0043] The update strategy for the cardiac CTA image cluster registration template is iterative: the sub-template is initialized with the mean or median image, the samples in the sub-cluster are registered to the coordinate system of the current sub-template, and the aggregated image is calculated as the new template. The process is iterated until the template change is lower than the preset threshold. Each sub-template is used as the input of the next level and this process is repeated until the global template is obtained.

[0044] S6: Symmetric training and optimization are performed using a combination loss of multi-scale reconstruction similarity and displacement field smoothing regularization within an end-to-end framework to evaluate template quality and deformation physics.

[0045] The loss function of the combined loss consists of a weighted sum of the reconstruction similarity loss at the fine, medium, and coarse scales and the gradient smoothing regularization of the displacement field. The reconstruction similarity is at least one of normalized cross-correlation or mean square error, and the three-scale supervision weights decrease sequentially from fine to coarse.

[0046] Symmetric training includes adopting a symmetric training strategy, which simultaneously optimizes the registration in both the x→y and y→x directions; and employing a progressive refinement method of upsampling and residual fusion for the displacement field in multi-resolution paths.

[0047] The proposed method for constructing a cardiac CTA image group registration template based on deep learning achieves the following performance metrics on the test set.

[0048] Therefore, the present invention adopts the above-mentioned objective of providing a method for constructing cardiac CTA image group registration templates based on deep learning, thus forming a set of methods for constructing cardiac CTA image group registration templates based on deep learning.

[0049] After introducing the method of exemplary embodiments of the present invention, the computer-readable storage medium of exemplary embodiments of the present invention will be described next. Please refer to [link / reference]. Figure 2 The computer-readable storage medium shown is an optical disc 30, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it implements the steps described in the above method implementation, such as: acquiring cardiac CTA images for template construction to form a set of fixed and moving images; preprocessing the cardiac CTA images, including intensity normalization and spatial normalization, to obtain uniform voxel sizes and cardiac regions of interest; dividing the images into subgroups based on sample feature similarity and topological persistent coherence to form a bottom-up hierarchical subgroup sequence; using a multi-resolution 3D registration network within each level to perform coarse-to-fine non-rigid registration of the moving / fixed images, outputting a multi-scale displacement field and completing differentiable deformation through a spatial transformer; generating sub-templates within the subgroups using an iterative strategy of registration-aggregation-update, and merging the lower-level sub-templates into a global template step by step; and performing symmetric training and optimization in an end-to-end framework using a combination loss of multi-scale reconstruction similarity and displacement field smoothing regularization to evaluate the template quality and deformation physicality. The specific implementation methods of each step will not be repeated here.

[0050] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0051] After introducing the methods and media of exemplary embodiments of the present invention, the following references are made. Figure 3 A computational device for adaptive recovery of low-voltage power grid self-healing control according to an exemplary embodiment of the present invention.

[0052] Figure 3 A block diagram is shown of an exemplary computing device 40 suitable for implementing embodiments of the present invention. The computing device 40 may be a computer system or a server. Figure 3 The computing device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0053] like Figure 3 As shown, the components of computing device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0054] The computing device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computing device 40, including volatile and non-volatile media, and removable and non-removable media.

[0055] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 4021 and / or cache memory 4022. Computing device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, ROM 4023 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 3 (Not shown in the image, usually referred to as "hard drive"). Although not shown in... Figure 3 The diagram illustrates that disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) can be provided. In these cases, each drive can be connected to bus 403 via one or more data media interfaces. System memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0056] A program / utility 4025 having a set (at least one) of program modules 4024 may be stored, for example, in system memory 402, and such program modules 4024 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment. Program modules 4024 typically perform the functions and / or methods described in the embodiments of the present invention.

[0057] The computing device 40 can also communicate with one or more external devices 404 (such as a keyboard, pointing device, display, etc.). This communication can be performed via the input / output (I / O) interface 405. Furthermore, the computing device 40 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 406. Figure 3 As shown, network adapter 406 communicates with other modules of computing device 40 (such as processing unit 401) via bus 403. It should be understood that, although... Figure 3 As not shown, it can be used in conjunction with computing device 40 with other hardware and / or software modules.

[0058] The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402. For example, it acquires cardiac CTA images for template construction, forming a set of fixed and moving images; preprocesses the cardiac CTA images, including intensity normalization and spatial normalization, to obtain uniform voxel sizes and cardiac regions of interest; divides the images into subgroups based on sample feature similarity and topological persistent coherence, forming a bottom-up hierarchical subgroup sequence; uses a multi-resolution 3D registration network to perform coarse-to-fine non-rigid registration of moving / fixed images within each level, outputs multi-scale displacement fields, and completes differentiable deformation through a spatial transformer; generates sub-templates within the subgroups using an iterative strategy of registration-aggregation-update, and aggregates from lower-level sub-templates to the global template; and performs symmetric training and optimization using a combination loss of multi-scale reconstruction similarity and displacement field smoothing regularization within an end-to-end framework to evaluate template quality and deformation physicality.

[0059] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0060] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0061] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0062] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0063] If the functionality is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions 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 within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0065] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a registration template for cardiac CTA image clusters based on deep learning, characterized in that, Includes the following steps: Acquire cardiac CTA images for template construction, forming a set of fixed and moving images; The cardiac CTA images are preprocessed, including intensity normalization and spatial normalization, to obtain uniform voxel sizes and cardiac regions of interest. Subgrouping is performed based on sample feature similarity and topological persistent cohomology to form a bottom-up hierarchical subgroup sequence; A multi-resolution 3D registration network is used to perform coarse-to-fine non-rigid registration of moving / fixed maps at each level, outputting a multi-scale displacement field and completing differentiable deformation through a spatial transformer. Within a subgroup, an iterative strategy of registration-aggregation-update is used to generate sub-templates, which are then aggregated from lower-level sub-templates to the global template. Symmetric training and optimization are performed using a combination loss of multi-scale reconstruction similarity and displacement field smoothing regularization within an end-to-end framework to evaluate template quality and deformation physics.

2. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The cardiac CTA includes ECG-gated arterial phase scanning, injection of a non-ionic contrast agent via a peripheral vein, acquisition via intelligent triggering, and flushing with saline solution after acquisition.

3. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The preprocessing includes windowing and Z-score normalization of CT values; voxel resampling to an isotropic resolution of 1.0–1.25 mm; and cropping the cardiac region of interest within the thoracic cavity and standardizing it to a size of 192×192×192 voxels.

4. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The subgroups are divided by calculating the feature similarity map between samples and combining it with topological persistent cohomology to obtain stable subgroups and hierarchical sequences. The features between samples include representation vectors and / or intensity-morphological statistics obtained by autoencoders or variational autoencoders, and the subgroups are expanded from bottom to top according to similarity thresholds and scale factors.

5. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The multi-resolution 3D registration network includes inputs with either a moving or fixed image in dual channels. The encoder employs a hierarchical window self-attention structure, and the decoder uses 3D convolutional upsampling with three-scale translation branches. The key hyperparameters are: The embedding dimension is 96, the layer depth is 2,2,4,2, the number of multi-heads is 4,4,8,8, the window size is 5,6,5, the drop path rate is 0.3, and the patch size includes a dual-path configuration of 2 and 4.

6. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, In the medium / high resolution stage, an adaptive weighting based on a normalized intensity difference map is introduced: the absolute value of the intensity difference between the coarsely registered moving map and the fixed map is taken and normalized to [0,1] according to the sample, and linearly mapped to a weight map of [0.5,1.0]. The two input channels are weighted voxel by voxel to highlight the myocardial-blood pool and coronary artery boundary regions.

7. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The spatial transformer includes trilinear interpolation deformation of three-dimensional volume data using unit coordinate grid and differentiable grid sampling, with coordinate normalization range of [-1,1] and multi-resolution voxel sizes of 48×48×48, 96×96×96 and 192×192×192.

8. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The update strategy for the cardiac CTA image group registration template is iterative: the sub-template is initialized with the mean or median image, the samples in the subgroup are registered to the coordinate system of the current sub-template, and the aggregated image is calculated as the new template. The process is iterated until the template change is lower than a preset threshold. Each sub-template is used as the input of the next level and this process is repeated until the global template is obtained.

9. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The loss function of the combined loss consists of a weighted sum of the reconstruction similarity loss at the fine, medium, and coarse scales and the gradient smoothing regularization of the displacement field. The reconstruction similarity is at least one of normalized cross-correlation or mean square error, and the three-scale supervision weights decrease sequentially from fine to coarse.

10. The method for constructing a cardiac CTA image cluster registration template based on deep learning according to claim 1, characterized in that, The symmetric training includes adopting a symmetric training strategy, that is, simultaneously optimizing the registration in both x→y and y→x directions; and using a progressive refinement method of upsampling and residual fusion for the displacement field in the multi-resolution path.

Images