A CT imaging method and system for cloud image centering for multiple imaging geometries

Abstract

The application discloses a CT imaging method and system for cloud image centers of various imaging geometries, wherein the CT imaging method for cloud image centers of various imaging geometries comprises four steps. The CT imaging method is specifically obtained by constructing an optimized intelligent imaging network of an imaging geometry modulation subnetwork and an intelligent reconstruction subnetwork, then inputting projection data collected by a CT scanning device into the optimized intelligent imaging network, and finally obtaining an optimized CT image. The application can be applied to CT image reconstruction imaging of different imaging geometries and doses and can obtain high-quality reconstructed images, and can greatly improve flexibility.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and computed tomography (CT) technology, and in particular to a CT imaging method and system for cloud imaging centers with multiple imaging geometries. Background Technology

[0002] Computed tomography (CT) has a significant advantage in reconstructing the anatomical structures within a patient's body without causing invasive damage. Due to the potential health risks of radiation exposure to patients during CT scans, low-dose CT imaging has been widely used in clinical diagnosis.

[0003] In recent years, the rapid development of deep learning technology has revealed broad prospects for the intelligentization of the medical industry. Inspired by this, some deep learning-based CT reconstruction methods have been proposed and are gradually becoming mainstream. However, current deep learning-based reconstruction methods are limited by the fact that they can only be trained using datasets with specific imaging geometry and dose levels, and cannot simultaneously process multiple imaging geometry data from different devices, which reduces the flexibility of deep learning reconstruction methods.

[0004] Recently, with the development of big data and cloud computing, cloud imaging centers are becoming a trend in the future development of medical imaging technology. Cloud imaging centers generally concentrate the detection data of different CT imaging equipment. In order to efficiently reconstruct CT images, the traditional deep learning CT reconstruction method based on a single imaging geometry is no longer applicable.

[0005] Therefore, in view of the shortcomings of the existing technology, it is necessary to provide a CT imaging method and a CT imaging system for cloud imaging centers with multiple imaging geometries to solve the shortcomings of the existing technology. Summary of the Invention

[0006] One objective of this invention is to overcome the shortcomings of existing technologies and provide a CT imaging method for cloud imaging centers with multiple imaging geometries. This CT imaging method for cloud imaging centers with multiple imaging geometries can be applied to CT images with different imaging geometries and doses to reconstruct high-quality reconstructed images, greatly improving flexibility.

[0007] The above-mentioned objectives of the present invention are achieved through the following technical measures: A CT imaging method for cloud image centers with various imaging geometries is provided, comprising the following steps: Step (1): Construct an imaging geometric modulation subnetwork for generating modulation feature vectors and an intelligent reconstruction subnetwork that can dynamically adjust the filtering according to the modulation feature vectors. Then, embed the imaging geometric modulation subnetwork into the intelligent reconstruction subnetwork. The imaging geometric modulation subnetwork includes a geometric mapping module and an imaging geometric modulation module. The intelligent reconstruction subnetwork is equipped with a chordal domain filtering module, an image domain filtering module, and a back projection module. Step (2): Perform downsampling acceleration and upsampling recombination to obtain the preliminary intelligent imaging network; Step (3): Train the preliminary intelligent imaging network obtained in step (2) to obtain the optimized intelligent imaging network; Step (4): Input the projection data collected by the CT scanning device into the optimized intelligent imaging network to obtain optimized CT images.

[0008] Preferably, step (1) above specifically includes: Step (1.1): A geometric mapping module is constructed using a multilayer perceptron with multiple fully connected layers. Two geometric mapping modules are constructed in total. The geometric mapping module will convert vectors... The input is mapped to a higher dimension to obtain a high-dimensional modulation feature vector. and high-dimensional modulation feature vector The vector Constructed from all imaging geometry and dose; Step (1.2): Construct a chord graph domain filtering module, an image domain filtering module, and a back projection module respectively. The back projection module is located between the last convolutional layer of the chord graph domain filtering module and the first convolutional layer of the image domain filtering module. Step (1.3): Construct an imaging geometry modulation module using multiple multilayer perceptrons. Two such imaging geometry modulation modules are constructed in total. Each imaging geometry modulation module modulates the high-dimensional feature vector. and high-dimensional modulation feature vector Modulation feature map Modulated feature map As input to the chordal domain filtering module and the image domain filtering module, the It can be expressed by equation (Ⅰ), ... Equation (I); in The feature maps output by the convolutional layer of the chord graph domain filtering module and the convolutional layer of the image domain filtering module.

[0009] Preferably, the chord graph domain filtering module and the image domain filtering module are each constructed from a convolutional residual network.

[0010] Preferably, the back projection module performs Fourier transform, inverse Fourier transform, interpolation, and inverse Leyden transform processing on the chord graph data of the chord graph domain filtering module, and then maps the processed chord graph data to the image domain filtering module.

[0011] Preferably, step (2) above specifically includes: Step (2.1): Construct two downsampling modules. One of the downsampling modules is deployed after the first convolutional layer of the chordal domain filtering module, and the other downsampling module is deployed after the first convolutional layer of the image domain filtering module. The downsampling module reduces the input data by a scaling factor of α, where 2≤α≤16. Step (2.2): Construct two upsampling modules. One of the upsampling modules is deployed before the last convolutional layer of the chordal domain filtering module, and the other upsampling module is deployed before the last convolutional layer of the image domain filtering module. The upsampling module amplifies the input data by a scaling factor α to obtain a preliminary intelligent imaging network.

[0012] Preferably, the above It can be expressed by equation (II), ...Formula (II) in, This refers to the number of projection angles in a CT scan. The number of CT detector units, To reconstruct the number of pixels in the image, To reconstruct the pixel size of the image, The size of the CT detector unit, This represents the distance from the radiation source to the detector. The distance from the ray source to the center of rotation. The incident photon quantity denoted as the standard deviation of Gaussian noise.

[0013] Preferably, step (3) above specifically includes: Step (3.1): Train the preliminary intelligent imaging network obtained in step (2) using normal dose data to obtain a pre-trained model; Step (3.2): The pre-trained model obtained in step (3.1) is further trained using low-dose data to fine-tune and optimize the parameters of the intelligent imaging network, thereby obtaining the optimized intelligent imaging network.

[0014] Preferably, in step (3.1) above, the following is used: L 2 The loss function is used to train the initial intelligent imaging network. L 2 The loss function is expressed by equation (Ⅲ): ... Formula (Ⅲ); in, for L 2 loss function constraints of the paradigm To reconstruct the image, For reference only.

[0015] Preferably, step (4) involves linearly preprocessing the projection data acquired by the CT scanning device and then inputting it into the optimized intelligent imaging network obtained in step (3) to obtain the filtered and reconstructed optimized CT image.

[0016] Preferably, the above linear preprocessing is normalization.

[0017] Preferably, the aforementioned low-dose data is dose data obtained by scanning under a low-dose scanning protocol with low tube voltage or low tube current, or low-dose data obtained through simulation.

[0018] Another objective of this invention is to overcome the shortcomings of existing technologies and provide a cloud-based imaging center CT imaging system with multiple imaging geometries. This cloud-based imaging center CT imaging system with multiple imaging geometries can be applied to CT images with different imaging geometries and doses to reconstruct images and obtain high-quality reconstructed images.

[0019] The above-mentioned objectives of the present invention are achieved through the following technical measures: A CT imaging system for cloud imaging centers with multiple imaging geometries is provided, employing the aforementioned CT imaging method for cloud imaging centers with multiple imaging geometries.

[0020] This invention discloses a CT imaging method and system for cloud imaging centers with multiple imaging geometries. The CT imaging method for cloud imaging centers with multiple imaging geometries includes the following steps: Step (1) Constructing an imaging geometry modulation subnetwork for generating modulation feature vectors and an intelligent reconstruction subnetwork that can dynamically adjust filtering according to the modulation feature vectors, and then embedding the imaging geometry modulation subnetwork into the intelligent reconstruction subnetwork. The imaging geometry modulation subnetwork includes a geometry mapping module and an imaging geometry modulation module, and the intelligent reconstruction subnetwork is equipped with a chordal domain filtering module, an image domain filtering module, and a back projection module; Step (2) Performing downsampling acceleration and upsampling recombination to obtain a preliminary intelligent imaging network; Step (3) Training the preliminary intelligent imaging network obtained in Step (2) to obtain an optimized intelligent imaging network; Step (4) Inputting the projection data acquired by the CT scanning device into the optimized intelligent imaging network to obtain an optimized CT image. This invention can be applied to CT images with different imaging geometries and doses for reconstruction imaging and obtain high-quality reconstructed images, which can greatly improve flexibility. Attached Figure Description

[0021] The invention will be further described with reference to the accompanying drawings, but the contents of the drawings do not constitute any limitation on the invention.

[0022] Figure 1 This is a diagram of the optimized intelligent imaging network structure of the present invention.

[0023] Figure 2 This is the reconstruction result of Example 2. Detailed Implementation

[0024] Example 1 A CT imaging method for cloud image centers with multiple imaging geometries includes the following steps: Step (1): Construct an imaging geometric modulation subnetwork for generating modulation feature vectors and an intelligent reconstruction subnetwork that can dynamically adjust the filtering according to the modulation feature vectors. Then, embed the imaging geometric modulation subnetwork into the intelligent reconstruction subnetwork. The imaging geometric modulation subnetwork includes a geometric mapping module and an imaging geometric modulation module. The intelligent reconstruction subnetwork is equipped with a chordal domain filtering module, an image domain filtering module, and a back projection module. Step (2): Perform downsampling acceleration and upsampling recombination to obtain the preliminary intelligent imaging network; Step (3): Train the preliminary intelligent imaging network obtained in step (2) to obtain an optimized intelligent imaging network, such as... Figure 1 ; Step (4): Input the projection data collected by the CT scanning device into the optimized intelligent imaging network to obtain optimized CT images.

[0025] Step (1) specifically includes: Step (1.1): Construct a geometry mapping module using a multilayer perceptron with multiple fully connected layers. Two geometry mapping modules are constructed in total. The geometry mapping module will convert vectors... The input is mapped to a higher dimension to obtain a high-dimensional modulation feature vector. and high-dimensional modulation feature vector ,vector Constructed from all imaging geometry and dose; Step (1.2): Construct the chord graph domain filtering module, the image domain filtering module, and the back projection module respectively. The back projection module is located between the last convolutional layer of the chord graph domain filtering module and the first convolutional layer of the image domain filtering module. Step (1.3): Construct an imaging geometry modulation module using multiple multilayer perceptrons. A total of two imaging geometry modulation modules are constructed. The imaging geometry modulation module modulates the high-dimensional feature vector. and high-dimensional modulation feature vector Modulation feature map Modulated feature map As input to the chordal domain filtering module and the image domain filtering module, It can be expressed by equation (Ⅰ), ... Equation (I); in The output features are the convolutional layers of the chordmap domain filtering module and the convolutional layers of the image domain filtering module. It should be noted that the imaging geometry modulation module receives the high-dimensional modulation vector from the geometry mapping module and the intermediate feature maps from the chordmap domain and image domain filtering modules. The high-dimensional modulation vector is divided into two parts, β and γ, equal to the number of channels in the feature maps, and then the above processing is performed.

[0026] The chord-graph domain filtering module and the image domain filtering module are each constructed from convolutional residual networks. It should be noted that both the chord-graph domain filtering module and the image domain filtering module contain a series of convolutional layers, regularization layers, and activation function layers.

[0027] The chord graph domain filtering module specifically uses a convolutional residual network to process the input chord graph data (i.e., the projection data collected in step (4)). After the chord graph data passes through the chord graph domain filtering module, the output is the filtered chord graph. The chord graph domain filtering module uses a neural network in the reconstruction process, which plays a role similar to traditional reconstruction filtering.

[0028] The image domain filtering module specifically uses a convolutional residual network to process the image domain data obtained through backprojection. After passing through this module, the final reconstructed image is output. The image domain filtering module corresponds to the post-processing operation of images in traditional reconstruction methods.

[0029] The backprojection module performs Fourier transform, inverse Fourier transform, interpolation, and inverse Lehman transform on the chordmap data from the chordmap domain filtering module, and then maps the processed chordmap data to the image domain filtering module. It's important to note that while the backprojection module performs a series of operations on the chordmap data to backproject it to the image domain filtering module, it does not train network parameters. However, it performs gradient backpropagation during neural network training. Specifically, during gradient backpropagation, the backprojection module performs a pre-projection operation on the gradients back from the image domain filtering module, using this as the gradient propagated back to the chordmap domain filtering module.

[0030] Step (2) of the present invention specifically includes: Step (2.1): Construct two downsampling modules. One downsampling module is deployed after the first convolutional layer of the chordal domain filtering module, and the other downsampling module is deployed after the first convolutional layer of the image domain filtering module. The downsampling module reduces the input data by a scaling factor of α, where 2≤α≤16. Step (2.2): Construct two upsampling modules. One upsampling module is deployed before the last convolutional layer of the chordal domain filtering module, and the other upsampling module is deployed before the last convolutional layer of the image domain filtering module. The upsampling module amplifies the input data by a scaling factor α to obtain the preliminary intelligent imaging network.

[0031] It should be noted that the downsampling module can reduce the input data by a scaling factor from 2 to 16, thereby accelerating the training speed of the entire network. This module adaptively scales the features according to the size of the input data without losing any details. The downsampling module generates different sampling indices from different sampling rates, and the sampling indices sample the data at equal intervals.

[0032] This embodiment describes the downsampling module's processing with a scaling factor of 2. The downsampling module reassembles the pixels in the original feature map without losing any detailed information, reducing the size to 2. The feature map is reshaped into a size of The downsampling feature map.

[0033] The upsampling module amplifies the input data by a scaling factor of 2 to 16, restoring data details, and then reassembles the downsampling data into pixels in the original feature map without losing any detail. Similarly, using a scaling factor of 2, the upsampling module process is explained: by reassembling the pixels in the original feature map without losing any detail, the data is upsampled to a size of... The feature map is reshaped into a size of The original feature map.

[0034] The present invention performs backprojection operation using normal-sized data. Before performing the backprojection operation, an upsampling module is added to the end of the chord graph domain filtering module to restore the chord graph domain data to normal size before performing the backprojection operation. Using normal-sized chord graph domain data for backprojection operation avoids the accumulation of errors caused by downsampling. At the same time, a downsampling module is added before the image domain filtering module to complete the subsequent acceleration.

[0035] Step (3) of the present invention specifically includes: Step (3.1): Train the preliminary intelligent imaging network obtained in step (2) using normal dose data to obtain a pre-trained model; Step (3.2): Use low-dose data to continue training the pre-trained model obtained in step (3.1) to fine-tune and optimize the parameters of the intelligent imaging network, and obtain the optimized intelligent imaging network; The low-dose data of the present invention is the dose data obtained by scanning under a low-dose scanning protocol with low tube voltage or low tube current or the low-dose data obtained by simulation.

[0036] In step (3.1), the following steps are used: L 2 The loss function is used to train the initial intelligent imaging network. L 2 The loss function is expressed by equation (Ⅲ): ... Formula (Ⅲ); in, for L 2 loss function constraints of the paradigm To reconstruct the image, The image shown is for reference. This invention utilizes pre-training with normal dose data to achieve good reconstruction performance in the network.

[0037] It should be noted that the training termination judgment method in step (3.1) can be one of the following two: 1. During the neural network training process, whether the error calculated by the loss function is stable and converges to a small value. The magnitude of this value is not the focus of this invention, and those skilled in the art can determine the magnitude of this value according to the actual situation. 2. After testing with test data, high-quality CT images can be reconstructed from chord graph data with different geometries (achieving good results in visual analysis and quantitative indicators). When one of the above two situations is met, training is stopped and a pre-trained model is obtained.

[0038] Step (4) of the present invention specifically involves linearly preprocessing the projection data acquired by the CT scanning device and then inputting it into the optimized intelligent imaging network obtained in step (3) to obtain the filtered and reconstructed optimized CT image, wherein the linear preprocessing is normalization.

[0039] in, It can be expressed by equation (II), ...Formula (II) in, This refers to the number of projection angles in a CT scan. The number of CT detector units, To reconstruct the number of pixels in the image, To reconstruct the pixel size of the image, The size of the CT detector unit, This represents the distance from the radiation source to the detector. The distance from the ray source to the center of rotation. The incident photon quantity denoted as the standard deviation of Gaussian noise.

[0040] It should be noted that the imaging geometry and dose of the present invention are the number of projection angles of the CT scan, the number of CT detector units, the number of pixels of the reconstructed image, the pixel size of the reconstructed image, the size of the CT detector unit, the distance from the X-ray source to the detector, the distance from the X-ray source to the center of rotation, the incident photon quantity, and the standard deviation of Gaussian noise.

[0041] The imaging geometry and dose are known when the projection data is acquired. This invention... This involves normalizing all parameters and doses in the imaging geometry and dosage, treating each parameter as a dimension to construct a vector. For the present invention... Both the projection data and the intelligent imaging network are used as inputs. The projection data will be input into the intelligent reconstruction subnetwork for filtering and reconstruction. The input is then used to the imaging geometry modulation subnetwork and mapped into a high-dimensional vector, which is used to modulate the high-dimensional feature map during the reconstruction process of the intelligent imaging subnetwork.

[0042] This cloud-based CT imaging method for various imaging geometries can be applied to CT images with different imaging geometries and doses to reconstruct high-quality reconstructed images, greatly improving flexibility.

[0043] Example 2 Application of a CT imaging method for cloud image centers with multiple imaging geometries.

[0044] (1) Data Construction The projection data in this embodiment is 3,000 CT images collected from a publicly available dataset, including 1,000 head images, 1,000 chest images, and 1,000 abdominal images.

[0045] The data in this embodiment consists of 3000 CT images collected from a publicly available dataset, including 1000 head images, 1000 chest images, and 1000 abdominal images. High / low dose projection data were obtained by simulating CT images of different body parts. The imaging geometry and dose used in the simulation are shown in the table below.

[0046] 90% of this data was used to train the neural network, and 10% was used to test and validate the performance of the intelligent imaging network.

[0047] (2) Implementation process S1. Construct the preliminary intelligent imaging network of Example 1; S2. Pre-train the preliminary intelligent imaging network using data without Poisson-Gaussian noise (i.e., normal dose data). Stop training when the network loss function converges to a small value. The resulting pre-trained model can be determined based on the actual situation.

[0048] S3. Using low-dose data, fine-tune the pre-trained model based on S2. Stop training when the network loss function converges to a small value again to obtain the optimized intelligent imaging network.

[0049] S4. Use test data to verify the performance of the optimized intelligent imaging network, calculate quantitative indicators of the reconstruction results, and obtain... Figure 2 .

[0050] like Figure 2 As shown, the embodiment results of reconstructing projection data with different imaging geometries and dose levels using the cloud imaging center CT imaging method for multiple imaging geometries of the present invention are significantly better than those of the traditional filtered back projection method, with a clear noise removal effect. The demonstration results are compared with the reference image by calculating three indicators: peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and root mean square error (RMSE). These three indicators show that the cloud imaging center CT imaging method for multiple imaging geometries of the present invention has a significant improvement over the traditional filtered back projection method.

[0051] Example 3 A cloud imaging center CT imaging system with multiple imaging geometries employs the cloud imaging method for multiple imaging geometries as described in Example 1.

[0052] The system of the present invention includes a CT device, a communication medium, a computer device, and a display.

[0053] This CT device can emit X-rays towards the patient or object being examined. The X-rays penetrate the human body and can be collected by the detector.

[0054] The communication medium is used to transmit information data, which includes, but is not limited to, projection data acquired by the CT scanning device and program instructions sent by the operator.

[0055] Computer devices are used to acquire CT projection data and perform reconstruction. The computer device is generally in the form of a high-performance computing server, which usually includes a central processing unit, a graphics computing unit, and can integrate storage devices for storing important data, including raw CT projection data, deep learning technology (e.g., deep learning model) data, reconstructed CT images, etc.

[0056] The display is used to show image data, including CT projection data, reconstructed CT images, residual images, etc.

[0057] A CT scanner consists of an X-ray tube, an X-ray detector, a high-voltage generator, and a rotating gantry. The X-ray tube emits an X-ray beam to scan an object. The X-ray detector receives the attenuated X-ray beam as it passes through the object. The high-voltage generator provides the necessary high-voltage electric field for the X-ray tube. The rotating gantry holds the X-ray tube and detector in place, and rotates the tube and detector around the object being examined.

[0058] This cloud imaging center CT imaging system with multiple imaging geometries can reconstruct images from CT images with different imaging geometries and doses, and obtain high-quality reconstructed images, which can greatly improve flexibility.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A CT imaging method for cloud image centers with multiple imaging geometries, characterized in that, Includes the following steps: Step (1): Construct an imaging geometric modulation subnetwork for generating modulation feature vectors and an intelligent reconstruction subnetwork that can dynamically adjust the filtering according to the modulation feature vectors. Then, embed the imaging geometric modulation subnetwork into the intelligent reconstruction subnetwork. The imaging geometric modulation subnetwork includes a geometric mapping module and an imaging geometric modulation module. The intelligent reconstruction subnetwork is equipped with a chordal domain filtering module, an image domain filtering module, and a back projection module. Step (2): Perform downsampling acceleration and upsampling recombination to obtain the preliminary intelligent imaging network; Step (3): Train the preliminary intelligent imaging network obtained in step (2) to obtain the optimized intelligent imaging network; Step (4): Input the projection data acquired by the CT scanning device into the optimized intelligent imaging network to obtain optimized CT images; Step (1) specifically includes: Step (1.1): A geometric mapping module is constructed using a multilayer perceptron with multiple fully connected layers. Two geometric mapping modules are constructed in total. The geometric mapping module will convert vectors... The input is mapped to a higher dimension to obtain a high-dimensional modulation feature vector. and high-dimensional modulation feature vector The vector Constructed from all imaging geometry and dose; Step (1.2): Construct a chord graph domain filtering module, an image domain filtering module, and a back projection module respectively. The back projection module is located between the last convolutional layer of the chord graph domain filtering module and the first convolutional layer of the image domain filtering module. Step (1.3): Construct an imaging geometry modulation module using multiple multilayer perceptrons. Two such imaging geometry modulation modules are constructed in total. Each imaging geometry modulation module modulates the high-dimensional feature vector. and high-dimensional modulation feature vector Modulation feature map Modulated feature map As the input to the chordal domain filtering module and the image domain filtering module, the It can be expressed by equation (Ⅰ), ...Formula (I); in The feature maps output by the convolutional layers of the chord graph domain filtering module and the image domain filtering module.

2. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 1, characterized in that: The string-map domain filtering module and the image domain filtering module are each constructed from convolutional residual networks; The back projection module performs Fourier transform, inverse Fourier transform, interpolation, and inverse Leyden transform processing on the chord graph data from the chord graph domain filtering module, and then maps the processed chord graph data to the image domain filtering module.

3. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 1, characterized in that, The specific steps (2) are as follows: Step (2.1): Construct two downsampling modules. One of the downsampling modules is deployed after the first convolutional layer of the chordal domain filtering module, and the other downsampling module is deployed after the first convolutional layer of the image domain filtering module. The downsampling module reduces the input data by a scaling factor of α, where 2≤α≤16. Step (2.2): Construct two upsampling modules. One of the upsampling modules is deployed before the last convolutional layer of the chordal domain filtering module, and the other upsampling module is deployed before the last convolutional layer of the image domain filtering module. The upsampling module amplifies the input data by a scaling factor α to obtain a preliminary intelligent imaging network.

4. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 1, characterized in that: The It can be expressed by equation (II), ...Formula (II) in, This refers to the number of projection angles in a CT scan. The number of CT detector units, To reconstruct the number of pixels in the image, To reconstruct the pixel size of the image, The size of the CT detector unit, This represents the distance from the radiation source to the detector. The distance from the X-ray source to the center of rotation. For the incident photon quantity, denoted as the standard deviation of Gaussian noise.

5. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 3, characterized in that, The specific steps (3) are as follows: Step (3.1): Train the preliminary intelligent imaging network obtained in step (2) using normal dose data to obtain a pre-trained model; Step (3.2): The pre-trained model obtained in step (3.1) is further trained using low-dose data to fine-tune and optimize the parameters of the intelligent imaging network, thereby obtaining the optimized intelligent imaging network.

6. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 5, characterized in that: In step (3.1), the following steps are used: L 2 The loss function is used to train the initial intelligent imaging network. L 2 The loss function is expressed by equation (Ⅲ): ...Formula (III); in, for L 2 loss function constraints of the paradigm To reconstruct the image, For reference only.

7. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 6, characterized in that: Step (4) specifically involves linearly preprocessing the projection data acquired by the CT scanning device and then inputting it into the optimized intelligent imaging network obtained in step (3) to obtain the filtered and reconstructed optimized CT image.

8. The CT imaging method for cloud image centers with multiple imaging geometries according to claim 7, characterized in that: The linear preprocessing is normalization; The low-dose data refers to dose data obtained by scanning under a low-dose scanning protocol with low tube voltage or low tube current, or low-dose data obtained through simulation.

9. A CT imaging system for cloud imaging centers with multiple imaging geometries, characterized in that: The CT imaging method for cloud image centers with multiple imaging geometries as described in any one of claims 1 to 8 is adopted.

Images