A computed tomography image reconstruction method and system based on extremely sparse scanning

Abstract

The present application relates to a kind of based on extremely sparse scanning computer tomography image reconstruction method and system, comprising: obtaining CT machine original scanning data and the reconstruction image of standard algorithm;Adopt the generation of adversarial network;Design loss function, train the generation of adversarial network;VdCS algorithm and CT-RDNet network are combined through feedback path, construct extremely sparse scanning CT image reconstruction framework, i.e.CSUF image reconstruction framework;Obtain the scanning data of CT machine extremely sparse scanning mode, i.e.extremely sparse scanning data;With extremely sparse scanning data as input, CSUF image reconstruction framework executes extremely sparse scanning CT image reconstruction, and outputs the reconstruction image satisfying the needs of radiology clinical diagnosis.The method of the present application can obtain CT image satisfying the needs of radiology clinical diagnosis under extremely sparse scanning condition, and has good robustness and engineering practicability.

Description

Technical Field

[0001] This invention relates to the field of medical imaging technology, and in particular to a method and system for reconstructing computed tomography images based on extremely sparse scanning. Background Technology

[0002] Computed tomography (CT), with its high resolution and rapid imaging capabilities, has become an indispensable technology in clinical diagnosis. However, the ionizing radiation from X-rays during CT scans poses a potential hazard to human health. Prolonged or excessive exposure to X-ray radiation increases the risk of cancer. Reducing the number of CT scans is an effective way to lower the X-ray radiation dose. However, in extremely sparse scans (scan data scale equivalent to no more than 20 fan-beam / cone-beam scan angles within the [0, 2π) range), severe undersampling leads to significant artifacts in the reconstructed images, and a high risk of losing high-frequency or detail information. This weakens the clinical diagnostic value and status of CT images and hinders the wider clinical application and promotion of CT imaging technology.

[0003] Traditional CT image reconstruction methods, including analytical algorithms such as Filtered Backscan (FBP) and iterative image reconstruction algorithms such as ART, perform poorly under extremely sparse scanning conditions, failing to provide reconstructed images that meet the needs of clinical radiological diagnosis. The paper "View-driven Compressed Sensing method of CT image reconstruction" (IOP Conference Series: Materials Science and Engineering, 768, 062003) proposes a CT iterative image reconstruction algorithm (Scan-driven Compressed Sensing CT Image Reconstruction Algorithm, VdCS algorithm) suitable for extremely sparse scanning based on compressed sensing theory. The VdCS algorithm can achieve image reconstruction relatively effectively under extremely sparse scanning conditions. However, the limitations of compressed sensing theory are also evident in the VdCS algorithm, such as the staircase effect, and the tendency for high-frequency information loss or blur artifacts in the reconstructed image. In recent years, deep learning-based CT image reconstruction methods have attracted attention. Compared with traditional algorithms, deep learning-based methods have shown significant advantages in noise suppression, artifact removal, and improved image detail preservation. However, most existing deep learning-based CT image reconstruction methods are limited to post-processing in the image domain, i.e., learning the mapping from low-quality reconstructed images to high-quality images through neural networks. Under extremely sparse scanning conditions, due to the poor quality of the original input images, existing deep learning methods struggle to accurately learn the aforementioned mapping relationship. This can lead to problems such as overly smooth reconstructed images, loss of detailed information, and even crucial lesion information. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a CT image reconstruction method that integrates iterative CT image reconstruction and deep learning. This method can obtain CT images that meet clinical diagnostic needs under extremely sparse scanning conditions (the number of scanning angles in fan-beam / cone-beam scanning within the range of [0,2π) is no more than 20), and has good robustness and engineering practicality.

[0005] The present invention also provides a computed tomography image reconstruction system based on extremely sparse scanning.

[0006] Terminology Explanation:

[0007] 1. VdCS algorithm, a CT image reconstruction algorithm driven by view-driven compressed sensing, abbreviated as VdCS algorithm.

[0008] 2. The standard scanning mode of a CT scanner refers to the typical or standard scanning parameters and sampling strategies used by the CT scanner when imaging the cross-section of the human body, which are within the range of current mainstream clinical practice or recommended by the equipment manufacturer, in order to obtain sufficiently dense and uniformly distributed scanning data, thereby enabling high-quality and complete image reconstruction.

[0009] 3. Extremely sparse scanning mode of CT scanner refers to a scanning strategy in which the CT scanner acquires only a very small amount of sparsely distributed scanning data (the scanning data scale is equivalent to no more than 20 scanning angles in the fan beam / cone beam scanning range [0,2π). It includes undersampling scanning mode and finite angle scanning mode. Among them, undersampling scanning mode means that the CT scanner acquires sparse scanning data only in a very small number of uniformly distributed scanning angles, and finite angle scanning mode means that the CT scanner acquires sparse scanning data only in a limited number of scanning angles.

[0010] The technical solution of this invention is as follows:

[0011] A computed tomography image reconstruction method based on extremely sparse scanning includes:

[0012] Step 1: Acquire the raw CT scan data and reconstructed images using standard algorithms;

[0013] The original scan data refers to the X-ray attenuation measurement data. Several scanning angles are uniformly selected from the original scan data to obtain sparse scan data, which refers to the scan data with a small number of scanning angles. The VdCS algorithm is used to obtain sparse reconstructed images with different numbers of scanning angles. The sparse reconstructed images are paired with the reconstructed images of the standard algorithm. Pairing means that the sparse reconstructed images with different numbers of scanning angles are matched one-to-one with the reconstructed images of the standard algorithm to obtain the training dataset for supervised deep learning.

[0014] Step 2: Construct a CT image restoration and noise reduction network, namely the CT-RDNet network;

[0015] Step 3: Using the CT-RDNet network as a generator, construct a generative adversarial network;

[0016] Step 4: Design the loss function and train the generative adversarial network;

[0017] Step 5: Combine the VdCS algorithm and the CT-RDNet network through the feedback path to construct an extremely sparse scanning CT image reconstruction framework, namely the CSUF image reconstruction framework.

[0018] Step 6: Obtain the scan data of the CT scanner in the extremely sparse scanning mode, i.e., the extremely sparse scan data;

[0019] Step 7: Using extremely sparse scan data as input, the CSUF image reconstruction framework performs extremely sparse scan CT image reconstruction and outputs reconstructed images that meet the needs of clinical imaging diagnosis.

[0020] According to a preferred embodiment of the present invention, the CT-RDNet network is a symmetrical U-shaped architecture of encoder-decoder;

[0021] The encoder of the CT-RDNet network includes eight downsampling modules for extracting features from sparse reconstructed images and downsampling the sparse reconstructed images.

[0022] The decoder of the CT-RDNet network includes seven upsampling modules and an output layer, which are used to reconstruct high-quality images from the features extracted from the sparse reconstructed images.

[0023] The CT-RDNet network uses skip connections to fuse features and a self-attention mechanism to focus on global features. Specifically, the output feature map of the upsampling module is concatenated with the output feature map of the downsampling module of the same size in the channel direction to obtain the fused feature. After being processed by the self-attention module, the attention feature map is obtained, and finally upsampling is performed.

[0024] The input feature map, i.e. the fused feature map, is processed by three 1×1 convolutions to obtain three corresponding feature maps, denoted as Q, K, and V. The obtained feature maps Q, K, and V are flattened into one-dimensional vectors of length H×W in the spatial dimension, and Q is transposed. The transposed Q and K are multiplied and processed using the SoftMax function to obtain the attention score. V is multiplied by the attention score and the vector is restored to a two-dimensional vector of size H×W to obtain the attention output. Finally, the attention output is processed by a 1×1 convolution to obtain the attention feature map after self-attention enhancement.

[0025] According to a preferred embodiment of the present invention, the discriminator of the generative adversarial network includes five discriminator modules, a zero-padding layer, and an output layer; the discriminator modules are used to extract image features; the zero-padding layer is used to adjust the feature map size.

[0026] According to a preferred embodiment of the present invention, a loss function is designed, and a generative adversarial network is trained; including:

[0027] loss function The definition is as follows:

[0028]

[0029] Among them, E (x～p(x)) Let x represent the expected value of a random variable x sampled from a distribution p(x), where x is the input image. Let y be the output image of the CT-RDNet network, y be the reconstructed image of the standard algorithm; D represents the discriminator; c, C1, and C2 are constants; N is the number of pixels in the image; μ " and These are the mean pixel values ​​of the reconstructed image from the standard algorithm and the output image from the CT-RDNet network, respectively. and These are the variances of pixel values ​​in the reconstructed image from the standard algorithm and the output image from the CT-RDNet network, respectively. β is the standard deviation of pixel values ​​in the reconstructed image from the standard algorithm and the output image from the CT-RDNet network; β is the weight parameter.

[0030] According to a preferred embodiment of the present invention, using extremely sparse scan data as input, the CSUF image reconstruction framework performs extremely sparse scan CT image reconstruction and outputs reconstructed images that meet the needs of clinical imaging diagnosis; including:

[0031] (1) Use the VdCS algorithm to perform iterative image reconstruction on extremely sparse scan data and pass the reconstructed image to the CT-RDNet network;

[0032] (2) The reconstructed image from step (1) is processed by the CT-RDNet network for image restoration and noise reduction optimization.

[0033] (3) The image optimized by the CT-RDNet network is provided to the VdCS algorithm as a prior image through the feedback path;

[0034] (4) Using the prior image as the initial estimate, iterative image reconstruction is performed again on the extremely sparse scan data, and the image reconstruction result is passed to the CT-RDNet network.

[0035] (5) The iterative image reconstruction results of the VdCS algorithm are further optimized through the CT-RDNet network, and the reconstructed images that meet the needs of clinical imaging diagnosis are output.

[0036] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a computed tomography image reconstruction method based on extremely sparse scanning.

[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a computed tomography image reconstruction method based on extremely sparse scanning.

[0038] A computed tomography image reconstruction system based on extremely sparse scanning, comprising:

[0039] The data acquisition module is configured to: acquire raw CT scan data and reconstructed images from a standard algorithm; the raw scan data refers to X-ray attenuation measurement data; uniformly select several scanning angles from the raw scan data to obtain sparse scan data, which refers to scan data with a small number of scanning angles; use the VdCS algorithm to obtain sparse reconstructed images with different numbers of scanning angles; pair the sparse reconstructed images with the reconstructed images from the standard algorithm, which means matching the sparse reconstructed images with different numbers of scanning angles one-to-one with the reconstructed images from the standard algorithm to obtain a training dataset for supervised deep learning.

[0040] The Generative Adversarial Network (GAN) building module is configured to: construct a CT image restoration and denoising network, namely the CT-RDNet network; and use the CT-RDNet network as a generator to build a GAN.

[0041] The Generative Adversarial Network (GAN) training module is configured to: design a loss function and train the GAN.

[0042] The image reconstruction module is configured to: combine the VdCS algorithm and the CT-RDNet network through a feedback path to construct an extremely sparse scanning CT image reconstruction framework, namely the CSUF image reconstruction framework; acquire scanning data of the CT machine's extremely sparse scanning mode, namely extremely sparse scanning data; and use the extremely sparse scanning data as input, the CSUF image reconstruction framework performs extremely sparse scanning CT image reconstruction and outputs reconstructed images that meet the needs of clinical imaging diagnosis.

[0043] The beneficial effects of this invention are as follows:

[0044] Compared with the prior art, the method of the present invention can obtain CT images that meet the needs of clinical imaging diagnosis under extremely sparse scanning conditions (the number of scanning angles of fan beam / cone beam scanning in the range of [0,2π) is no more than 20), and has good robustness and engineering applicability. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the computed tomography image reconstruction method based on extremely sparse scanning according to the present invention.

[0046] Figure 2 This is a schematic diagram of the process for creating the dataset of this invention;

[0047] Figure 3(a) is a schematic diagram of the structure of the CT-RDNet network of the present invention;

[0048] Figure 3(b) is a schematic diagram of the self-attention module structure of the CT-RDNet network of the present invention;

[0049] Figure 3(c) is a schematic diagram of the discriminator of the generative adversarial network of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of the extremely sparse scanning CT image reconstruction framework (CSUF image reconstruction framework) of the present invention; Detailed Implementation

[0051] The present invention will be further defined below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0052] Example 1

[0053] A computed tomography image reconstruction method based on extremely sparse scanning, such as Figure 1 As shown, it includes:

[0054] Step 1: Acquire the raw scan data from the CT scanner's standard scanning mode and the reconstructed images from the standard algorithm;

[0055] Raw scan data refers to X-ray attenuation measurement data directly acquired from the CT scanner detector, without preprocessing or image reconstruction. Reconstructed images using standard algorithms typically refer to images obtained using image reconstruction algorithms (such as the filtered back-scan algorithm, FBP algorithm) preset or defaulted by the CT equipment manufacturer for general human tomographic tissue imaging. Standard CT algorithm reconstructed images often strike a relatively balanced trade-off between spatial resolution and noise suppression, suitable for most clinical imaging needs. Sparse scan data is obtained by uniformly selecting several scanning angles from the raw scan data. Sparse scan data refers to scan data with a small number of scanning angles, for example, 20, 45, 90, and 180 degrees. Compared to the raw scan data, the amount of data in sparse scan data is significantly reduced. The VdCS algorithm is used to obtain sparse reconstructed images with different numbers of scanning angles. These sparse reconstructed images are then paired with reconstructed images from the standard algorithm. Pairing involves matching each sparse reconstructed image with a corresponding image from the standard algorithm, resulting in a supervised deep learning training dataset. The training dataset creation process is as follows: Figure 2 As shown.

[0056] Step 2: Construct a CT image restoration and noise reduction network, namely the CT-RDNet network;

[0057] Step 3: Using the CT-RDNet network as a generator, construct a generative adversarial network;

[0058] Step 4: Design the loss function and train the generative adversarial network;

[0059] Step 5: Combine the VdCS algorithm and the CT-RDNet network through a feedback path to construct an extremely sparse scanning CT image reconstruction framework, namely the CSUF image reconstruction framework; the structure of the CSUF image reconstruction framework is as follows: Figure 4 As shown; the CT-RDNet network provides prior image information to the VdCS algorithm through the feedback path.

[0060] Step 6: Obtain the scan data of the CT scanner in the extremely sparse scanning mode, i.e., the extremely sparse scan data;

[0061] Step 7: Using extremely sparse scan data as input, the CSUF image reconstruction framework performs extremely sparse scan CT image reconstruction (i.e., CSUF image reconstruction) and outputs reconstructed images that meet the needs of clinical imaging diagnosis.

[0062] Example 2

[0063] The difference between the computed tomography image reconstruction method based on extremely sparse scanning described in Example 1 and the method described in Example 1 is as follows:

[0064] As shown in Figure 3(a), the CT-RDNet network is a symmetrical U-shaped architecture of encoder-decoder;

[0065] The encoder of the CT-RDNet network includes eight downsampling modules for extracting features from sparse reconstructed images and downsampling the sparse reconstructed images.

[0066] The eight downsampling modules are connected in sequence: Down1, Down2, Down3, Down4, Down5, Down6, Down7, and Down8.

[0067] Down1 includes convolutional layers and activation functions;

[0068] Both Down2 and Down3 include convolutional layers, activation functions, and batch normalization layers;

[0069] Down4, Down5, Down6, and Down7 all include convolutional layers, activation functions, batch normalization layers, and dropout layers;

[0070] Down8 includes convolutional layers, activation functions, and dropout layers;

[0071] The convolutional layers have a kernel size of 4×4, a stride of 2, and padding of 1; the activation function used is the LeakyReLU function with a slope of 0.2; and dropout layers are used to reduce overfitting.

[0072] The decoder of the CT-RDNet network includes seven upsampling modules and an output layer, which are used to reconstruct high-quality images from the features extracted from the sparse reconstructed images.

[0073] The seven upsampling modules are Up1, Up2, Up3, Up4, Up5, Up6, and Up7, which are connected in sequence.

[0074] Up1, Up2, and Up3 all include transposed convolutional layers, batch normalization layers, and ReLU activation functions;

[0075] Up4, Up5, Up6, and Up7 all include transposed convolutional layers, batch normalization layers, ReLU activation functions, and dropout layers;

[0076] The transposed convolutional layer has a kernel size of 4×4, a stride of 2, and padding of 1; a dropout layer is used to reduce overfitting.

[0077] The output layer consists of a transposed convolutional layer with a kernel size of 4×4, a stride of 2, padding of 1, and a Tanh activation function.

[0078] The CT-RDNet network uses skip connections to fuse features and a self-attention mechanism to focus on global features. Specifically, the output feature map of the first upsampling module is concatenated with the output feature map of the same size of the seventh downsampling module in the channel direction to obtain the fused feature. After being processed by the self-attention module, the attention feature map is obtained, and finally, upsampling is performed.

[0079] Here, the implementation process of the self-attention module of the present invention (as shown in Figure 3(b)) is as follows:

[0080] The input feature map, i.e. the fused feature map, is processed by three 1×1 convolutions to obtain three corresponding feature maps, denoted as Q, K, and V. The obtained feature maps Q, K, and V are flattened into one-dimensional vectors of length H×W in the spatial dimensions (i.e., height (H) and width (W)), and Q is transposed. The transposed Q and K are multiplied and processed using the SoftMax function to obtain the attention score. V is multiplied by the attention score and the vector is restored to a two-dimensional size of H×W to obtain the attention output. Finally, the attention output is processed by a 1×1 convolution to obtain the attention feature map after self-attention enhancement.

[0081] As shown in Figure 3(c), the discriminator of the generative adversarial network includes five discriminator modules, a zero-padding layer, and an output layer; the discriminator modules are used to extract image features; the zero-padding layer is used to adjust the feature map size.

[0082] The five discrimination modules include Dise1, Dise2, Dise3, Dise4, and Dise5.

[0083] Dise1 includes convolutional layers, LeakyReLU activation function, and dropout layers;

[0084] Dise2, Dise3, Dise4, and Dise5 all include convolutional layers, LeakyReLU activation function, dropout layers, and batch normalization layers;

[0085] The convolutional layers have a kernel size of 4×4, a stride of 2, and padding of 1. The slope of the LeakyReLU activation function is 0.2. Zero-padding layers are used to adjust the feature map size. The output layer uses a convolutional layer without bias, with a kernel size of 4×4, a stride of 1, and padding of 1. Here, the output size of the discriminator and the corresponding label size are both 16×16.

[0086] Designing a loss function and training a generative adversarial network; including:

[0087] The overall loss function of the CT-RDNet network in this invention integrates LSGAN loss, L1 loss, and SSIM loss. The definition is as follows:

[0088]

[0089] Among them, E (x～p(x)) Let x represent the expected value of a random variable x sampled from a distribution p(x), where x is the input image. Let y be the output image of the CT-RDNet network, and y be the reconstructed image from the standard algorithm; D represents the discriminator; c, C1, and C2 are constants, 100, 0.01, and 0.03 respectively; N is the number of pixels in the image; μ y and These are the mean pixel values ​​of the reconstructed image from the standard algorithm and the output image from the CT-RDNet network, respectively. and These are the variances of pixel values ​​in the reconstructed image from the standard algorithm and the output image from the CT-RDNet network, respectively. β is the standard deviation of pixel values ​​in the reconstructed image from the standard algorithm and the output image from the CT-RDNet network; β is a weight parameter, typically set to 0.4.

[0090] The discriminator loss function of the generative adversarial network of this invention adopts LSGAN loss, with generated samples having a label of 0 and real samples having a label of 1.

[0091] The specific steps for training a generative adversarial network are as follows:

[0092] Data preparation: Using the training dataset constructed in step one, sparsely reconstructed images are used as input images, and images reconstructed by the standard algorithm are used as target images. Standardizing the input and target images before training ensures consistent data distribution, which helps accelerate network convergence and improve training stability.

[0093] The specific training process includes:

[0094] (1) Initialize the weights of the generator and discriminator.

[0095] (2) For each batch of input images, execute:

[0096] i. Image generation: Input the standardized sparse reconstructed image into the generator to generate a preliminary reconstructed image (generated sample);

[0097] ii. Update the discriminator: Input the target image (real sample, labeled 1) and the generated image (generated sample, labeled 0) into the discriminator, calculate the discriminator loss and perform backpropagation, update the discriminator weights to improve its ability to distinguish between real and generated images.

[0098] iii. Update the generator: Calculate the loss between the generated image and the target image (LSGAN loss, L1 loss, and SSIM loss), perform backpropagation, update the generator weights, and optimize the quality of the generated image.

[0099] (3) Training stops when the termination condition is met.

[0100] Using extremely sparse scan data as input, the CSUF image reconstruction framework performs reconstruction of extremely sparse scan CT images, outputting reconstructed images that meet the needs of clinical radiological diagnosis; including:

[0101] (1) Use the VdCS algorithm to perform iterative image reconstruction on extremely sparse scan data. The VdCS algorithm is an iterative image reconstruction algorithm. It reconstructs the extremely sparse scan data to obtain the reconstructed image and passes the reconstructed image to the CT-RDNet network.

[0102] (2) The reconstructed images from step (1) are restored and denoised using the CT-RDNet network; the sparsely scanned CT reconstructed images are restored and denoised using CT-RDNet.

[0103] (3) The image optimized by the CT-RDNet network is provided to the VdCS algorithm as a prior image through the feedback path;

[0104] (4) Using the prior image as the initial estimate, the VdCS algorithm iteratively reconstructs the extremely sparse scan data again. Similar to step (1), the difference is that the input of the VdCS algorithm used here is an additional prior image, and the image reconstruction result is passed to the CT-RDNet network.

[0105] (5) Further optimize the iterative image reconstruction results of the VdCS algorithm through the CT-RDNet network. Similar to step (2), CT-RDNet is used for restoration and noise reduction, and the reconstructed image that meets the needs of clinical imaging diagnosis is output.

[0106] Example 3

[0107] A computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the computed tomography image reconstruction method based on extremely sparse scanning as described in Embodiment 1 or 2.

[0108] Example 4

[0109] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the computed tomography image reconstruction method based on extremely sparse scanning as described in Embodiment 1 or 2.

[0110] Example 5

[0111] A computed tomography image reconstruction system based on extremely sparse scanning, comprising:

[0112] The data acquisition module is configured to: acquire raw CT scan data and reconstructed images from a standard algorithm; the raw scan data refers to X-ray attenuation measurement data; uniformly select several scanning angles from the raw scan data to obtain sparse scan data, which refers to scan data with a small number of scanning angles; use the VdCS algorithm to obtain sparse reconstructed images with different numbers of scanning angles; pair the sparse reconstructed images with the reconstructed images from the standard algorithm, which means matching the sparse reconstructed images with different numbers of scanning angles one-to-one with the reconstructed images from the standard algorithm to obtain a training dataset for supervised deep learning.

[0113] The Generative Adversarial Network (GAN) building module is configured to: construct a CT image restoration and denoising network, namely the CT-RDNet network; and use the CT-RDNet network as a generator to build a GAN.

[0114] The Generative Adversarial Network (GAN) training module is configured to: design a loss function and train the GAN.

[0115] The image reconstruction module is configured to: combine the VdCS algorithm and the CT-RDNet network through a feedback path to construct an extremely sparse scanning CT image reconstruction framework, namely the CSUF image reconstruction framework; acquire scanning data of the CT machine's extremely sparse scanning mode, namely extremely sparse scanning data; and use the extremely sparse scanning data as input, the CSUF image reconstruction framework performs extremely sparse scanning CT image reconstruction and outputs reconstructed images that meet the needs of clinical imaging diagnosis.

Claims

1. A computed tomography image reconstruction method based on extremely sparse scanning, characterized by, The application relates to a CT image reconstruction method based on a supervised deep learning method. Step one, obtaining CT machine original scanning data and a standard algorithm reconstruction image; The original scanning data refers to X-ray attenuation measurement data; a plurality of scanning angles are uniformly selected from the original scanning data to obtain sparse scanning data, the sparse scanning data refers to scanning data under a small number of scanning angles, and a VdCS algorithm is used to obtain sparse reconstruction images under different scanning angle numbers; the sparse reconstruction images are paired with the standard algorithm reconstruction image, the pairing refers to one-to-one correspondence between the sparse reconstruction images under different scanning angle numbers and the standard algorithm reconstruction image, and a training data set of the supervised deep learning is obtained; Step two, constructing a CT image restoration and denoising network, namely a CT-RDNet network; Step three, taking the CT-RDNet network as a generator to build a generative adversarial network; the CT-RDNet network is a symmetrical U-shaped architecture of an encoder-decoder; The encoder of the CT-RDNet network comprises eight down-sampling modules, which are used for extracting features of the sparse reconstruction image and down-sampling the sparse reconstruction image; The decoder of the CT-RDNet network comprises seven up-sampling modules and an output layer, which are used for reconstructing a high-quality image from the extracted features of the sparse reconstruction image; The CT-RDNet network uses a skip connection to fuse features and a self-attention mechanism to focus on global features; wherein the output feature map of the up-sampling module is spliced with the output feature map of the down-sampling module in the channel direction to obtain fused features, the attention feature map is obtained after the fused features are processed by the self-attention module, and finally the attention feature map is up-sampled; Step four, designing a loss function to train the generative adversarial network; Step five, combining the VdCS algorithm and the CT-RDNet network through a feedback path to construct a very sparse scanning CT image reconstruction framework, namely a CSUF image reconstruction framework; Step six, obtaining scanning data of a CT machine very sparse scanning mode, namely very sparse scanning data; Step seven, taking the very sparse scanning data as input, and executing very sparse scanning CT image reconstruction by the CSUF image reconstruction framework to output a reconstruction image meeting the needs of clinical diagnosis in radiology; comprising: (1) using the VdCS algorithm to perform iterative image reconstruction on the very sparse scanning data, and delivering the reconstruction image to the CT-RDNet network; (2) performing image restoration and denoising optimization processing on the reconstruction image of step (1) by the CT-RDNet network; (3) the image optimized by the CT-RDNet network is provided as a prior image to the VdCS algorithm through the feedback path; (4) taking the prior image as an initial estimation, performing iterative image reconstruction on the very sparse scanning data again, and delivering the image reconstruction result to the CT-RDNet network; (5) further optimizing the iterative image reconstruction result of the VdCS algorithm by the CT-RDNet network, and outputting a reconstruction image meeting the needs of clinical diagnosis in radiology.

2. The computed tomography image reconstruction method based on extremely sparse scan according to claim 1, characterized in that, The input feature map, i.e., the fused feature, is obtained through three 1x1 convolutions to obtain three corresponding feature maps, denoted as Q, K and V; the obtained feature maps Q, K and V are flattened in the spatial dimension into one-dimensional vectors with a length of HxW, and Q is transposed; the transposed Q and K are multiplied and processed using a SoftMax function to obtain attention scores; V and the attention scores are multiplied, and the vector is restored to a two-dimensional size of HxW to obtain attention output; finally, the attention output is processed using a 1x1 convolution to obtain an attention feature map after self-attention reinforcement.

3. The computed tomography image reconstruction method based on extremely sparse scan according to claim 1, characterized in that, The discriminator of the generative adversarial network comprises five discrimination modules, a zero padding layer and an output layer; the discrimination module is used for extracting image features; The zero padding layer is used for adjusting the size of the feature map.

4. The computed tomography image reconstruction method based on extremely sparse scan according to any one of claims 1-3, characterized in that, A loss function is designed to train the generative adversarial network; Comprise: Loss function is defined as follows: where E (x～p(x)) denotes the expectation value of the random variable x sampled from the distribution p(x), x is the input image, is the output image of the CT-RDNet network, y is the reconstructed image of the standard algorithm; D represents the discriminator; c, C1, C2 are constants, N is the number of pixels of the image; μ y and are the mean of the pixel values of the reconstructed image of the standard algorithm and the output image of the CT-RDNet network, respectively, and are the variance of the pixel values of the reconstructed image of the standard algorithm and the output image of the CT-RDNet network, respectively, is the standard deviation of the pixel values of the reconstructed image of the standard algorithm and the output image of the CT-RDNet network; β is the weight parameter. 5.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-4 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the computer tomography image reconstruction method based on extremely sparse scanning according to any one of claims 1-4.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the computer tomography image reconstruction method based on extremely sparse scanning according to any one of claims 1-4.

7. A computer tomography image reconstruction system based on extremely sparse scanning, characterized in that Comprise: The data acquisition module is configured to: acquire CT machine original scanning data and reconstruction images of a standard algorithm; the original scanning data refers to X-ray attenuation measurement data; a plurality of scanning angles are uniformly selected from the original scanning data to obtain sparse scanning data, the sparse scanning data refers to scanning data under a small number of scanning angles, and the VdCS algorithm is used to obtain sparse reconstruction images under different numbers of scanning angles; the sparse reconstruction images are paired with the reconstruction images of the standard algorithm, which means that the sparse reconstruction images under different numbers of scanning angles are respectively and correspondingly paired with the reconstruction images of the standard algorithm, to obtain a training data set of supervised deep learning; The generative adversarial network construction module is configured to: construct a CT image restoration and denoising network, i.e., a CT-RDNet network; use the CT-RDNet network as a generator to build a generative adversarial network; the CT-RDNet network is a symmetrical U-shaped architecture of an encoder-decoder; The encoder of the CT-RDNet network comprises eight down-sampling modules, which are used for extracting features of the sparse reconstruction images and down-sampling the sparse reconstruction images; The decoder of the CT-RDNet network comprises seven up-sampling modules and an output layer, which are used for reconstructing high-quality images from the extracted features of the sparse reconstruction images; The CT-RDNet network uses a skip connection to fuse features and a self-attention mechanism to focus on global features; wherein, the output feature map of the up-sampling module is spliced with the output feature map of the down-sampling module in the channel direction to obtain fused features, and then the fused features are processed by a self-attention module to obtain an attention feature map, and finally the attention feature map is up-sampled; The generative adversarial network training module is configured to: design a loss function to train the generative adversarial network; The image reconstruction module is configured to combine the VdCS algorithm and the CT-RDNet network through a feedback path to construct a CT image reconstruction framework for extremely sparse scanning, i.e., a CSUF image reconstruction framework; acquire scanning data of a CT machine in an extremely sparse scanning mode, i.e., extremely sparse scanning data; input the extremely sparse scanning data, and perform extremely sparse scanning CT image reconstruction by using the CSUF image reconstruction framework to output a reconstructed image meeting the needs of clinical diagnosis in radiology; The input of the extremely sparse scanning data into the CSUF image reconstruction framework to perform extremely sparse scanning CT image reconstruction and output a reconstructed image meeting the needs of clinical diagnosis in radiology includes: (1) performing iterative image reconstruction on the extremely sparse scanning data by using the VdCS algorithm, and delivering the reconstructed image to the CT-RDNet network; (2) performing image restoration and noise reduction optimization processing on the reconstructed image of step (1) by using the CT-RDNet network; (3) providing the image optimized by the CT-RDNet network as a priori image to the VdCS algorithm through the feedback path; (4) performing iterative image reconstruction on the extremely sparse scanning data again by using the priori image as initial estimation, and delivering the image reconstruction result to the CT-RDNet network; (5) further optimizing the iterative image reconstruction result of the VdCS algorithm by using the CT-RDNet network, and outputting a reconstructed image meeting the needs of clinical diagnosis in radiology.

Images