Tomographic imaging methods, devices, tomographic scanning equipment and storage media

Abstract

This invention relates to the field of computed tomography (CT) technology, specifically to a CT imaging method, apparatus, CT scanning device, and storage medium. By constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate systems of the high-resolution and large-field-of-view imaging links are aligned in a common reconstruction space, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. By integrating the projection matrix into the process of minimizing the reconstruction residual and combining iterative reconstruction with image structure regularization constraints, global geometric consistency and local detail fidelity are balanced without the need for external high-resolution labels. This is beneficial for improving overall image resolution and imaging quality while maintaining large-field-of-view imaging capabilities.

Description

Technical Field

[0001] This invention relates to the field of computed tomography (CT) technology, and in particular to a CT imaging method, apparatus, CT equipment, and storage medium. Background Technology

[0002] Computed tomography (CT) technology, especially cone-beam computed tomography (CBCT), has been widely used in human clinical practice, veterinary medicine, and pathology. Currently, the mainstream structures of CBCT equipment are C-type and O-type. However, existing CBCT equipment has significant limitations in imaging performance: while C-type equipment offers a large field of view, its spatial resolution is low due to the use of large-focus X-ray tubes and large-pixel detectors; while O-type equipment achieves excellent detail imaging capabilities through small-focus X-ray tubes and high-resolution detectors, it is limited by a narrow field of view, making it difficult to meet the needs of large-area scanning. Therefore, current technology struggles to balance high spatial resolution and a large imaging range, restricting its comprehensive application in various scenarios. Summary of the Invention

[0003] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a tomographic imaging method, apparatus, tomographic scanning device, and storage medium, which can improve the overall image resolution and imaging quality while maintaining a large field of view imaging capability.

[0004] In a first aspect, embodiments of the present invention provide a computed tomography imaging method, comprising:

[0005] Obtain a locally high-resolution first projection dataset and a globally low-resolution second projection dataset;

[0006] Load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space.

[0007] Based on the first projection matrix and the second projection matrix, the reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset is minimized and combined with the regularization constraint of the image structure to obtain a three-dimensional reconstructed image through iterative reconstruction.

[0008] Secondly, embodiments of the present invention provide a computed tomography imaging method, comprising:

[0009] Obtain the first projection dataset with local high resolution and the second projection dataset from the overall low resolution;

[0010] A three-dimensional reconstructed image is obtained by jointly reconstructing the first projection dataset and the second projection dataset.

[0011] The 3D reconstructed image is input into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels;

[0012] The first projection dataset is input into the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level;

[0013] Based on the attention mechanism, the global features and the local features are spatially aligned and weighted and fused at the corresponding resolution level to generate a 3D enhanced image.

[0014] Thirdly, embodiments of the present invention provide a tomographic imaging apparatus, comprising:

[0015] The first acquisition module is used to acquire a local high-resolution first projection dataset and a global low-resolution second projection dataset.

[0016] The first loading module is used to load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space.

[0017] The first reconstruction module is used to minimize the weighted reconstruction residual between the target voxel field and the first and second projection datasets based on the first and second projection matrices, and combine it with the regularization constraints of the image structure to obtain a three-dimensional reconstructed image through iterative reconstruction.

[0018] Fourthly, embodiments of the present invention provide a CT imaging device, applied to a CT equipment equipped with a high-resolution imaging link and a large field-of-view imaging link, comprising:

[0019] The second acquisition module is used to acquire a first projection dataset from the high-resolution imaging link and a second projection dataset from the wide-field imaging link.

[0020] The second reconstruction module is used to perform joint reconstruction based on the first projection dataset and the second projection dataset to obtain a three-dimensional reconstructed image.

[0021] The first extraction module is used to input the three-dimensional reconstructed image into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels.

[0022] The second extraction module is used to input the first projection dataset into the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level.

[0023] The enhancement fusion module is used to spatially align and weightedly fuse the global features and the local features at the corresponding resolution level based on the spatial attention mechanism to generate a 3D enhanced image.

[0024] Fifthly, embodiments of the present invention provide a tomographic scanning device, including a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program to implement the above-described tomographic imaging method.

[0025] Sixthly, embodiments of the present invention provide a storage medium storing a computer program that, when run, implements the above-described computed tomography imaging method.

[0026] The embodiments of the present invention have at least the following beneficial effects:

[0027] Firstly, by constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate system alignment of the high-resolution and large-field-of-view imaging links in the common reconstruction space is achieved, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. The projection matrix is ​​integrated into the reconstruction residual minimization process, and iterative reconstruction is carried out in combination with image structure regularization constraints. Without the need for external high-resolution labels, global geometric consistency and local detail fidelity are taken into account, which is conducive to improving the overall image resolution and imaging quality while maintaining the large-field-of-view imaging capability.

[0028] Secondly, joint reconstruction is performed based on the locally high-resolution first projection dataset and the globally low-resolution second projection dataset to obtain a globally consistent 3D reconstructed image with preserved local details. The 3D reconstructed image and the first projection dataset are then input into the global feature extraction branch and local feature extraction branch of the pre-trained fusion enhancement model, respectively, to extract global and local features at different resolution levels. Through an attention mechanism, spatial alignment and weighted fusion of global and local features are achieved at the corresponding resolution levels to generate a 3D enhanced image. This achieves a deep integration of global structure and local high-resolution details, injecting local high-resolution details into the global image while maintaining the consistency of the overall scene structure. This is beneficial for improving the overall image resolution and imaging quality while maintaining the large field-of-view imaging capability.

[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0031] Figure 1 This is a schematic diagram of the structure of a tomography scanning device according to an embodiment of the present invention;

[0032] Figure 2 This is one of the flowcharts of the tomographic imaging method according to an embodiment of the present invention;

[0033] Figure 3 This is the second flowchart of the tomographic imaging method according to an embodiment of the present invention;

[0034] Figure 4 This is the third step of the tomographic imaging method according to an embodiment of the present invention;

[0035] Figure 5 This is an architecture diagram of the fusion enhancement model of the tomographic imaging method according to an embodiment of the present invention;

[0036] Figure 6 These are schematic block diagrams illustrating two different examples of a computed tomography imaging device according to embodiments of the present invention.

[0037] Figure 7 This is the fourth step of the tomographic imaging method according to an embodiment of the present invention;

[0038] Figure 8 This is a second principle block diagram of the tomographic imaging device according to an embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the tomographic scanning device and storage medium according to an embodiment of the present invention.

[0040] Figure label:

[0041] Rack 10, turntable 11, first transmitter 21, first detector 22, second transmitter 31, second detector 32, first acquisition module 110a, first loading module 120a, second loading module 130b, first reconstruction module 130a, second acquisition module 210, second reconstruction module 220, first extraction module 230, second extraction module 240, enhancement fusion module 250, processor 310, memory 320, storage medium 410, computer program 420. Detailed Implementation

[0042] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0043] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and 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 limiting this invention.

[0044] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," "exceeding," etc. are understood to exclude the stated number, and "above," "below," "within," etc. are understood to include the stated number. If "first," "second," etc. are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of the indicated technical features.

[0045] In the description of this invention, unless otherwise explicitly defined, terms such as "set", "install", and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0046] Tomographic scanning is an imaging technique that generates two-dimensional or three-dimensional views by acquiring cross-sectional images of the interior of an object. The main types include: computed tomography (CT), magnetic resonance imaging (MRI), optical coherence tomography (OCT), positron emission tomography (PET), and thermal tomography (TTM). Computed tomography uses an X-ray beam to perform a tomographic scan of the target object and then uses a computer to process the data to generate detailed images of the body's internal structures. Cone-beam computed tomography (CBCT) has been widely used in human clinical practice, veterinary medicine, and pathology due to its significant advantages such as high spatial resolution, low radiation dose, rapid scanning, and three-dimensional imaging. The application requirements are mostly focused on imaging range and spatial resolution.

[0047] Currently, mainstream CBCT equipment is mainly divided into C-type and O-type structures. C-type equipment typically features a high-power X-ray tube and a wide-field detector. This configuration, due to its larger focal spot and relatively larger detector pixel size, sacrifices spatial resolution while providing a wider imaging range. Conversely, O-type equipment combines a low-power, small-focal-point X-ray tube with a high-resolution detector. While this design can achieve imaging of fine structures, its field of view is relatively narrow.

[0048] Therefore, the main challenge facing current CBCT technology lies in balancing the trade-off between field of view size and spatial resolution. On the one hand, to obtain a large imaging range (over 200 mm), a larger detector and a longer source-to-image distance (SID) are required. With increasing imaging distance, the X-ray tube needs greater power to penetrate the scanned object, often resulting in a larger focal spot size and lower spatial resolution, which cannot meet the requirements for imaging microstructures. On the other hand, to achieve higher spatial resolution (greater than 10 line pairs / cm), existing technologies tend to use microfocal tubes and high-resolution detectors, achieving sub-millimeter spatial resolution. However, this approach is limited by the smaller SID and detector size, resulting in a limited imaging range that cannot meet the needs of applications requiring a wide imaging range.

[0049] To simultaneously meet the demands of large-area imaging and high resolution, it is typically necessary to prepare multiple tomographic scanning devices of different specifications. These include large-area CBCT devices for image-guided imaging and target localization, and high-resolution tomographic scanning devices for microscopic lesion analysis and diagnosis. However, the generally large size and footprint of CBCT devices, coupled with the need for operation in specialized shielded rooms, not only increase the overall cost of the equipment but also add to the financial burden on users. Furthermore, the transfer of the scanned object between devices and the repeated localization processes result in significant positioning errors, long operation times, and low model reconstruction efficiency, making it difficult to meet the ever-increasing demands for image quality and scanning modeling efficiency.

[0050] Please refer to Figure 1This invention provides a tomographic scanning device, including a gantry 10, a high-resolution imaging link, a wide-field-of-view imaging link, and a controller. The gantry 10 is equipped with a rotating turntable 11 and a drive assembly connected to the turntable 11. The high-resolution imaging link is mounted on the turntable 11 and is used to acquire a first projection dataset. The wide-field-of-view imaging link is also mounted on the turntable 11 and is used to acquire a second projection dataset. The controller is communicatively connected to both the high-resolution imaging link and the wide-field-of-view imaging link. The controller receives the first and second projection datasets and, based on a preloaded first projection matrix, a first projection noise weighting matrix, a second projection matrix, and a second projection noise weighting matrix, minimizes the weighted reconstruction residual between the target voxel field and the first and second projection datasets. Combining this with regularization constraints of the image structure, iterative reconstruction yields a three-dimensional reconstructed image. The high-resolution imaging link includes a first transmitter 21 and a first detector 22, both mounted on the turntable 11. The wide-field-of-view imaging link includes a second transmitter 31 and a second detector 32, both mounted on the turntable 11. Both the first transmitter 21 and the second transmitter 31 are X-ray transmitters, but the high-resolution imaging link and the wide-field-of-view imaging link differ significantly in key performance parameters, such as the output power and focal size of the X-ray transmitters, and the spatial resolution, pixel size, and effective imaging area of ​​the detectors. Based on these structural and performance differences, the high-resolution imaging link can acquire high-resolution projection image data with sub-millimeter level detail resolution for local regions of interest of the target object (i.e., the object to be detected); while the wide-field-of-view imaging link can cover the entire anatomical range of the target object, acquiring projection image data with a wider field of view but relatively lower spatial resolution.

[0051] During equipment operation, the controller controls the rotation of turntable 11 via the drive component. Since both the high-resolution imaging link and the wide-field-of-view imaging link are mounted on turntable 11, synchronous rotation of the two links is possible. This allows for the simultaneous acquisition of two independent but spatially registered sets of projection data around the same target object during a single scan. Specifically, a first projection dataset and a second projection dataset are acquired from the same target object during the synchronous rotation of the high-resolution and wide-field-of-view imaging links. The first projection dataset is acquired by the high-resolution imaging link, and the second projection dataset is acquired by the wide-field-of-view imaging link. This simultaneous acquisition of high-resolution and wide-field-of-view data avoids the positioning errors and time losses caused by multiple scans and repetitive positioning in traditional methods, and also improves imaging efficiency and data consistency. Specifically, the tomographic imaging method executed by the controller is described below.

[0052] Please refer to Figure 2The tomographic imaging method disclosed in this embodiment is applied to a tomographic scanning device equipped with a high-resolution imaging link and a wide-field-of-view imaging link, and includes steps S110a to S130a. It should be noted that the numbering of the steps in this embodiment is only for ease of review and understanding, and does not limit the execution order of the steps. The details of each step are described below:

[0053] S110a, Obtain the local high-resolution first projection dataset and the overall low-resolution second projection dataset;

[0054] For example, the first projection dataset is obtained by scanning a local structure of the target object, and the first projection dataset has a high imaging resolution; therefore, the first projection dataset is a "local" and "high-resolution" dataset. Compared with the first projection dataset, the second projection dataset has a larger imaging range, so it can image the "overall" structure of the target object. However, the second projection dataset has a lower imaging resolution; therefore, the second projection dataset is a "global" and "low-resolution" dataset. It should be understood that in this embodiment, "local" and "global," "high-resolution" and "low-resolution" are relative to the first projection dataset and the second projection dataset.

[0055] In some application examples, step S110a includes: acquiring a first projection dataset of local high resolution from a high-resolution imaging link and a second projection dataset of overall low resolution from a wide-field-of-view imaging link.

[0056] The high-resolution imaging link is used to acquire the first projection dataset, and the wide field-of-view imaging link is used to acquire the second projection dataset. The high-resolution imaging link and the wide field-of-view imaging link can be separate and located in different tomographic scanning devices.

[0057] Alternatively, in some other application examples, step S110a includes: acquiring a first projection dataset of local high resolution from the high-resolution imaging link and a second projection dataset of overall low resolution from the wide field-of-view imaging link, wherein the high-resolution imaging link and the wide field-of-view imaging link are integrated in the same tomographic scanning device.

[0058] In conventional technical solutions, high-resolution projection data and wide-field-of-view projection data typically require two separate tomographic scanning devices to acquire. One device is equipped with a high-resolution imaging component for detailed local imaging, while the other is equipped with a wide-field-of-view imaging component for overall structural coverage. This approach not only requires multiple positioning and repeated scanning of the same target object but also easily introduces inconsistencies in projection data due to factors such as mechanical positioning deviations, time intervals, or physiological movements, thereby affecting the accuracy and reliability of subsequent 3D reconstruction. This embodiment integrates the high-resolution imaging link and the wide-field-of-view imaging link into a single tomographic scanning device. This avoids the positioning errors and time losses caused by multiple scans and repeated positioning in traditional solutions and also improves imaging efficiency and data consistency.

[0059] Alternatively, in some application examples, step S110a includes: acquiring a locally high-resolution first projection dataset from the high-resolution imaging link and a globally low-resolution second projection dataset from the wide-field-of-view imaging link. The high-resolution imaging link and the wide-field-of-view imaging link are integrated into the same tomographic scanning device, and the first and second projection datasets are acquired separately for the same target object during the synchronous rotation of the high-resolution imaging link and the wide-field-of-view imaging link. Thus, high-resolution projection data and wide-field-of-view projection data can be acquired simultaneously for the same target object in a single scan, avoiding repeated positioning of the target object and improving scanning efficiency. Simultaneously, since the two types of projection data are acquired within the same mechanical motion trajectory and time window, their spatial geometric relationships are highly consistent, effectively reducing systematic deviations caused by equipment differences or operational timing. This provides a high-quality, high-registration-accuracy data foundation for subsequent fusion reconstruction, which is beneficial for improving the overall quality of 3D image reconstruction.

[0060] In a specific example, when the region of interest is known, such as when the location of a lesion has been determined through prior information, the high-resolution imaging link and the wide-field-of-view imaging link can be controlled to rotate synchronously around the target object during a single scan, acquiring projection data separately. The high-resolution imaging link is used to perform a local scan of the region of interest to obtain a high spatial resolution projection image; the wide-field-of-view imaging link is used to scan the entire anatomical region of the target object to obtain a projection image with a larger coverage area, thus obtaining a first projection dataset and a second projection dataset. It should be noted that both the first and second projection datasets contain multiple projection images arranged in order of acquisition angle, used for subsequent 3D image reconstruction.

[0061] In another specific example, when the region of interest (ROI) is unknown, a wide-field-of-view imaging link can be used to quickly scan the target object to obtain overall anatomical information, and the location of the ROI can be determined based on this information. Subsequently, a precise scan is performed without changing the target object's location. This precise scan can be achieved by controlling the high-resolution imaging link and the wide-field-of-view imaging link to scan simultaneously, where the high-resolution imaging link acquires high-resolution data of the identified ROI, while the wide-field-of-view imaging link continues to acquire data of the entire region; or, the scanning field of view of the high-resolution imaging link can be fine-tuned to cover the identified ROI, allowing for high-resolution scanning of the ROI independently via the high-resolution imaging link.

[0062] S120a, Load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space.

[0063] For example, during the fusion of projection data, due to significant differences in imaging geometry and noise characteristics between large-scale low-resolution projection data and small-scale high-resolution projection data, direct feature fusion can easily lead to feature mismatch during the fusion process. Specifically, high-resolution projection data has a stronger spatial frequency response capability and can reflect local detail information; while low-resolution projection data covers a more complete anatomical structure and provides global spatial context. Maintaining the consistency of the overall structure while preserving high-resolution details during image reconstruction constitutes a key technical challenge. This embodiment constructs a first projection matrix and a second projection matrix based on the geometric calibration structure of the same calibration phantom, integrating them into the subsequent 3D image reconstruction process. This achieves coordinate system alignment between the high-resolution imaging link and the large-field-of-view imaging link in the common reconstruction space, ensuring the "geometric correspondence" in physical consistency. Specifically, the method for constructing the first projection matrix and the second projection matrix includes:

[0064] Construct a first initial projection matrix and a second initial projection matrix; wherein the initial values ​​of the first initial projection matrix and the second initial projection matrix can be set according to the parameters of the tomographic scanning equipment, such as the source distance, detector offset and rotation angle of both matrices.

[0065] A high-precision calibration phantom containing known spatial markers is used as the target object. The same calibration phantom is scanned through a high-resolution imaging link and a wide-field-of-view imaging link to obtain first calibration projection data and second calibration projection data. The number of spatial markers can be one or more. The three-dimensional absolute coordinates of each spatial marker on the calibration phantom are known. The three-dimensional relative coordinates of the spatial markers during the scanning process (i.e., the three-dimensional coordinates of the spatial markers relative to the coordinate system of the tomographic scanning equipment) can be calculated based on the three-dimensional absolute coordinates. In other words, both the three-dimensional absolute coordinates and the three-dimensional relative coordinates of the calibration phantom are known quantities.

[0066] The first projection coordinates of the spatial marker points in the first calibration projection data and the second projection coordinates in the second calibration projection data are determined. During the scanning process, the position of the calibration phantom remains unchanged. In this way, scanning the same calibration phantom ensures that the three-dimensional coordinates of the spatial marker points of the calibration phantom remain unchanged, and the corresponding two-dimensional projection coordinates of the spatial marker points can be determined by detection in both the first marker projection data and the second calibration projection data.

[0067] A geometric projection model is established based on the spatial coordinates of the spatial markers in the calibration phantom, as well as the first and second projected coordinates. The first projected coordinates are the projected coordinates of the spatial markers in the high-resolution imaging link, and the second projected coordinates are the projected coordinates of the spatial markers in the large field-of-view imaging link. There is a geometric mapping relationship between the spatial coordinates of the spatial markers and the first and second projected coordinates, which allows for the establishment of a geometric projection model to facilitate the analysis of the spatial offset between the coordinate systems of the first and second projected coordinates.

[0068] Error assessment is performed based on the geometric projection model to determine parameter errors, including source-detector distance error, detector offset error, and rotation angle error. Error assessment can be based on historical experience or optimization algorithms. Through error assessment, the errors of the high-resolution imaging link and the wide-field imaging link on key parameters can be determined.

[0069] The first and second initial projection matrices are obtained by compensating for parameter errors using the first and second initial projection matrices. Specifically, error compensation based on parameter errors, resulting in the first and second projection matrices, corrects the deviation between the projection coordinate systems of the two imaging links, achieving precise spatial alignment and ensuring the "geometric correspondence" in physical consistency. In subsequent 3D image reconstruction, the first and second projection matrices are used to fuse the geometric structure information of the imaging links into the image, achieving physical consistency constraints.

[0070] S130a: Based on the first projection matrix and the second projection matrix, the reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset is minimized and combined with the regularization constraint of the image structure to obtain a three-dimensional reconstructed image through iterative reconstruction.

[0071] For example, a voxel field is a data representation method used to represent physical or geometric properties in three-dimensional space. A voxel field divides a continuous three-dimensional space into regularly arranged volumetric units (called voxels). Each voxel stores one or more attribute values ​​(such as X-ray attenuation coefficient, density, semantic labels, etc.), thereby describing the internal structural information of a three-dimensional object or scene in a discretized manner. A voxel can be viewed as an extension of a two-dimensional pixel in three-dimensional space and is typically modeled as a regular cubic unit. The spatial resolution of a voxel field is determined by the voxel size: the smaller the voxel, the finer the geometric details that can be expressed, and the higher the spatial resolution of the reconstructed image.

[0072] In this embodiment, the target voxel field is used to characterize the discretized representation of the 3D image to be reconstructed in a common reconstruction space. The 3D image reconstruction process can be formalized as an optimization problem, namely, finding a target voxel field that minimizes the residual between the simulated projection data generated by the forward projection of the target voxel field through the imaging model and the actually acquired projection data (such as the first projection dataset and the second projection dataset). In related technologies, low-resolution projection data is usually used for reconstruction in order to obtain a 3D reconstructed image with high spatial resolution. However, this embodiment fuses the high-resolution projection data (first projection dataset) and the large field-of-view projection data (second projection dataset) to reconstruct the global features and local details of the 3D image.

[0073] To achieve effective fusion of data acquired by the high-resolution imaging link and the wide-field-of-view imaging link, this embodiment introduces a first projection matrix and a second projection matrix, which correspond to the imaging geometry of the high-resolution imaging link and the wide-field-of-view imaging link, respectively. This ensures the physical consistency of the two projection data (the first projection dataset and the second projection dataset) in the imaging process and applies high-resolution sampling constraints to the local region of interest to preserve fine structural information.

[0074] Furthermore, image structure-related regularization constraints are introduced during the reconstruction optimization process to incorporate prior knowledge and suppress noise or artifacts. For example, a Total Variation (TV) regularization term can be used, leveraging the prior knowledge of image gradient sparsity to effectively preserve edge structures while suppressing noise in smooth regions. By jointly optimizing data fidelity terms (i.e., residuals) and regularization terms (i.e., regularization constraints), high-quality 3D image reconstruction can be achieved while maintaining global anatomical integrity and local high-resolution details.

[0075] Therefore, by constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate system alignment of the high-resolution and large-field-of-view imaging links in the common reconstruction space is achieved, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. By incorporating the projection matrix into the reconstruction residual minimization process and combining iterative reconstruction with image structure regularization constraints, global geometric consistency and local detail fidelity can be balanced without the need for external high-resolution labels. This is beneficial for improving the overall image resolution and imaging quality while maintaining the large-field-of-view imaging capability.

[0076] Please refer to Figure 3 To further improve the resolution and imaging quality of the reconstructed image, this embodiment provides a further improved method, namely, a tomographic imaging method including:

[0077] S110b: Obtain the first projection dataset with local high resolution and the second projection dataset with overall low resolution;

[0078] S120b: Load the first projection matrix corresponding to the high-resolution imaging link and the second projection matrix corresponding to the large field-of-view imaging link. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration phantom to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space.

[0079] S130b: Load the first projection noise weighting matrix and the second projection noise weighting matrix. Each weight element of the first projection noise weighting matrix is ​​used to characterize the noise confidence of each projection measurement point in the first projection dataset. Each weight element of the second projection noise weighting matrix is ​​used to characterize the noise confidence of each projection measurement in the second projection dataset.

[0080] For example, in a dual-source CT system with a high-resolution imaging link and a wide-field-of-view imaging link, the first projection dataset (acquired by the high-resolution imaging link) and the second projection dataset (acquired by the wide-field-of-view imaging link) differ significantly in noise characteristics. To suppress the impact of noise on the quality of the reconstructed image, this embodiment constructs a projection noise weighting matrix to differentially weight the projection data from different sources: higher weights are assigned to projection data with lower noise levels to enhance their contribution to the reconstruction process; lower weights are assigned to projection data with higher noise levels, thereby weakening the adverse effects of projection data on the reconstruction results and improving the overall signal-to-noise ratio and reliability of the reconstructed image.

[0081] Specifically, a first projection noise weighting matrix is ​​constructed for the first projection dataset acquired by the high-resolution imaging link; a second projection noise weighting matrix is ​​constructed for the second projection dataset acquired by the large-field-of-view imaging link. Both the first and second projection noise weighting matrices can be generated based on the Poisson-Gaussian mixed noise model. This model can more accurately characterize the Poisson noise caused by photon statistical fluctuations and the Gaussian noise introduced by the detector electronics system in the X-ray projection data, thus providing a physical basis and numerical foundation for the calculation of the weighting matrix.

[0082] S140b: Based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix, the weighted reconstruction residual between the target voxel field and the first and second projection datasets is minimized, and combined with the regularization constraints of the image structure, a three-dimensional reconstructed image is obtained through iterative reconstruction. Steps S110b, S120b, and S140b correspond to steps S110a, S120a, and S130a as described above, respectively. Detailed explanations and corresponding beneficial effects are provided above and will not be repeated here.

[0083] Therefore, by constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate system alignment of the high-resolution and large-field-of-view imaging links in the common reconstruction space is achieved, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. Moreover, the noise reliability of each projection measurement point is quantified by the first and second projection noise weighting matrices, and the projection matrix and noise weighting matrix are integrated into the weighted reconstruction residual minimization process. Combined with image structure regularization constraints, iterative reconstruction is carried out. Without the need for external high-resolution labels, global geometric consistency and local detail fidelity are taken into account, which is conducive to improving the overall image resolution and imaging quality while maintaining the large-field-of-view imaging capability.

[0084] In step S140b, based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix, the weighted reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset is minimized, including:

[0085] Based on the first projection matrix and the first projection noise weighting matrix, the first weighted reconstruction residual between the simulated projection data generated by the target voxel field and the first projection dataset is determined.

[0086] Based on the second projection matrix and the second projection noise weighting matrix, the second weighted reconstruction residual between the simulated projection data generated by the target voxel field and the second projection dataset is determined.

[0087] Minimize the sum of the square norms of the first weighted reconstruction residual and the second weighted reconstruction residual.

[0088] For example, 3D image reconstruction is performed based on a high-resolution first projection dataset and a large-field-of-view second projection dataset. As mentioned above, the reconstruction process can be formalized as an optimization problem, the goal of which is to solve for a target voxel field such that the residual between the simulated projection data generated by the imaging model from the target voxel field and the actually acquired projection data is minimized.

[0089] Specifically, for the first projection dataset acquired by the high-resolution imaging link, the first weighted reconstruction residual is defined as: the difference between the simulated data obtained by projecting the target voxel field forward through the first projection matrix and the first projection dataset, and then the result after being weighted by the first projection noise weighting matrix. The first projection matrix is ​​used to model the ray path and geometric relationship of the high-resolution imaging link, ensuring the consistency of the physical imaging process; the first projection noise weighting matrix assigns different weights to each projection channel according to the noise statistical characteristics of the high-resolution imaging link, in order to suppress the influence of high-noise data and enhance the contribution of low-noise data.

[0090] Similarly, for the second projection dataset acquired by the large field-of-view imaging link, the second weighted reconstruction residual is defined as: the difference between the simulated data obtained by projecting the same target voxel field forward through the second projection matrix and the second projection dataset, and then the result after being weighted by the second projection noise weighting matrix. The second projection matrix is ​​used to describe the imaging geometry of the large field-of-view imaging link, while the second projection noise weighting matrix reflects the noise distribution characteristics of the link, thereby achieving a reasonable balance of noise in the reconstruction.

[0091] 3D image reconstruction is achieved by jointly minimizing the sum of the square norms of the first weighted reconstruction residual and the second weighted reconstruction residual. This not only ensures the uniformity of the two heterogeneous projection data in the physical model, but also effectively integrates high-resolution detail information with the large field-of-view structural context, thereby simultaneously improving the spatial resolution and overall anatomical consistency of the image under a single reconstruction framework.

[0092] The above steps, based on the first projection matrix and the first projection noise weighting matrix, determine the first reconstruction residual between the simulated projection data generated from the target voxel field and the first projection dataset, including:

[0093] The target voxel field is projected forward based on the first projection matrix to generate the first simulated projection data;

[0094] The residuals between the first simulated projection data and the first projection dataset are weighted based on the first projection noise weighting matrix to determine the first weighted reconstruction residual.

[0095] For example, let the target voxel field be x, the first projection matrix be A1, the first projection noise weighting matrix be W1, and the projected data of the first projection dataset be y1. The first projection matrix A1 represents the mapping relationship from voxel space to projection data space. A simulated first projection data A1x can be generated by forward projection of the target voxel field x using the first projection matrix A1. The first projection noise weighting matrix W1 is used to assign different weights to different measurements to reflect the reliability of each measurement or the corresponding noise level. For example, high-noise data is given a lower weight, and low-noise data is given a higher weight. Therefore, the first weighted reconstruction residual can be expressed as W1(A1x). y1), here A1x y1 is used to represent the residual (i.e., the unweighted reconstruction residual) between the simulated projection data generated by the target voxel field and the first actual observed projection data. Then, the residual is weighted by the first projection noise weighting matrix W1 so that the reliability and noise level of each measurement can be properly considered in the subsequent optimization process, which helps to improve the quality of the final reconstruction result, especially in the presence of non-uniform noise or different measurements with different confidence levels.

[0096] The above steps, based on the second projection matrix and the second projection noise weighting matrix, determine the second weighted reconstruction residual between the simulated projection data generated from the target voxel field and the second projection dataset, including:

[0097] The target voxel field is projected forward based on the second projection matrix to generate second simulated projection data;

[0098] The residuals between the second simulated projection data and the second projection dataset are weighted based on the second projection noise weighting matrix to determine the second weighted reconstruction residual.

[0099] For example, let the target voxel field be x, the second projection matrix be A2, the second projection noise weighting matrix be W2, and the projection data of the second projection dataset be y2. Similar to the principle of the first weighted reconstruction residual, the second simulated projection data can be represented as A2x, and the second weighted reconstruction residual can be represented as W2(A2x). y2).

[0100] Therefore, the optimization problem of 3D image reconstruction can be defined as a joint optimization model. Specifically, step S140 includes: minimizing the weighted reconstruction residuals between the target voxel field and the first and second projection datasets based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix, and combining this with the regularization constraints of the image structure to construct a joint optimization model. The mathematical expression of the joint optimization model is as follows:

[0101]

[0102] Among them, ||W1(A1x y1)||2 2 The square norm used to represent the first weighted reconstruction residual, ||W2(A2x) y2)||2 2 The square norm of the second-weighted reconstruction residual is used to represent the square norm, and λ is used to represent the regularization weight parameter. This is used to represent regularization terms, i.e., applying regularization constraints, such as TV regularization. The joint optimization model can guarantee the physical consistency of the two projected data globally, while taking into account high-resolution sampling constraints in local regions.

[0103] The solution process for the aforementioned joint optimization model is quite complex. It can be decomposed into a data consistency subproblem and a regularization problem using the Alternating Direction Multiplier Method (ADMM). Specifically, in step S140, after constructing the joint optimization model, the 3D reconstructed image is obtained through iterative reconstruction using the Alternating Direction Multiplier Method. The mathematical expression of the joint optimization model after ADMM decomposition is as follows:

[0104]

[0105] Here, z is an intermediate variable in the regularization subproblem, used to decompose the original optimization problem into data consistency and regularization subproblems, facilitating alternating solutions. u is a Lagrange multiplier used to handle constraints in the ADMM framework, ensuring convergence of the solution through iterative updates. k is the iteration number, representing the k-th step in the iterative solution process (e.g., k represents the k-th iteration). ρ is a penalty parameter used to balance the weights of the "data consistency term" and the "intermediate variable constraint term" in the optimization problem, controlling the penalty intensity during the iteration process. Parameters such as λ, k, and ρ can be determined through model training.

[0106] Please refer to Figure 4 In some application examples, computed tomography imaging methods also include:

[0107] S150. Input the 3D reconstructed image into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels.

[0108] S160. Input the first projection dataset into the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level;

[0109] S170. Based on the attention mechanism, global and local features are spatially aligned and weighted and fused at the corresponding resolution level to generate a 3D enhanced image.

[0110] For example, the 3D reconstructed image obtained by jointly reconstructing the first projection dataset and the second projection dataset has fused global structural information and local high-resolution features. However, in detailed regions, it may still be affected by system blurring effects and projection noise, resulting in insufficient representation of some fine structures. Therefore, this embodiment further inputs the 3D reconstructed image into a pre-trained fusion enhancement model to perform super-resolution reconstruction and denoising. The fusion enhancement model includes parallel global feature extraction branches and local feature extraction branches. The global feature extraction branch receives the 3D reconstructed image and extracts multi-scale global contextual features of the 3D reconstructed image to characterize large-scale spatial relationships and overall anatomical structures in the image. The local feature extraction branch receives the first projection dataset and extracts high-frequency detail features to preserve fine structural information from the high-resolution imaging link. The fusion enhancement model performs feature fusion based on an attention mechanism, which can be a spatial attention mechanism, a cross-attention mechanism, a window attention mechanism, or a collaborative attention mechanism, etc. This embodiment uses a spatial attention mechanism, spatially aligning global and local features at the corresponding spatial resolution level, and dynamically assigning fusion weights according to the feature saliency at each location to generate a weighted fused 3D enhanced image. This approach maintains overall scene structural consistency while enhancing the response to local detail areas and suppressing noise propagation in non-critical regions. The resulting 3D enhanced image inherits the integrity of the overall anatomical structure from the joint reconstruction results, while simultaneously being effectively enhanced in local areas by high-resolution detail information from the first projection dataset, and with reduced noise levels. Thus, the obtained 3D enhanced image possesses both global structural consistency and high-resolution local clarity.

[0111] In step S150 above, the 3D reconstructed image is input into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels, including:

[0112] The 3D reconstructed image is divided into blocks and embedded to obtain the first feature block;

[0113] The first feature block is sequentially processed through multiple sets of context aggregation modules to perform global context modeling, and the spatial resolution is gradually reduced through downsampling operations to extract global features at different resolution levels.

[0114] For example, the context aggregation module can use VMamba, Vision Transformer, SwinTransformer, Swin, ResNet residual convolutional blocks, MLP-Mixer, or MobileNet depthwise separable convolutional blocks. Downsampling operations can employ block-based merging downsampling, Conv2d downsampling, pooling downsampling, or stride convolution downsampling. Please refer to [reference needed]. Figure 5The global feature extraction branch includes a first image patch embedding module, a VMamba module, and a first image patch merging module, connected sequentially. The first image patch embedding module is used to embed the input 3D reconstructed image... x joint The image is divided into multiple non-overlapping patches in the spatial dimension, and a linear transformation is performed on each patch to map it into a low-dimensional vector representation, thereby generating an initial sequence of image patches.

[0115] The VMamba module, built upon the Vision Mamba model, is used for sequence modeling of image patch sequences. It possesses long-distance dependency modeling capabilities, effectively representing spatial relationships across regions in an image. It can extract features from 3D reconstructed images, such as global structure, full-scene semantic relationships, and large-scale morphological distributions, for example, the overall contour layout of organs in medical imaging and the global structural relationships of scenes in industrial imaging. In this embodiment, multiple VMamba modules are used, preferably two (VMamba×2), to enhance feature abstraction capabilities.

[0116] The first image block merging module is used to aggregate multiple adjacent image blocks (e.g., 2×2 blocks) into a single merged block in the spatial dimension of the feature map, and to downsample and adjust the channel dimension of the merged features to reduce spatial resolution and increase channel dimension, thereby achieving hierarchical feature extraction and computational efficiency optimization.

[0117] The global feature extraction branch can contain multiple cascaded structures, each including a first image patch merging module and at least one VMamba module. By repeatedly setting up cascaded structures containing multiple sets of VMamba modules and the first image patch merging module, multi-scale global features can be extracted step-by-step at different resolution levels, providing hierarchical contextual information for subsequent feature fusion.

[0118] In step S160 above, the first projection dataset is input to the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level, including:

[0119] The first projection dataset is divided into blocks and embedded to obtain the second feature block;

[0120] The second feature block is sequentially passed through multiple sets of convolutional modules for local feature extraction and then segmented and merged to extract local features at the corresponding resolution level.

[0121] For example, please continue to refer to Figure 5The local feature extraction branch includes a second image patch embedding module, a convolutional neural network module (CNN), and a second image patch merging module. The second image patch embedding module functions the same as the first image patch embedding module in the aforementioned global feature extraction branch, and is used to merge the input local high-resolution sub-images. x HR The projection map of the first projection dataset is divided into multiple non-overlapping image blocks, and each image block is mapped to a low-dimensional vector representation through linear transformation. The second image block merging module has the same function as the first image block merging module. It is used to aggregate multiple adjacent image blocks into a merged block and downsample and adjust the channel dimension of the merged features to reduce the spatial resolution and increase the number of feature channels.

[0122] The convolutional module is built upon a convolutional neural network and possesses a strong ability to perceive local spatial patterns. It is suitable for extracting edge textures, fine structures, and local high-frequency details from high-resolution sub-images with small field of view. The convolutional module can employ ResNet residual convolutional blocks, MobileNet depthwise separable convolutional blocks, DenseNet dense convolutional blocks, or ShuffleNet shuffling convolutional blocks, etc. In this embodiment, multiple convolutional modules are used, preferably two, i.e., CNN×2, to enhance the expressive power of local features.

[0123] Furthermore, the local feature extraction branch can contain multiple cascaded structures, each including a second image patch merging module and at least one convolutional module. By repeatedly setting multiple sets of convolutional modules and second image patch merging modules, multi-scale local features can be extracted step-by-step at different resolution levels, providing hierarchical detailed information for subsequent feature fusion.

[0124] The global feature extraction branch outputs global features to maintain the consistency of the overall anatomical structure of the reconstructed image, preventing global geometric distortion caused by local enhancements. The local feature extraction branch outputs local features, providing high-resolution detail information to supplement fine structures. Specifically, global and local features employ a multi-scale feature fusion strategy combined with a spatial attention mechanism. During the fusion process, the global features output from the VMamba module of the global feature extraction branch and the local features output from the convolutional module of the local feature extraction branch are processed through patch merging in their respective paths to form multiple feature maps at different resolution levels. Subsequently, the attention module aligns and weights the local and global features at their corresponding spatial locations based on learned spatial weights. This not only integrates global context and local detail information at different resolution levels but also enhances the feature responses of key detail regions through adaptive weight allocation.

[0125] The fused multi-scale features are sequentially upsampled and refined through multiple cascaded decoding modules. Each decoding module includes a patch expand module and two convolutional modules (CNN×2) to progressively restore spatial resolution and optimize feature quality. The final projection module projects the image onto the target voxel space, generating a 3D enhanced image. x enhanced While maintaining overall structural consistency, it effectively integrates local details derived from high-resolution projection data, thereby improving both global integrity and local clarity.

[0126] Please refer to Figure 6 Based on the above-described tomographic imaging method, this embodiment also provides a tomographic imaging device, comprising:

[0127] The first acquisition module 110a is used to acquire a local high-resolution first projection dataset and an overall low-resolution second projection dataset.

[0128] The first loading module 120a is used to load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space.

[0129] The first reconstruction module 130a is used to minimize the reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset based on the first projection matrix and the second projection matrix, and combine the regularization constraints of the image structure to obtain a three-dimensional reconstructed image through iterative reconstruction.

[0130] The inventive concept of this tomographic imaging device embodiment is the same as that of the tomographic imaging method embodiment described above. Contents not covered in this tomographic imaging device embodiment can be referred to in the tomographic imaging method embodiment described above, and will not be repeated here. By constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate systems of the high-resolution and large-field-of-view imaging links are aligned in the common reconstruction space, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. Integrating the projection matrix into the reconstruction residual minimization process, combined with image structure regularization constraints for iterative reconstruction, balances global geometric consistency and local detail fidelity without the need for external high-resolution labels, which is beneficial for improving overall image resolution and imaging quality while maintaining large-field-of-view imaging capabilities.

[0131] For further details, please refer to Figure 6 (b) The computed tomography imaging apparatus also includes:

[0132] The second loading module 130b is used to load the first projection noise weighting matrix and the second projection noise weighting matrix. Each weight element of the first projection noise weighting matrix is ​​used to characterize the noise confidence of each projection measurement point in the first projection dataset, and each weight element of the second projection noise weighting matrix is ​​used to characterize the noise confidence of each projection measurement in the second projection dataset.

[0133] The function of the first reconstruction module 130a is also adjusted accordingly. It is used to minimize the weighted reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix, and combined with the regularization constraints of the image structure, to obtain a three-dimensional reconstructed image through iterative reconstruction.

[0134] By constructing a first projection matrix and a second projection matrix based on the same calibration phantom, the coordinate system alignment of the high-resolution and wide-field-of-view imaging links in the common reconstruction space is achieved, providing a geometric consistency basis for the joint reconstruction of dual-source projection data. Moreover, the noise confidence of each projection measurement point is quantified by the first and second projection noise weighting matrices, and the projection matrix and noise weighting matrix are integrated into the weighted reconstruction residual minimization process. Combined with image structure regularization constraints, iterative reconstruction is carried out. Without the need for external high-resolution labels, global geometric consistency and local detail fidelity are taken into account, which is conducive to improving the overall image resolution and imaging quality while maintaining the wide-field-of-view imaging capability.

[0135] Please refer to Figure 7This embodiment provides a computed tomography imaging method, including steps S210-S250. It should be noted that the numbering of the steps in this embodiment is only for ease of review and understanding, and not to limit the execution order of the steps. The details of each step are described below:

[0136] S210. Obtain the local high-resolution first projection dataset and the overall low-resolution second projection dataset;

[0137] S220. Perform joint reconstruction based on the first projection dataset and the second projection dataset to obtain a three-dimensional reconstructed image;

[0138] S230. Input the 3D reconstructed image into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels;

[0139] S240. Input the first projection dataset into the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level;

[0140] S250: Based on the attention mechanism, global and local features are spatially aligned and weighted and fused at the corresponding resolution level to generate a 3D enhanced image.

[0141] For example, in CT image reconstruction applications, the conventional approach typically employs one of the following two methods:

[0142] First, high-resolution 3D image reconstruction based directly on low-resolution or sparsely sampled projection data is limited by insufficient information in the original data, making it difficult to effectively recover fine structures.

[0143] Second, applying noise reduction processing only to high-resolution projection data during reconstruction can improve local clarity, but it often sacrifices the imaging field of view and cannot take into account the integrity of the overall anatomical structure.

[0144] This embodiment utilizes both high-resolution projection data and wide-field projection data for 3D reconstruction, and uses high-resolution projection data to enhance local details and denoise the reconstructed 3D image, thereby obtaining a 3D enhanced image that combines wide-field coverage with high local resolution.

[0145] Specifically, by acquiring a high-resolution small-field-of-view first projection dataset from the high-resolution imaging link and a large-field-of-view low-resolution second projection dataset from the large-field-of-view imaging link, joint reconstruction is then performed based on the first and second projection datasets. The joint reconstruction process can adopt, but is not limited to, the tomographic imaging method described above. That is, the details of steps S210 to S220 can be found in steps S110a to S130a or steps S110b to S140b above. For example, the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix of the corresponding imaging link are introduced respectively to construct a joint optimization model based on the objective function of minimizing the weighted residual. While taking into account physical consistency, the noise characteristics of different data sources are balanced, thereby achieving high-quality panoramic reconstruction through the constraint of physical consistency.

[0146] After reconstruction, the obtained 3D reconstructed image and the first projection dataset are input into a pre-trained fusion enhancement model. This model includes parallel global feature extraction and local feature extraction branches. The global feature extraction branch extracts global features at different resolution levels from the 3D reconstructed image, such as the overall contour. The local feature extraction branch extracts local features corresponding to the global feature resolution level from the high-resolution first projection dataset, such as edges, textures, and fine structures. Then, based on attention mechanisms, such as spatial attention, global and local features are spatially aligned at each corresponding resolution level, and adaptive weighted fusion is performed according to learned spatial weights. This effectively enhances the feature response of key detail areas while suppressing the propagation of noise and unstructured artifacts, thus achieving detail enhancement and denoising of the 3D reconstructed image. Finally, feature fusion and subsequent decoding processes generate the 3D enhanced image. In this way, near-high-resolution scanning overall imaging quality can be obtained while maintaining a wide field-of-view imaging capability, effectively overcoming the technical bottleneck of traditional cone-beam CT imaging technology where "imaging range and resolution are difficult to achieve simultaneously," significantly improving the reconstruction quality and application value of CT images. For details of steps S230 to S240, please refer to steps S150 to S170 above.

[0147] Therefore, by jointly reconstructing the first projection dataset with local high resolution and the second projection dataset with overall low resolution, a globally consistent 3D reconstructed image with preserved local details is obtained. The 3D reconstructed image and the first projection dataset are then input into the global feature extraction branch and the local feature extraction branch of the pre-trained fusion enhancement model, respectively, to extract global and local features at different resolution levels. Through an attention mechanism, spatial alignment and weighted fusion of global and local features are achieved at the corresponding resolution levels to generate a 3D enhanced image. This achieves a deep integration of global structure and local high-resolution details, injecting local high-resolution details into the global image while maintaining the consistency of the overall scene structure. This is beneficial for improving the overall image resolution and imaging quality while maintaining the large field-of-view imaging capability.

[0148] Please refer to Figure 8 This embodiment also provides a tomographic imaging device, applied to a tomographic scanning equipment with a high-resolution imaging link and a wide field-of-view imaging link, including:

[0149] The second acquisition module 210 is used to acquire a local high-resolution first projection dataset and an overall low-resolution second projection dataset.

[0150] The second reconstruction module 220 is used to perform joint reconstruction based on the first projection dataset and the second projection dataset to obtain a three-dimensional reconstructed image.

[0151] The first extraction module 230 is used to input the 3D reconstructed image into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels.

[0152] The second extraction module 240 is used to input the first projection dataset into the local feature extraction branch of the fusion enhancement model in order to extract local features at the corresponding resolution level.

[0153] The enhanced fusion module 250 is used to spatially align and weightedly fuse global and local features at corresponding resolution levels based on an attention mechanism to generate a 3D enhanced image.

[0154] The inventive concept of this tomographic imaging device embodiment is the same as that of the tomographic imaging method embodiment described above. Content not covered in this tomographic imaging device embodiment can be referred to in the tomographic imaging method embodiment described above, and will not be repeated here. Joint reconstruction is performed based on a locally high-resolution first projection dataset and a globally low-resolution second projection dataset to obtain a globally consistent 3D reconstructed image with preserved local details. The 3D reconstructed image and the first projection dataset are respectively input into the global feature extraction branch and local feature extraction branch of a pre-trained fusion enhancement model to extract global and local features at different resolution levels. An attention mechanism is used to align and weightedly fuse global and local features at corresponding resolution levels, generating a 3D enhanced image. This achieves a deep integration of global structure and local high-resolution details, injecting local high-resolution details into the global image while maintaining overall scene structural consistency. This is beneficial for improving overall image resolution and imaging quality while maintaining large field-of-view imaging capabilities.

[0155] Please refer to Figure 9 (a) In this embodiment, a tomographic scanning device is also provided, including a processor 310 and a memory 320. The memory 320 stores a computer program, and the processor 310 runs the computer program to implement the above-described tomographic scanning imaging method. The details and beneficial effects of the tomographic scanning imaging method can be found above and will not be repeated here.

[0156] Please refer to Figure 9 (b) In this embodiment, a storage medium 410 is also provided, which stores a computer program 420. When the computer program 420 is run, it implements the above-described tomographic imaging method. The details and beneficial effects of the tomographic imaging method can be found above and will not be repeated here.

[0157] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A tomographic imaging method, characterized in that, include: Obtain a locally high-resolution first projection dataset and a globally low-resolution second projection dataset; Load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space. Load a first projection noise weighting matrix and a second projection noise weighting matrix. Each weight element of the first projection noise weighting matrix is ​​used to characterize the noise confidence level of each projection measurement point in the first projection dataset, and each weight element of the second projection noise weighting matrix is ​​used to characterize the noise confidence level of each projection measurement in the second projection dataset. Based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix, the weighted reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset is minimized and combined with the regularization constraints of the image structure. The three-dimensional reconstructed image is obtained through iterative reconstruction.

2. The computed tomography imaging method according to claim 1, characterized in that, The step of minimizing the weighted reconstruction residual between the target voxel field and the first and second projection datasets based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix, and the second projection noise weighting matrix includes: Based on the first projection matrix and the first projection noise weighting matrix, the first weighted reconstruction residual between the simulated projection data generated by the target voxel field and the first projection dataset is determined. Based on the second projection matrix and the second projection noise weighting matrix, a second weighted reconstruction residual is determined between the simulated projection data generated by the target voxel field and the second projection dataset. Minimize the sum of the square norms of the first weighted reconstruction residual and the second weighted reconstruction residual.

3. The computed tomography imaging method according to claim 2, characterized in that, The step of determining the first reconstruction residual between the simulated projection data generated from the target voxel field and the first projection dataset based on the first projection matrix and the first projection noise weighting matrix includes: Based on the first projection matrix, the target voxel field is projected forward to generate first simulated projection data; The residuals between the first simulated projection data and the first projection dataset are weighted based on the first projection noise weighting matrix to determine the first weighted reconstruction residual; And / or, The step of determining the second weighted reconstruction residual between the simulated projection data generated from the target voxel field and the second projection dataset based on the second projection matrix and the second projection noise weighting matrix includes: Based on the second projection matrix, the target voxel field is projected forward to generate second simulated projection data; The residuals between the second simulated projection data and the second projection dataset are weighted based on the second projection noise weighting matrix to determine the second weighted reconstruction residual.

4. The computed tomography imaging method according to claim 1 or 2, characterized in that, The acquisition of the locally high-resolution first projection dataset and the globally low-resolution second projection dataset includes: Acquire a local high-resolution first projection dataset from the high-resolution imaging link and a global low-resolution second projection dataset from the wide-field-of-view imaging link; or, Acquire a local high-resolution first projection dataset from a high-resolution imaging link and a global low-resolution second projection dataset from a wide-field-of-view imaging link, wherein the high-resolution imaging link and the wide-field-of-view imaging link are integrated in the same tomographic scanning device. or, A first high-resolution projection dataset from a high-resolution imaging link and a second low-resolution projection dataset from a wide-field-of-view imaging link are acquired. The high-resolution imaging link and the wide-field-of-view imaging link are integrated into the same tomographic scanning device, and the first projection dataset and the second projection dataset are acquired from the same target object during the synchronous rotation of the high-resolution imaging link and the wide-field-of-view imaging link.

5. The computed tomography imaging method according to claim 1, characterized in that, The tomographic imaging method further includes: The 3D reconstructed image is input into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels; The first projection dataset is input into the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level; Based on the attention mechanism, the global features and the local features are spatially aligned and weighted and fused at the corresponding resolution level to generate a 3D enhanced image.

6. The computed tomography imaging method according to claim 5, characterized in that, The step of inputting the 3D reconstructed image into the global feature extraction branch of the pre-trained fusion enhancement model to extract global features at different resolution levels includes: The three-dimensional reconstructed image is divided into blocks and embedded to obtain the first feature block; The first feature block is sequentially processed through multiple sets of context aggregation modules to perform global context modeling, and the spatial resolution is gradually reduced through downsampling operations to extract global features at different resolution levels. And / or, The first projection dataset is input to the local feature extraction branch of the fusion enhancement model to extract local features at the corresponding resolution level, including: The first projection dataset is segmented and embedded to obtain the second feature block; The second feature block is sequentially passed through multiple sets of convolution modules for local feature extraction and block merging to extract local features at the corresponding resolution level.

7. A tomographic imaging device, characterized in that, include: The first acquisition module is used to acquire a local high-resolution first projection dataset and a global low-resolution second projection dataset. The first loading module is used to load the first projection matrix corresponding to the first projection dataset and the second projection matrix corresponding to the second projection dataset. The first projection matrix and the second projection matrix are constructed based on the geometric calibration structure of the same calibration model to align the coordinate systems of the first projection dataset and the second projection dataset in the common reconstruction space. The second loading module is used to load the first projection noise weighting matrix and the second projection noise weighting matrix. Each weight element of the first projection noise weighting matrix is ​​used to characterize the noise confidence level of each projection measurement point in the first projection dataset, and each weight element of the second projection noise weighting matrix is ​​used to characterize the noise confidence level of each projection measurement in the second projection dataset. The first reconstruction module is used to minimize the weighted reconstruction residual between the target voxel field and the first projection dataset and the second projection dataset based on the first projection matrix, the first projection noise weighting matrix, the second projection matrix and the second projection noise weighting matrix, and combine the regularization constraints of the image structure to obtain a three-dimensional reconstructed image through iterative reconstruction.

8. A computed tomography (CT) scanner, comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the processor runs the computer program, it is used to implement the computed tomography imaging method as described in any one of claims 1 to 6.

9. A storage medium storing a computer program, characterized in that, When the computer program is run, it implements the computed tomography imaging method as described in any one of claims 1 to 6.

Images