Projection data processing method and device for cone beam CT half-fan scan

Abstract

This invention relates to the field of image processing technology and discloses a method and apparatus for processing projection data in cone-beam CT half-fan scanning. The method includes: acquiring raw projection data obtained from cone-beam CT half-fan scanning; inputting a first diagonal projection data pair from the raw projection data into a pre-trained artificial intelligence model to obtain a pair of synthetic projection data, wherein the size of the pair of synthetic projection data is larger than the raw projection data and equivalent to the projection data obtained from full-fan scanning. This invention, through AI, learns complete projection data from two basic diagonal (approximately 180° interval) projection data, obtaining projection data equivalent to that acquired by a large-size detector, thereby increasing the scanning field of view of cone-beam CT.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a projection data processing method and apparatus for cone-beam CT half-fan scanning. Background Technology

[0002] Cone-beam computed tomography (CBCT) based on a flat-panel detector offers advantages such as lightweight construction and high image resolution, making it widely used in medical specialties (e.g., orthopedics, dentistry) and scientific research (e.g., materials science, biology). A cone-beam CT scanner primarily consists of an X-ray source (including the X-ray tube and high-voltage generator), a flat-panel detector, a scanning gantry, a computer, and a housing. The flat-panel detector, which converts X-rays into projection data, is typically the most expensive component. CBCT usually acquires hundreds of projection data points of the object being scanned along a circular trajectory. These data are then processed by a computer (including CT image reconstruction, image denoising, and artifact removal) to obtain a CT image containing the object's three-dimensional structure.

[0003] Due to limitations in detector manufacturing processes, data transmission speeds, and overall system production costs, cone-beam CT scanners cannot use excessively large detectors. For example, in microscopic imaging, high-resolution detectors have relatively small pixel sizes (≤75μm), but the stitching process cannot meet the production requirements for large-size detectors (e.g., ≥230mm), and CT manufacturers cannot afford the high price of large-size detectors. In orthopedic surgery guidance, to ensure CT scanning speed, the detector frame rate must be sufficiently high (e.g., ≥20fps), but the data transmission bottleneck in the acquisition circuit cannot guarantee the high frame rate performance of large-size detectors (e.g., ≥43cm). Detector size directly determines the upper limit of the field of view in cone-beam CT scanning; for example, the scanning field of view in microscopic imaging is generally ≤80mm. Summary of the Invention

[0004] This invention provides a projection data processing method and apparatus for cone-beam CT half-fan scanning to solve the problem of the low upper limit of the field of view in cone-beam CT scanning.

[0005] In a first aspect, the present invention provides a projection data processing method for cone-beam CT half-fan scanning, the method comprising: Acquire raw projection data from cone-beam CT half-fan scans; The first diagonal projection data pair in the original projection data is input into a pre-trained artificial intelligence model to obtain a pair of synthetic projection data. The size of the pair of synthetic projection data is larger than that of the original projection data and is equivalent to the projection data obtained by full-fan scanning.

[0006] Secondly, the present invention provides a projection data processing apparatus for cone-beam CT half-fan scanning, the apparatus comprising: The raw projection data acquisition module is used to acquire the raw projection data obtained from cone-beam CT half-fan scans; The synthesis module is used to input the first diagonal projection data pair in the original projection data into a pre-trained artificial intelligence model to obtain a pair of synthesized projection data. The size of the pair of synthesized projection data is larger than that of the original projection data and is equivalent to the projection data obtained by full-fan scanning.

[0007] Thirdly, the present invention provides an electronic device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the projection data processing method for cone-beam CT half-fan scanning described in the first aspect or any corresponding embodiment thereof.

[0008] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the projection data processing method for cone-beam CT half-fan scanning according to the first aspect or any corresponding embodiment thereof.

[0009] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the projection data processing method for cone-beam CT half-fan scanning according to the first aspect or any corresponding embodiment described above.

[0010] This invention utilizes an artificial intelligence model to synthesize diagonal projection data pairs obtained from half-fan scans of cone-beam CT. The resulting synthesized projection data is larger than the original projection data obtained from half-fan scans and equivalent to the projection data obtained from full-fan scans, thus increasing the scanning field of view of cone-beam CT. Therefore, this embodiment achieves a significant increase in the scanning field of view of cone-beam CT without increasing the detector size or increasing the overall cost of the cone-beam CT system, while ensuring that the CT image quality is not degraded. This will greatly expand the application areas of cone-beam CT and enhance its product competitiveness, for example, it can be used for peripheral orthopedic microscopy to promote the early and accurate diagnosis of osteoporosis. Attached Figure Description

[0011] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the geometric structure of cone-beam CT imaging according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the field of view in cone-beam CT scanning according to an embodiment of the present invention; Figure 3 These are schematic diagrams of Full-fan scanning and Half-fan scanning according to embodiments of the present invention; Figure 4 This is a schematic diagram of a Half-fan diagonal projection according to an embodiment of the present invention; Figure 5 This is a schematic flowchart of a projection data processing method for cone-beam CT half-fan scanning according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the Half-fan projection data processing flow according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the network framework of an artificial intelligence model according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the network architecture of Conv_Mamba_block according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the network architecture of the VssBlock module according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the projection data acquisition and processing flow according to an embodiment of the present invention; Figure 11 This is a schematic diagram illustrating the creation of the model training dataset according to an embodiment of the present invention; Figure 12 This is a structural block diagram of a projection data processing apparatus for cone-beam CT half-fan scanning according to an embodiment of the present invention; Figure 13 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0015] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0016] The principle of CT imaging is to use a radiation source to generate X-rays at hundreds of angles along a circular trajectory. These rays penetrate the object being acquired (which can be a person) and are collected by a detector. The detector converts the X-rays into digital signals, i.e., projection data. Then, using CT image reconstruction algorithms, a three-dimensional image of the object is calculated.

[0017] The geometry of cone-beam CT imaging is as follows Figure 1 As shown, L represents the distance from X-ray source 201 to detector 202, S represents the distance from X-ray source 201 to the rotation axis, D represents the distance from the rotation axis to detector 202, O represents the rotation axis of the scanning gantry (also known as the "rotation center"), T represents the effective length of the detector, and SFOV represents the CT scanning field of view. Based on the imaging geometry, the formula for calculating the cone-beam CT scanning field of view (SFOV) is as follows: .

[0018] As can be seen from the above formula, without changing the imaging geometry (L, S, D), increasing the SFOV requires increasing the T value, i.e., using a larger detector (i.e., a longer one). For example... Figure 2As shown, the SFOV here is located laterally, that is, coplanar with the gantry rotation, and is called the lateral SFOV. It determines the "thickness" of objects that cone-beam CT can scan. The axial SFOV is orthogonal to the gantry rotation and determines the "length" of objects that cone-beam CT can scan in a single circular rotation. The lateral SFOV is more important, so when SFOV is mentioned, it generally refers to the lateral SFOV.

[0019] Based on whether the focal point of the X-ray source 201 and the center of the detector 202 are aligned in the CT imaging structure, CT scanning methods are divided into Full-fan scanning and Half-fan scanning. Figure 3 As shown.

[0020] In a full-fan scan, the focal point of the X-ray source is aligned with the center of the flat panel detector. This is typically achieved using a full-bowtie (thin in the middle, thicker on the sides) made of metal to adjust the energy distribution of the X-ray beam (higher energy X-rays penetrating the center of the object, and lower energy X-rays at the edges). Due to detector size limitations, this CT scanning method results in a small field of view. The projection data generated by a full-fan scan can be directly input into a reconstruction algorithm to generate CT images.

[0021] During half-fan scanning, the flat panel detector has a lateral offset (t) relative to the X-ray source, and the maximum value of t does not exceed half the effective length of the detector (i.e., The term "detector half-length" is used for short. The detector half-length minus the lateral offset t equals the overlap length of the two sets of projected images diagonally (i.e., in the 180° direction). ),like Figure 4 As shown. At this point, in the Full-fan scan mode, the equivalent detector half-length is (Tt), the equivalent detector full length is (2×(Tt)), and the detector width remains unchanged. The projection data generated by the Half-fan scan needs to be preprocessed (to solve the lateral translation problem) before it can be fed into the reconstruction algorithm to generate CT images. Because the equivalent detector full length (2×(Tt)) is much larger than the original detector full length T, Half-fan scanning can increase the scanning field of cone-beam CT. Typically, the overlap length ( The field of view needs to be large enough, otherwise artifacts will appear in the reconstructed image, which limits the expansion of the Half-fan scanning field of view.

[0022] According to an embodiment of the present invention, a projection data processing method for cone-beam CT half-fan scanning is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0023] This embodiment provides a projection data processing method for cone-beam CT half-fan scanning, which can be used in various electronic devices, including cone-beam CT devices and other computer devices. Figure 5 This is a flowchart of a projection data processing method for cone-beam CT half-fan scanning according to an embodiment of the present invention, as shown below. Figure 5 As shown, the process includes the following steps: Step S501: Obtain the raw projection data obtained from cone-beam CT half-fan scan.

[0024] Step S502: Input the first diagonal projection data pair in the original projection data into the pre-trained artificial intelligence model to obtain a pair of synthetic projection data. The size of the pair of synthetic projection data is larger than the original projection data and is equivalent to the projection data obtained by full-fan scanning.

[0025] The diagonal projection data pair refers to a pair of diagonally opposite projection data. Diagonal projection data is the projection data obtained by scanning at opposite scanning angles, or in other words, the projection data obtained by scanning at scanning angles spaced 180 degrees apart. In this embodiment, the scanning angles of the diagonal projection data are not strictly limited to a 180-degree interval; the scanning interval angle can be 180 degrees or close to 180 degrees, meaning the scanning angles only need to be approximately diagonal.

[0026] For example, such as Figure 6 As shown, the input to the artificial intelligence model is the first diagonal projection data pair A and P in the original projection data, and then the output is a pair of synthetic projection data sA and sP. Of course, the original projection data obtained from cone-beam CT half-fan scanning can form multiple diagonal projection data pairs, which can be synthesized separately using the artificial intelligence model in this embodiment to obtain multiple pairs of synthetic projection data.

[0027] In this embodiment, the synthesized projection data has the same size as the projection data acquired by the equivalent large-size detector. Compared to the original detector, the equivalent large-size detector only has a larger lateral dimension (i.e., length), while the axial dimension (i.e., width) remains unchanged.

[0028] The projection data processing method for half-fan scanning in cone-beam CT provided in this embodiment utilizes an artificial intelligence model to synthesize diagonal projection data pairs obtained from half-fan scanning. The resulting synthesized projection data is larger than the original projection data obtained from half-fan scanning and is equivalent to the projection data obtained from full-fan scanning, thus increasing the scanning field of view of cone-beam CT. Therefore, this embodiment achieves a significant increase in the scanning field of view of cone-beam CT without increasing the detector size or increasing the overall cost of the cone-beam CT system. For example, the scanning field of view can be increased from 80mm to 140mm, while ensuring that the CT image quality is not degraded. This will greatly expand the application areas of cone-beam CT and enhance its product competitiveness. For example, it can be used for peripheral orthopedic microscopy to promote the early and accurate diagnosis of osteoporosis.

[0029] Furthermore, CT image reconstruction algorithms are divided into two categories: analytical reconstruction algorithms and iterative reconstruction algorithms. Analytical reconstruction algorithms are based on the Fourier central slice theorem, directly solving for the reconstructed image through processes such as filtering, transformation, and interpolation. This method is computationally efficient and can quickly obtain CT images, and is adopted by most CT manufacturers. Iterative reconstruction algorithms are based on optimization theory, transforming the CT image to be solved into an objective function containing a projection model, projection data, and prior knowledge. Using optimization methods, through multiple iterations, the objective function converges to a preset minimum value, thus obtaining the CT image. This method can achieve higher image quality, but it has high computational complexity and is very time-consuming, requiring the use of CPU or GPU parallel computing technology to accelerate reconstruction.

[0030] Specifically, analytical reconstruction algorithms are based on the Fourier transform theory. Classical algorithms mainly include the two-dimensional CT image reconstruction algorithm FBP (Filtered Back Projection) and the three-dimensional image reconstruction algorithm FDK (the three-dimensional form of FBP). Among them, the FDK algorithm is the most commonly used, and its calculation formula is as follows:

[0031] Where S is the distance from the ray source to the center of rotation. The angle of rotation between the X-ray source and the detector. For the collected projection data, These are the detector pixel coordinates (i.e., the projection data coordinates). For a one-dimensional filter in the horizontal direction, This represents one-dimensional convolution. As can be seen from the formula, the projected data at each angle is first processed by convolution filtering and then back-projected, exhibiting relative independence.

[0032] The objective function of the iterative reconstruction algorithm can be expressed as:

[0033] in, Let g represent the CT image that minimizes the objective function, g represent the acquired projection data, and M represent the system matrix (obtained from the CT imaging system modeling, representing the process of transforming the CT image domain to the projection data domain). Represents the regularization term (optional).

[0034] Classical iterative reconstruction algorithms mainly include algorithms such as the Simultaneous Algebraic Reconstruction Technique (SART) and the Expectation Maximization Algorithm (EM). The calculation formula for the SART algorithm is as follows:

[0035] in, This represents the image from the nth iteration. This is the relaxation factor, which controls the convergence of the SART algorithm; The system matrix was used to construct the projection relationship between image pixels on the detector; It is a diagonal matrix, and its elements are... ,Right now The row and the reciprocal; It is a diagonal matrix, and its elements are... ,Right now The column and its reciprocal; express The transpose of .

[0036] To reconstruct Half-fan projection data, related technologies require weighted preprocessing of the projection data before reconstruction, followed by image reconstruction using filtered back projection (FBP or FDK) algorithms. In other words, reconstruction of Half-fan projection data can only be performed using analytical reconstruction algorithms (such as FBP and FDK), and iterative reconstruction algorithms are not feasible. This method of reconstructing Half-fan projection data using an analytical reconstruction algorithm after weighted preprocessing requires that the detector offset not be too large, meaning the overlapping area of ​​the diagonal projection data must be sufficiently large; otherwise, a large bright spot (i.e., an artifact) will appear in the center of the reconstructed CT image, limiting the maximum scanning field of view corresponding to the Half-fan scanning mode. Furthermore, this method of reconstructing Half-fan projection data using an analytical reconstruction algorithm heavily relies on the accuracy of CT geometric parameter calibration, specifically the actual geometric parameters (distance from the X-ray source to the detector (L), distance from the X-ray source to the rotation axis (S), center alignment deviation between the X-ray source and the detector, etc.). U、 The accuracy of the values ​​for V and detector tilt (Tilt, Yaw, Roll) is crucial. Errors in these geometric parameters will result in severe artifacts in the reconstructed CT images.

[0037] However, in this embodiment, by using an artificial intelligence model to synthesize the diagonal projection data pairs obtained from cone-beam CT half-fan scanning, the resulting synthesized projection data is equivalent to the projection data obtained from full-fan scanning. Therefore, it is not limited to the method described above of first performing weighted preprocessing on the half-fan projection data and then using an analytical reconstruction algorithm for reconstruction; iterative reconstruction algorithms can also be used to reconstruct images from the half-fan projection data. Moreover, compared to the method described above that uses an analytical reconstruction algorithm, using an iterative reconstruction algorithm to reconstruct images from the half-fan projection data can obtain higher quality CT images.

[0038] In summary, the projection data processing method for cone-beam CT Half-fan scanning provided in this embodiment, for existing Half-fan projection data, learns (similar to interpolation and synthesis) complete projection data from two basic diagonal (approximately 180° interval) projection data using AI, thus obtaining projection data equivalent to that acquired by a large-size detector. Compared to existing projection processing methods (such as analytical weighted methods), this embodiment allows for greater offset (i.e., smaller overlap length) when cone-beam CT uses Half-fan scanning, thereby providing a larger scanning field of view and simultaneously supporting analytical and iterative reconstruction algorithms.

[0039] In some optional implementations, the artificial intelligence model is constructed based on a combination of convolutional neural network model and Mamba model.

[0040] This embodiment proposes a novel artificial intelligence model, derived from a combination of a CNN and a Mamba model, named CMsH2F. This name signifies the synthesis (synthesize, abbreviated as s) of projection data from Half-fan (H) to Full-fan (F) based on the CNN and Mamba models (CM). The network framework of this new model could be, for example, as shown below. Figure 7 The network architecture shown.

[0041] In some optional implementations, the artificial intelligence model includes an encoder and a decoder; The encoder includes a patch embedding module, a convolutional-Mamba block module, and a patch merging module. The convolutional-Mamba block module is a feature processing module that fuses the convolutional neural network model and the Mamba model. The decoder includes a patch expanding module, a convolutional-Mamba block module, an upsampling module, and a convolutional layer module.

[0042] Specifically, such as Figure 7 As shown, the artificial intelligence model provided in this embodiment basically follows the Unet network architecture, including an encoder on the left, a decoder on the right, and multiple residual connections (also known as "jump connections") between them.

[0043] In some optional implementations, the block embedding module is used to segment the input raw projection data into multiple blocks (patch) and encode them into feature vectors to obtain an initial feature representation; the convolution-Mamba module in the encoder integrates the local modeling of the convolutional neural network model and the global sequence modeling of the Mamba model to extract contextual information; the block merging module is used to merge multiple blocks to reduce the feature size. The block expansion module gradually restores spatial resolution to reconstruct details through upsampling and projection; the convolution-Mamba module in the decoder has its input skipped to the output of the corresponding convolution-Mamba module in the encoder, and the convolution-Mamba module utilizes its ability to fuse local and global information to optimize the upsampled features through coordinated skip connections.

[0044] In this embodiment, the block embedding module converts the input raw projection data into serialized feature tokens, providing an initial feature representation for the encoder. The convolutional-Mamba module in the encoder efficiently extracts contextual information by fusing local modeling from convolution and global sequence modeling from Mamba. The block expansion module gradually restores spatial resolution to reconstruct details through upsampling and projection; the convolutional-Mamba module in the decoder utilizes its ability to fuse local and global information, and collaboratively optimizes the upsampled features using skip connections. Residual connections concatenate feature maps along the channel dimension, fusing local detail features from different levels of the encoder with deep semantic features from the decoder across levels to compensate for spatial information loss during upsampling, ensuring the edge and detail accuracy of the model output.

[0045] In some optional implementations, the convolutional-mamba module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a visual state-space block (VSS block). The output of the first convolutional layer is connected to the second convolutional layer and the visual state space block, respectively. The output of the second convolutional layer is input to the third convolutional layer after nonlinearization. The output of the first convolutional layer is connected to the output of the third convolutional layer in a skip connection and then connected to the input of the fourth convolutional layer. The output of the visual state space block is connected to the input of the fourth convolutional layer. The second and third convolutional layers are used for local modeling, and the fourth convolutional layer is used for global sequence modeling.

[0046] Specifically, the network architecture of the convolutional-Mamba module is as follows: Figure 8 As shown, the Convolution-Mamba module splits the input through a 1×1 convolution (first convolutional layer), and uses parallel branches (3×3 convolutions (second and third convolutional layers) for local modeling and VSS Block for global sequence modeling) to fuse the results. Then, it achieves efficient collaboration between local features and long-range dependencies through a 1×1 convolution (fourth convolutional layer) and residual connections.

[0047] In some optional implementations, the Visual State Space Block (VSS Block) includes a first normalized (LayerNorm) layer, a first linear layer, a depthwise separable convolutional (DW_Conv) layer, a first weighted linear unit (SiLU), a 2D selective scan (SS2D) layer, a second normalized layer, and a second linear layer connected in sequence. The output of the first normalized layer is connected to the output of the second linear layer in a skip connection; the output of the first linear layer is connected to the output of the second normalized layer in a skip connection after passing through the second weighted linear unit.

[0048] In summary, the artificial intelligence model provided in this embodiment is a network structure with dual inputs and dual outputs, which is formed by combining and improving CNN and Mamba.

[0049] Specifically, the network architecture of VSS Block is as follows: Figure 9 As shown in the figure. In this embodiment, VssBlock splits features through LayerNorm and linear transformation, and uses a dual-branch structure (depth convolution + SS2D for multi-directional sequence modeling and linear projection to preserve global information). Finally, selective feature enhancement is achieved through feature fusion and gating mechanisms.

[0050] In some optional implementations, before inputting the first diagonal projection data pair from the original projection data into a pre-trained artificial intelligence model to obtain a pair of synthetic projection data, the method further includes: Step S01: Construct the source domain dataset and the corresponding target domain dataset; Step S02: Train the artificial intelligence model using the source domain dataset and the corresponding target domain dataset; The source domain data in the source domain dataset includes multiple pairs of second diagonal projection data obtained by scanning through the target experimental platform. The target experimental platform and the target cone-beam CT use the same imaging components and imaging geometric parameters. The target cone-beam CT is used for Half-fan scanning to obtain the original projection data. The target domain data in the target domain dataset includes a first equivalent Full-fan projection data and a second equivalent Full-fan projection data corresponding to the second diagonal projection data pair. The first equivalent Full-fan projection data and the second equivalent Full-fan projection data are obtained by image processing through one of the second diagonal projection data pairs and the corresponding third projection data. The third projection data has the same scanning angle as the corresponding second diagonal projection data, and the detector has the same translation stroke but opposite direction.

[0051] Based on whether the X-ray source and flat panel detector in the scanning gantry rotate synchronously, cone-beam CT scanners are divided into desktop and gantry types. Desktop cone-beam CT scanners use a rotating stage to acquire hundreds of projection data points of the object being scanned, and are typically used in industrial inspection, biological sample testing, and other fields. Gantry cone-beam CT scanners use a ring-shaped gantry with a fixed X-ray source and flat panel detector to acquire hundreds of projection data points of the object being scanned, and are typically used in medical imaging diagnosis, surgical image guidance, and other fields.

[0052] In this embodiment, in order to collect training data for the artificial intelligence model, a corresponding CT experimental platform (i.e., the target experimental platform) is built based on the target cone-beam CT. Both use the same imaging components (including X-ray source, detector, filter, etc.) and imaging geometric parameters (including the distance from the X-ray source to the detector (L), the distance from the X-ray source to the rotation axis (S), and the center alignment deviation between the X-ray source and the detector). U、 The difference lies in the detector support components of the experimental platform, which have a large-stroke, high-precision translation function, and the detector tilt angle (Tilt, Yaw, Roll). Furthermore, the CT experimental platform uses a desktop gantry, which is easier and more convenient than a rack-mounted gantry in terms of scan control and data acquisition.

[0053] After the CT experimental platform was set up, Half-fan projection data and reverse Half-fan projection data were acquired at each scanning angle of the target cone-beam CT. The Half-fan projection data is in the same orientation as the target cone-beam CT, meaning the translational travel and direction of the detector support assembly are the same; the reverse Half-fan projection data is in the opposite orientation to the target cone-beam CT, meaning the translational travel of the detector support assembly is the same but the direction is opposite. The acquired projection data were processed (cropping, stitching, etc.) to obtain the equivalent large-size detector's Full-fan projection data (referred to as "equivalent Full-fan projection data"), as shown below. Figure 10 As shown.

[0054] Then, the Half-fan and equivalent Full-fan projection data from two basic diagonals (approximately 180 degrees apart) are combined to form paired data, such as... Figure 11 As shown. Two of the Half-fan projection datasets belong to the source domain dataset used for AI model training, and two equivalent Full-fan projection datasets belong to the target domain dataset used for AI model training.

[0055] In summary, this embodiment utilizes a cone-beam CT experimental platform with a high-precision detector translation structure to acquire two Half-fan projection data sets and process them to form equivalent Full-fan projection data.

[0056] This embodiment also provides a projection data processing device for cone-beam CT half-fan scanning, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0057] This embodiment provides a projection data processing device for cone-beam CT half-fan scanning, such as... Figure 12 As shown, it includes: The raw projection data acquisition module 1201 is used to acquire the raw projection data obtained from cone-beam CT half-fan scanning. The synthesis module 1202 is used to input the first diagonal projection data pair in the original projection data into a pre-trained artificial intelligence model to obtain a pair of synthesized projection data. The size of the pair of synthesized projection data is larger than the original projection data and is equivalent to the projection data obtained by Full-fan scanning.

[0058] In some alternative implementations, the artificial intelligence model is constructed based on a combination of convolutional neural network model and Mamba model.

[0059] In some alternative implementations, the artificial intelligence model includes an encoder and a decoder; The encoder includes a block embedding module, a convolutional-Mamba module, and a block merging module. The convolutional-Mamba module is a feature processing module that fuses the convolutional neural network model and the Mamba model. The decoder includes a block expansion module, a convolutional-Mamba module, an upsampling module, and a convolutional layer module.

[0060] In some optional implementations, the block embedding module is used to segment the input raw projection data into multiple blocks and encode them into feature vectors to obtain an initial feature representation; the convolution-Mamba module in the encoder fuses the local modeling of the convolutional neural network model and the global sequence modeling of the Mamba model to extract contextual information; the block merging module is used to merge multiple blocks to reduce the feature size. The block expansion module gradually restores spatial resolution to reconstruct details through upsampling and projection; the convolution-Mamba module in the decoder connects the input to the output residual of the corresponding convolution-Mamba module in the encoder, and the convolution-Mamba module utilizes the ability to fuse local and global information to optimize the upsampled features in coordination with the residual connection.

[0061] In some alternative implementations, the convolutional-Mamba module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a visual state space block; The output of the first convolutional layer is connected to the second convolutional layer and the visual state space block, respectively. The output of the second convolutional layer is input to the third convolutional layer after nonlinearization. The output of the first convolutional layer is connected to the output residual of the third convolutional layer and then connected to the input of the fourth convolutional layer. The output of the visual state space block is connected to the input of the fourth convolutional layer. The second and third convolutional layers are used for local modeling, and the fourth convolutional layer is used for global sequence modeling.

[0062] In some optional implementations, the visual state space block includes a first normalized layer, a first linear layer, a depth-separable convolutional layer, a first weighted linear unit, a 2D selective scan layer, a second normalized layer, and a second linear layer connected in sequence. The output of the first normalized layer is connected to the output residual of the second linear layer; the output of the first linear layer is connected to the output residual of the second normalized layer after passing through the second weighted linear unit.

[0063] In some optional embodiments, the projection data processing apparatus for cone-beam CT half-fan scanning further includes a model training module, the model training module being used for: Construct the source domain dataset and the corresponding target domain dataset; The artificial intelligence model is trained using the source domain dataset and the corresponding target domain dataset; The source domain data in the source domain dataset includes multiple pairs of second diagonal projection data obtained by scanning through the target experimental platform. The target experimental platform and the target cone-beam CT use the same imaging components and imaging geometric parameters. The target cone-beam CT is used for Half-fan scanning to obtain the original projection data. The target domain data in the target domain dataset includes a first equivalent Full-fan projection data and a second equivalent Full-fan projection data corresponding to the second diagonal projection data pair. The first equivalent Full-fan projection data and the second equivalent Full-fan projection data are obtained by image processing through one of the second diagonal projection data pairs and the corresponding third projection data. The third projection data has the same scanning angle as the corresponding second diagonal projection data, and the detector has the same translation stroke but opposite direction.

[0064] The projection data processing apparatus for cone-beam CT half-fan scanning provided in this embodiment of the invention can execute the projection data processing method for cone-beam CT half-fan scanning provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0065] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0066] The following is a detailed reference. Figure 13 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 1301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1302 or a program loaded from memory 1308 into random access memory (RAM) 1303. The RAM 1303 also stores various programs and data required for the operation of the electronic device. The processor 1301, ROM 1302, and RAM 1303 are interconnected via a bus 1304. An input / output (I / O) interface 1305 is also connected to the bus 1304.

[0067] Typically, the following devices can be connected to I / O interface 1305: input devices 1306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 1308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1309. Communication device 1309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 13 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0068] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 1309, or installed from memory 1308, or installed from ROM 1302. When the computer program is executed by processor 1301, it performs the functions defined in the projection data processing method for cone-beam CT half-fan scanning according to embodiments of the present invention.

[0069] Figure 13 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0070] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the projection data processing method for cone-beam CT half-fan scanning shown in the above embodiments is implemented.

[0071] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0072] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for processing projection data for cone-beam CT half-fan scanning, characterized in that, The method includes: Acquire raw projection data from cone-beam CT half-fan scans; The first diagonal projection data pair in the original projection data is input into a pre-trained artificial intelligence model to obtain a pair of synthetic projection data. The size of the pair of synthetic projection data is larger than that of the original projection data and is equivalent to the projection data obtained by full-fan scanning.

2. The method according to claim 1, characterized in that, The artificial intelligence model is constructed based on a combination of convolutional neural network model and Mamba model.

3. The method according to claim 2, characterized in that, The artificial intelligence model includes an encoder and a decoder; The encoder includes a block embedding module, a convolutional-Mamba module, and a block merging module. The convolutional-Mamba module is a feature processing module that fuses the convolutional neural network model and the Mamba model. The decoder includes a block expansion module, a convolutional-Mamba module, an upsampling module, and a convolutional layer module.

4. The method according to claim 3, characterized in that, The block embedding module is used to segment the input raw projection data into multiple blocks and encode them into feature vectors to obtain an initial feature representation; the convolution-Mamba module in the encoder integrates the local modeling of the convolutional neural network model and the global sequence modeling of the Mamba model to extract contextual information; the block merging module is used to merge multiple blocks to reduce the feature size. The block expansion module gradually restores spatial resolution to reconstruct details through upsampling and projection; the convolution-Mamba module in the decoder connects the input to the output residual of the corresponding convolution-Mamba module in the encoder, and the convolution-Mamba module utilizes the ability to fuse local and global information to optimize the upsampled features in coordination with the residual connection.

5. The method according to claim 3, characterized in that, The convolutional-Mamba module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, and a visual state space block; The output of the first convolutional layer is connected to the second convolutional layer and the visual state space block, respectively. The output of the second convolutional layer is input to the third convolutional layer after nonlinearization. The output of the first convolutional layer is connected to the output residual of the third convolutional layer and then connected to the input of the fourth convolutional layer. The output of the visual state space block is connected to the input of the fourth convolutional layer. The second and third convolutional layers are used for local modeling, and the fourth convolutional layer is used for global sequence modeling.

6. The method according to claim 5, characterized in that, The visual state space block includes a first normalized layer, a first linear layer, a depth-separable convolutional layer, a first weighted linear unit, a 2D selective scan layer, a second normalized layer, and a second linear layer connected in sequence. The output of the first normalized layer is connected to the output residual of the second linear layer; the output of the first linear layer is connected to the output residual of the second normalized layer after passing through the second weighted linear unit.

7. The method according to any one of claims 1 to 6, characterized in that, Before inputting the first diagonal projection data pair from the original projection data into a pre-trained artificial intelligence model to obtain a pair of synthetic projection data, the method further includes: Construct the source domain dataset and the corresponding target domain dataset; The artificial intelligence model is trained using the source domain dataset and the corresponding target domain dataset; The source domain data in the source domain dataset includes multiple pairs of second diagonal projection data obtained by scanning through the target experimental platform. The target experimental platform and the target cone-beam CT use the same imaging components and imaging geometric parameters. The target cone-beam CT is used for Half-fan scanning to obtain the original projection data. The target domain data in the target domain dataset includes a first equivalent Full-fan projection data and a second equivalent Full-fan projection data corresponding to the second diagonal projection data pair. The first equivalent Full-fan projection data and the second equivalent Full-fan projection data are obtained by image processing through one of the second diagonal projection data pairs and the corresponding third projection data. The third projection data has the same scanning angle as the corresponding second diagonal projection data, and the detector has the same translation stroke but opposite direction.

8. A projection data processing device for cone-beam CT half-fan scanning, characterized in that, The device includes: The raw projection data acquisition module is used to acquire the raw projection data obtained from cone-beam CT half-fan scans; The synthesis module is used to input the first diagonal projection data pair in the original projection data into a pre-trained artificial intelligence model to obtain a pair of synthesized projection data. The size of the pair of synthesized projection data is larger than that of the original projection data and is equivalent to the projection data obtained by full-fan scanning.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the projection data processing method for cone-beam CT half-fan scanning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the projection data processing method for cone-beam CT Half-fan scanning as described in any one of claims 1 to 7.

Images