Sparse helical CT image reconstruction method based on differentiable helical reconstruction operator

Abstract

A sparse helical CT image reconstruction method based on a differentiable helical reconstruction operator. First, actual helical scanning geometric parameters of a subject and corresponding full-angle helical projection data are acquired, and a final reconstructed image is acquired by means of seven steps. In the present invention, actual scanning geometry is used to perform forward projection on a reconstructed sparse-angle image, thereby providing geometric prior guidance for missing projections; moreover, on the basis of the similarity and redundancy characteristics of adjacent projections in helical scanning, a projection completion network is constructed, and by learning bidirectional motion fields of adjacent angles and in combination with geometric prior projections, intermediate missing projection data is jointly synthesized; in addition, the global streak artifact restoration of the image is realized; and finally, the joint training of a projection domain and an image domain is realized, thereby facilitating integral restoration by using projection-image dual-domain information, and data collected within two pitches is used for restoration, thereby effectively avoiding an excessive computational load.

Description

A sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator Technical Field

[0001] This invention relates to the fields of deep learning and computed tomography, and in particular to a sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator. Background Technology

[0002] Computed tomography (CT), a high-contrast, non-invasive, and rapid imaging technology, has been widely used in medical imaging diagnosis, surgical navigation, and radiotherapy planning. However, studies have shown that exposure to large amounts of X-ray radiation can cause systemic acute and chronic radiation damage, DNA and cell damage, significantly increasing the risk of future cancer. Reducing X-ray radiation dose inevitably leads to degradation of CT image quality, affecting diagnostic and therapeutic performance. Therefore, achieving high-quality low-dose CT imaging to meet clinical needs is a cutting-edge scientific issue in the field of advanced medical CT research. To mitigate the radiation dose hazards of medical CT, various low-dose CT scanning protocols have been proposed. Currently, low-millisecond scanning protocols combined with advanced reconstruction algorithms can achieve routine low-dose clinical task scanning. However, these methods are limited by hardware. When the X-ray photon exposure is further reduced, the measured projection data may experience a "photon starvation" effect, meaning that the amount of X-ray photons generated at ultra-low milliseconds cannot penetrate the object or the amount of X-ray photons that pass through is extremely low. This results in the detector not acquiring effective and sufficient X-ray signals, causing the reconstructed image tissue structure to be overwhelmed by noise artifacts and difficult to recover.

[0003] In achieving ultra-low-dose CT imaging, sparse angle scanning protocols are among the most promising technologies. They reduce radiation dose by decreasing the projection exposure angle, maintaining tube voltage and current at normal dose levels for each projection exposure. Compared to low-mA-second scanning protocols, this protocol offers the following advantages: 1. It effectively avoids photon starvation and modeling complex noise; 2. It avoids performance degradation caused by domain differences between simulated and real low-dose data; 3. It effectively shortens scan time. However, existing sparse angle scanning still cannot meet current clinical needs. When the scanning projection angle is reduced, the sampling rate does not satisfy the classical Shannon sampling theorem, causing image reconstruction to degenerate into an ill-posed inverse problem. When using classical filtered backprojection for reconstruction, artifacts are easily introduced into the image.

[0004] With the rapid development of deep learning, a large number of deep learning-based methods have been proposed to effectively improve image quality. According to the different methods and modeling objects, these methods can be divided into four categories: (1) Sparse angle CT imaging based on projection interpolation. This type of method relies on the correlation between projection angles, learns the mapping between sparse angle projection and full angle projection, and then uses a filtered back projection algorithm to reconstruct the target image. However, due to the characteristics of the FBP reconstruction algorithm, a data point in the projection data corresponds to all pixel values ​​in a ray path through the CT image. Therefore, the error caused by projection interpolation can easily introduce new artifacts that are difficult to handle in the image domain, which is the main reason why this type of method is difficult to improve. 2. Sparse angle CT imaging based on image domain post-processing: This type of method directly learns the mapping between sparse angle images and full-angle images. Although this type of method can effectively remove most of the noise and artifacts in the image, since the convolution in the restoration process is equivalent to the filtering process, it inevitably causes the loss of structural detail information in the image, and the improvement of image quality is limited. 3. Sparse angle CT imaging based on projection-image cross-domain restoration: This type of method learns the mapping between sparse angle projection data and full-angle image data, effectively combining the data information of the projection domain and the image domain, and can obtain imaging performance that is better than that of the projection domain or the image domain alone.

[0005] Current research on sparse angles largely focuses on fan-beam projections under circular scanning trajectories. The sparse angle simulation process mainly includes two methods: 1. Starting from the acquired image data, projecting the image into a fan-beam projection to obtain simulated fan-beam projection data, and then uniformly sampling this projection data to obtain the corresponding sparse angle projection. This method lacks analysis of the actual scanning geometry and helical projection data, making it difficult to directly apply clinically. 2. Starting from the actual helical projection data, rearranging the helical projection data into a fan-beam projection, and then uniformly sampling this fan-beam projection to obtain the corresponding sparse angle projection. Although this method starts from the actual projection data, its recovery performance drops significantly under large pitch conditions. Currently, there is almost no detailed work on sparse angle CT reconstruction algorithms and systems directly based on helical projection data, therefore, they cannot be directly applied to clinical diagnosis and treatment.

[0006] Therefore, it is essential to provide a sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator to address the shortcomings of existing technologies. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator. This sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator constructs a gradient-backpropagable differentiable spiral reconstruction operator and builds a cross-domain deep learning framework for projecting sparse spirals onto high-quality reconstructed images, thereby achieving diagnostic-grade sparse spiral CT reconstruction under low-dose scanning protocols.

[0008] The above-mentioned objectives of the present invention are achieved through the following technical measures:

[0009] A sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator is provided, comprising the following steps:

[0010] S1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data;

[0011] S2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in S1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters.

[0012] S3. Based on the real spiral scanning geometric parameters and full-angle spiral projection data from S1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from S2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image.

[0013] S4. Using the actual spiral scanning geometric parameters of S1, perform 3D front projection on the sparse angle spiral CT image obtained in S3 to obtain the corresponding geometric prior projection.

[0014] S5. The projection data of the missing angles in the geometric prior projection obtained in S4 is recovered by the trained projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in S3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data.

[0015] S6. Using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in S1 and the final completed projection data obtained in S5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion.

[0016] S7. The sparse angle spiral CT image obtained in S3 and the initial restored image data obtained in S6 are processed by the trained image domain deep learning model to restore details and restore residual artifacts, and the final reconstructed image is obtained.

[0017] The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator of the present invention obtains the trained projection completion deep learning model and the trained image domain deep learning model through joint training, wherein each joint training is performed according to the following steps:

[0018] A1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data;

[0019] A2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in A1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters.

[0020] A3. Based on the actual spiral scanning geometric parameters and full-angle spiral projection data from A1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from A2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image.

[0021] A4. Using the actual spiral scanning geometric parameters of A1, perform 3D front projection on the sparse angle spiral CT image obtained in A3 to obtain the corresponding geometric prior projection.

[0022] A5. The projection data of the missing angles in the geometric prior projection obtained in A4 is recovered by the projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in A3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data.

[0023] A6. By using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in A1 and the final completed projection data obtained in A5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion.

[0024] A7. Using an image domain deep learning model, detail restoration and residual artifact restoration are performed on the sparse angle spiral CT image obtained in A3 and the initial restored image data obtained in A6 to obtain the final reconstructed image.

[0025] A8. Calculate the loss using the loss function, where the final completed projection data obtained from A5 is compared with the full-angle spiral projection data from A1 to obtain the loss. The gradient of the final completed projection data obtained from A5 is compared with the gradient of the full-angle spiral projection data from A1 to obtain the loss. The initial restored image data from A6 was compared with the full-angle spiral CT image from A3 to obtain the loss. The final reconstructed image from A7 was compared with the full-angle spiral CT image from A3 to obtain the loss. ;

[0026] A9. The update of an image domain deep learning model involves applying the loss... Gradient backpropagation is performed directly to the image domain deep learning model to update its parameters; the update of the projection-complete deep learning model involves applying the loss function. and loss The gradient is directly backpropagated to the projection-complete deep learning model, while the loss is... and loss The gradient is backpropagated to the projection-complete deep learning model through the 3D differentiable gradient backpropagation operator, and finally the parameters of the projection-complete deep learning model are updated.

[0027] In S2 or A2, uniform sparse sampling specifically involves uniformly sampling the full-angle projection data along the angular direction, as represented by equation (1):

[0028] ...Equation (1);

[0029] in, To simulate sparse angle spiral projection data, This is full-angle spiral projection data. For a uniform sampling process in the angular direction, N is the sparse sampling interval.

[0030] The reconstruction process of the weighted filtering back projection algorithm in S3 or A3 is represented by equations (2) and (3):

[0031] ...Equation (2);

[0032] ...Equation (3);

[0033] in, Full-angle spiral CT images, These are sparse angle spiral CT images. These are the actual geometric parameters for helical scanning. To simulate sparse scan geometry parameters, To achieve full-angle reconstruction using true helical scanning geometry parameters, This is for sparse angle reconstruction using simulated sparse scan geometry parameters.

[0034] In S4 or A4, the geometric prior projection is obtained by using the prior information of the real spiral scan geometric parameters and sparse angle CT images to obtain the initial projection value of the missing angle in the sparse angle scan, which is expressed by equation (4):

[0035] ...Equation (4);

[0036] in, For geometric prior projection, To perform 3D front projection under the true helical scanning geometry parameters.

[0037] Projection completion of deep learning models in S5 or A5 The projection completion deep learning model synthesizes the missing intermediate projection by learning the motion field between adjacent angles. The projection completion deep learning model is represented by equation (5):

[0038] ...Equation (5);

[0039] in, To finally complete the projection data, To complete the learnable parameters of a deep learning model for projection.

[0040] In S6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator, while in A6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator and a 3D forward projection operator.

[0041] Preferably, the above-mentioned 3D spiral reconstruction operator transforms the completed projection data from the projection domain to the image domain to obtain the initial restored image data. The 3D spiral reconstruction operator is represented by equation (6).

[0042] ...Equation (6);

[0043] in, To initially recover image data, For 3D spiral reconstruction operators.

[0044] Preferably, the above-mentioned 3D forward projection operator transforms the initial restored image data from the image domain to the projection domain, as represented by equation (7);

[0045] ...Equation (7);

[0046] in, For 3D front projection operator, Projection data obtained by 3D front projection of the initial restored image data. This is for data that needs to be backpropagated from the image domain to the projection domain using gradients.

[0047] In S7 or A7, the image domain deep learning model utilizes sparse angled CT images and projected completed image data for detail restoration and residual artifact recovery. It employs a multi-scale algorithm to extract global image information and utilizes a channel attention module to extract information between different layers to obtain the final reconstructed image. The image domain deep learning model is expressed by equation (8):

[0048] ...Equation (8);

[0049] in, For image domain deep learning models, These are the learnable parameters for deep learning models in the image domain. For the final reconstructed image.

[0050] The loss function described in A8 is represented by equations (9)-(13):

[0051] ...Equation (9);

[0052] ...Equation (10);

[0053] ...Equation (11);

[0054] ...Equation (12);

[0055] ...Equation (13);

[0056] in, For the total loss function, For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights This represents the total number of angles corresponding to the full-angle spiral projection data. This represents the number of images corresponding to a full-angle spiral CT image. To reconstruct the operator.

[0057] In A9, gradient backpropagation involves backpropagating the gradients of learnable parameters to the projected domain. With loss The relevant learnable parameters are defined as follows: , project the domain With loss ,loss and loss The relevant learnable parameters are defined as follows: ;

[0058] In A9, the update of the image domain deep learning model is represented by equation (14):

[0059] ...Equation (14);

[0060] In A9, the update of the projection completion deep learning model is represented by equations (15) and (16):

[0061] ...Equation (14);

[0062] ...Equation (15);

[0063] in, The loss between the current image and the full-angle spiral CT image, and in middle for ,exist middle for ;

[0064] The actual helical scanning geometry parameters in S1 or A1 are the distance from the X-ray source to the detector and the center of rotation, the detector size, the detector element size, the number of exposure angles per revolution, the collimator width, the pitch, and the starting scan angle.

[0065] This invention discloses a sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator, comprising the following steps: S1, acquiring the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data; S2, uniformly sparsely sampling the full-angle spiral projection data obtained in S1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters; S3, reconstructing the full-angle spiral CT image using a weighted filtering back projection algorithm based on the true spiral scanning geometric parameters and full-angle spiral projection data obtained in S1; and reconstructing the sparse angle spiral CT image using a weighted filtering back projection algorithm based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained in S2; S4, reconstructing the sparse angle spiral CT image obtained in S3 using the true spiral scanning geometric parameters obtained in S1. S5. The T-image is subjected to 3D front projection to obtain the corresponding geometric prior projection; S6. The projection data of the missing angles in the geometric prior projection obtained in S4 is recovered by the trained projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in S3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data; S7. The real spiral scanning geometric parameters in S1 and the final completed projection data obtained in S5 are connected across the projection domain to the image domain by the differentiable spiral reconstruction operator to obtain the initial restored image data after projection completion; S8. The sparse angle spiral CT image obtained in S3 and the initial restored image data obtained in S6 are subjected to detail restoration and residual artifact restoration by the trained image domain deep learning model to obtain the final reconstructed image. Compared with the prior art, the beneficial effects of the present invention are:

[0066] 1. By constructing a geometric prior preprojection, the reconstructed sparse angle image is preprojected using the real scanning geometry, providing geometric prior guidance for missing projections; 2. Based on the similarity and redundancy characteristics of adjacent projections in spiral scanning, a projection completion network is constructed. By learning the bidirectional motion field of adjacent angles and combining it with the geometric prior projection, intermediate missing projection data is synthesized; 3. An image domain fine-tuning module is constructed to restore global stripe artifacts in the image; 4. To address the problems of loss gradient not being able to be backpropagated and high computational cost during dual-domain training, on the one hand, a differentiable spiral reconstruction operator with gradient backpropagation is constructed to achieve joint training of the projection domain and the image domain, which is beneficial for overall restoration by combining projection-image dual-domain information; on the other hand, during training, only data collected within two pitches is used for restoration, effectively avoiding excessive computational cost. Attached Figure Description

[0067] 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.

[0068] Figure 1 is a flowchart of a sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator.

[0069] Figure 2 is a flowchart of the joint training process.

[0070] Figure 3 is a comparison of the results before and after reconstruction using the algorithm of this invention. Detailed Implementation

[0071] The technical solution of the present invention will be further described in conjunction with the following embodiments.

[0072] Example 1

[0073] A sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator, as shown in Figure 1, includes the following steps:

[0074] S1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data;

[0075] S2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in S1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters.

[0076] S3. Based on the real spiral scanning geometric parameters and full-angle spiral projection data from S1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from S2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image.

[0077] S4. Using the actual spiral scanning geometric parameters of S1, perform 3D front projection on the sparse angle spiral CT image obtained in S3 to obtain the corresponding geometric prior projection.

[0078] S5. The projection data of the missing angles in the geometric prior projection obtained in S4 are recovered by the trained projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in S3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data, thereby maintaining the consistency of the measurement data before and after recovery. The projection completion deep learning model is constructed based on the similarity and redundancy characteristics between adjacent projections in the spiral scan.

[0079] S6. Using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in S1 and the final completed projection data obtained in S5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion.

[0080] S7. The sparse angle spiral CT image obtained in S3 and the initial restored image data obtained in S6 are processed by the trained image domain deep learning model to restore details and restore residual artifacts, and the final reconstructed image is obtained.

[0081] The trained projection completion deep learning model and the trained image domain deep learning model of this invention are obtained through joint training, wherein each joint training is performed according to the following steps, as shown in Figure 2:

[0082] A1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data;

[0083] A2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in A1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters.

[0084] A3. Based on the actual spiral scanning geometric parameters and full-angle spiral projection data from A1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from A2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image.

[0085] A4. Using the actual spiral scanning geometric parameters of A1, perform 3D front projection on the sparse angle spiral CT image obtained in A3 to obtain the corresponding geometric prior projection.

[0086] A5. The projection data of the missing angles in the geometric prior projection obtained by A4 is recovered by the projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data of A3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data, thereby maintaining the consistency of the measurement data before and after recovery. The projection completion deep learning model is constructed based on the similarity and redundancy characteristics between adjacent projections of spiral scanning.

[0087] A6. By using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in A1 and the final completed projection data obtained in A5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion.

[0088] A7. Using an image domain deep learning model, detail restoration and residual artifact restoration are performed on the sparse angle spiral CT image obtained in A3 and the initial restored image data obtained in A6 to obtain the final reconstructed image.

[0089] A8. Calculate the loss using the loss function, where the final completed projection data obtained from A5 is compared with the full-angle spiral projection data from A1 to obtain the loss. The gradient of the final completed projection data obtained from A5 is compared with the gradient of the full-angle spiral projection data from A1 to obtain the loss. The initial restored image data from A6 was compared with the full-angle spiral CT image from A3 to obtain the loss. The final reconstructed image from A7 was compared with the full-angle spiral CT image from A3 to obtain the loss. ;

[0090] A9. The update of an image domain deep learning model involves applying the loss... Gradient backpropagation is performed directly to the image domain deep learning model to update its parameters; the update of the projection-complete deep learning model involves applying the loss function. and loss The gradient is directly backpropagated to the projection-complete deep learning model, while the loss is... and loss The gradient is backpropagated to the projection-complete deep learning model through the 3D differentiable gradient backpropagation operator, and finally the parameters of the projection-complete deep learning model are updated.

[0091] It should be noted that, in this invention, the training is considered complete when the projection completion deep learning model and the image domain deep learning model converge after multiple training sessions. The convergence condition can be that the number of training sessions reaches a predetermined number, which is determined based on the amount of data and the network size.

[0092] In S2 or A2, uniform sparse sampling specifically involves uniformly sampling the full-angle projection data along the angular direction, as represented by equation (1):

[0093] ...Equation (1);

[0094] in, To simulate sparse angle spiral projection data, This is full-angle spiral projection data. For a uniform sampling process in the angular direction, N is the sparse sampling interval.

[0095] The reconstruction process of the weighted filtering back projection algorithm in S3 or A3 is represented by equations (2) and (3):

[0096] ...Equation (2);

[0097] ...Equation (3);

[0098] in, Full-angle spiral CT images, These are sparse angle spiral CT images. These are the actual geometric parameters for helical scanning. To simulate sparse scan geometry parameters, To achieve full-angle reconstruction using true helical scanning geometry parameters, This is for sparse angle reconstruction using simulated sparse scan geometry parameters.

[0099] It should be noted that, since the sampling rate of sparse angular projection data does not satisfy the Shannon sampling theorem, directly using the traditional filtering back projection algorithm for reconstruction will result in a large number of stripe artifacts and noise in the image, affecting clinical diagnosis. Therefore, this invention uses a weighted filtering back projection algorithm for reconstruction.

[0100] In S4 or A4, the geometric prior projection is obtained by using the prior information of the real spiral scan geometric parameters and sparse angle CT images to obtain the initial projection value of the missing angle in the sparse angle scan, which is expressed by equation (4):

[0101] ...Equation (4);

[0102] in, For geometric prior projection, To perform 3D front projection under the true helical scanning geometry parameters.

[0103] It should be noted that geometric prior projection is helpful in guiding the recovery of missing angles.

[0104] Projection completion of deep learning models in S5 or A5 The projection completion deep learning model synthesizes the missing intermediate projection by learning the motion field between adjacent angles. The projection completion deep learning model is represented by equation (5):

[0105] ...Equation (5);

[0106] in, To finally complete the projection data, The projected data completes the learnable parameters of the deep learning model. The final completed projection data still retains the original measurement data at the actual sparse scan angle positions, thus maintaining the consistency of the data before and after the projection restoration.

[0107] In S6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator, while in A6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator and a 3D forward projection operator.

[0108] The 3D spiral reconstruction operator transforms the completed projection data from the projection domain to the image domain to obtain the initial restored image data. The 3D spiral reconstruction operator is represented by equation (6).

[0109] ...Equation (6);

[0110] in, To initially recover image data, For 3D spiral reconstruction operators,

[0111] The 3D forward projection operator transforms the initial restored image data from the image domain to the projection domain, as expressed by equation (7);

[0112] ...Equation (7);

[0113] in, For 3D front projection operator, Projection data obtained by 3D front projection of the initial restored image data. This is for data that needs to be backpropagated from the image domain to the projection domain using gradients.

[0114] It should be noted that this projection data Because the image has data truncation on the Z-axis, therefore... In contrast, it has data truncation in the direction of the detector channel, but this data truncation has no effect on the image to be reconstructed.

[0115] In S7 or A7, the image domain deep learning model utilizes sparse angled CT images and projected completed image data for detail restoration and residual artifact recovery. It employs a multi-scale algorithm to extract global image information and utilizes a channel attention module to extract information between different layers to obtain the final reconstructed image. The image domain deep learning model is expressed by equation (8):

[0116] ...Equation (8);

[0117] in, For image domain deep learning models, These are the learnable parameters for deep learning models in the image domain. For the final reconstructed image.

[0118] The loss function described in A8 is represented by equations (9)-(13):

[0119] ...Equation (9);

[0120] ...Equation (10);

[0121] ...Equation (11);

[0122] ...Equation (12);

[0123] ...Equation (13);

[0124] in, For the total loss function, For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights This represents the total number of angles corresponding to the full-angle spiral projection data. This represents the number of images corresponding to a full-angle spiral CT image. To reconstruct the operator.

[0125] It should be noted that in A9, gradient backpropagation refers to backpropagating the gradients of learnable parameters, and the projection domain is used to perform gradient backpropagation. Learnable parameters loss Related, the projection domain Learnable parameters With loss ,loss and loss Related. Due to and For learnable parameters, it is necessary to calculate and .

[0126] In A9, the update of the image domain deep learning model is represented by equation (14):

[0127] ...Equation (14);

[0128] In A9, the update of the projection completion deep learning model is represented by equations (15) and (16):

[0129] ...Equation (14);

[0130] ...Equation (15);

[0131] in, The loss between the current image and the full-angle spiral CT image, and in middle for ,exist middle for ;

[0132] The actual helical scanning geometric parameters in S1 or A1 are the distance from the X-ray source to the detector and the center of rotation, the detector size, the detector unit size, the number of exposure angles in one revolution, the collimator width, the pitch, and the starting scanning angle; wherein the helical projection data of the present invention, such as full-angle helical projection data, sparse-angle helical projection data, etc., refers to the projection data after logarithmic transformation.

[0133] Both the projection completion deep learning model and the image domain deep learning model are U-net models.

[0134] This sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator achieves joint training from real spiral projection data to 3D images by constructing a differentiable spiral reconstruction operator. It fully utilizes the information from both domains of the projection image for diagnostic-level image reconstruction. Simultaneously, it leverages the redundant information between adjacent spiral projection data to recover missing angles, fully utilizing the information in the projection data and significantly improving projection recovery performance. Furthermore, this invention can be directly applied to clinical scanning. During clinical spiral scanning, by inputting the scanning parameters and sparse spiral projection data for each case into this invention, the final CT image can be obtained. This invention is directly applicable to clinical CT scanning and is suitable for different scanning conditions and scanning models.

[0135] The sparse spiral CT image reconstruction method based on the differentiable spiral reconstruction operator has the following advantages: 1. By constructing a geometric prior preprojection, the reconstructed sparse angle image is preprojected using the real scan geometry, providing geometric prior guidance for missing projections; 2. Based on the similarity and redundancy characteristics of adjacent projections in spiral scanning, a projection completion network is constructed. By learning the bidirectional motion field of adjacent angles and combining it with the geometric prior projection, intermediate missing projection data is synthesized together; 3. An image domain fine-tuning module is constructed to realize the recovery of global strip artifacts in the image; 4. To address the problems of loss gradient not being able to be backpropagated and large computational load during dual-domain training, on the one hand, a differentiable spiral reconstruction operator with gradient backpropagation is constructed to realize joint training of the projection domain and the image domain, which is beneficial for overall recovery by combining projection-image dual-domain information; on the other hand, during the training process, only data collected within two pitches is used for recovery, effectively avoiding excessive computational load.

[0136] Example 2

[0137] A sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator is provided. Other features are the same as in Example 1. In this example, the actual spiral scanning geometric parameters in S1 and A1 are: the distances from the X-ray source to the rotation center and the detector are 1085.6 mm and 595 mm, respectively; the exposure angle for one revolution is 1152°; the detector size is 736×64; and the detector element size is 1.2858×1.0947.

[0138] In S2 and A2, uniform sparse sampling is performed using a sparse factor of 1 / 6 (192 angles, N=6). In S4 and A4, to balance memory usage and network performance, two-ring projection data is used for restoration, i.e., sparse projection data 384*736*64 is completed into full-angle projection data 2304*736*64. During model training, a strategy of interpolating the missing frame by projecting between two adjacent frames is adopted. Due to the large amount of data, the data is adjusted to 384*1*736*64 during training, and the data of different adjacent frames are restored in parallel to reduce the amount of computation and achieve spiral projection completion, laying the foundation for subsequent reconstruction.

[0139] In S6 and A6, the final size of the completed projection data is 32*512*512. In S7 and A7, in order to reduce the number of parameters of the image domain deep learning model during image domain training, this embodiment selects a 2.5D training strategy, that is, placing the data of different slices in the channel direction.

[0140] Figure 3 shows the abdominal transverse and coronal sectional images from a full-angle spiral CT scan (S3), the second column shows the abdominal transverse and coronal sectional images from a sparse-angle (N=6) spiral CT scan (S3), and the third column shows the abdominal transverse and coronal sectional images reconstructed using the method of this invention (N=6). As can be seen from Figure 3, the transverse and coronal sections of the sparse-angle spiral CT images contain significant noise and artifacts, leading to image distortion. These artifacts overlap and obscure the real tissue structures, affecting the doctor's diagnosis. After processing using the method of this invention, the images in Figure 3 show significant changes. The corrected images effectively remove a large amount of noise and stripe artifacts from the sparse-angle spiral CT images, making the detailed tissue structures clearly visible.

[0141] 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 sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator, characterized in that, Includes the following steps: S1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data; S2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in S1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters. S3. Based on the real spiral scanning geometric parameters and full-angle spiral projection data from S1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from S2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image. S4. Using the actual spiral scanning geometric parameters of S1, perform 3D front projection on the sparse angle spiral CT image obtained in S3 to obtain the corresponding geometric prior projection. S5. The projection data of the missing angles in the geometric prior projection obtained in S4 is recovered by the trained projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in S3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data. S6. Using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in S1 and the final completed projection data obtained in S5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion. S7. The sparse angle spiral CT image obtained in S3 and the initial restored image data obtained in S6 are processed by the trained image domain deep learning model to restore details and restore residual artifacts, and the final reconstructed image is obtained.

2. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 1, characterized in that: The trained projection completion deep learning model and the trained image domain deep learning model are obtained through joint training, with each joint training session performed according to the following steps: A1. Obtain the true spiral scanning geometric parameters of the object and the corresponding full-angle spiral projection data; A2. Perform uniform sparse sampling on the full-angle spiral projection data obtained in A1 to obtain simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters. A3. Based on the actual spiral scanning geometric parameters and full-angle spiral projection data from A1, a weighted filtering back projection algorithm is used to reconstruct the full-angle spiral CT image; and based on the simulated sparse angle spiral projection data and simulated sparse scanning geometric parameters obtained from A2, a weighted filtering back projection algorithm is used to reconstruct the sparse angle spiral CT image. A4. Using the actual spiral scanning geometric parameters of A1, perform 3D front projection on the sparse angle spiral CT image obtained in A3 to obtain the corresponding geometric prior projection. A5. The projection data of the missing angles in the geometric prior projection obtained in A4 is recovered by the projection completion deep learning model to obtain the initial completed projection data. Then, the simulated sparse angle spiral projection data in A3 is used to replace the corresponding positions in the completed projection data to obtain the final completed projection data. A6. By using the differentiable spiral reconstruction operator, the real spiral scanning geometric parameters in A1 and the final completed projection data obtained in A5 are connected across the projection domain to the image domain to obtain the initial restored image data after projection completion. A7. Using an image domain deep learning model, detail restoration and residual artifact restoration are performed on the sparse angle spiral CT image obtained in A3 and the initial restored image data obtained in A6 to obtain the final reconstructed image. A8. Calculate the loss using the loss function, where the final completed projection data obtained from A5 is compared with the full-angle spiral projection data from A1 to obtain the loss. The gradient of the final completed projection data obtained from A5 is compared with the gradient of the full-angle spiral projection data from A1 to obtain the loss. The initial restored image data from A6 was compared with the full-angle spiral CT image from A3 to obtain the loss. The final reconstructed image from A7 was compared with the full-angle spiral CT image from A3 to obtain the loss. ； A9. The update of an image domain deep learning model involves applying the loss... Gradient backpropagation is performed directly to the image domain deep learning model to update its parameters; the update of the projection-complete deep learning model involves applying the loss function. and loss The gradient is directly backpropagated to the projection-complete deep learning model, while the loss is... and loss The gradient is backpropagated to the projection-complete deep learning model through the 3D differentiable gradient backpropagation operator, and finally the parameters of the projection-complete deep learning model are updated.

3. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 2, characterized in that: In S2 or A2, uniform sparse sampling specifically involves uniformly sampling the full-angle projection data along the angular direction, as represented by equation (1): ...Equation (1); in, To simulate sparse angle spiral projection data, This is full-angle spiral projection data. For a uniform sampling process in the angular direction, N is the sparse sampling interval.

4. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 3, characterized in that, The reconstruction process of the weighted filtering back projection algorithm in S3 or A3 is represented by equations (2) and (3): ...Equation (2); ...Equation (3); in, Full-angle spiral CT images, These are sparse angle spiral CT images. These are the actual geometric parameters for helical scanning. To simulate sparse scan geometry parameters, To achieve full-angle reconstruction using true helical scanning geometry parameters, This is for sparse angle reconstruction using simulated sparse scan geometry parameters.

5. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 4, characterized in that: In S4 or A4, the geometric prior projection is obtained by using the prior information of the real spiral scan geometric parameters and sparse angle CT images to obtain the initial projection value of the missing angle in the sparse angle scan, which is expressed by equation (4): ...Equation (4); in, For geometric prior projection, To perform 3D front projection under the true helical scanning geometry parameters.

6. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 5, characterized in that: Projection completion of deep learning models in S5 or A5 The projection completion deep learning model synthesizes the missing intermediate projection by learning the motion field between adjacent angles. The projection completion deep learning model is represented by equation (5): ...Equation (5); in, To finally complete the projection data, To complete the learnable parameters of a deep learning model for projection.

7. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 6, characterized in that: In S6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator, while in A6, the differentiable spiral reconstruction operator consists of a 3D spiral reconstruction operator and a 3D forward projection operator. The 3D spiral reconstruction operator transforms the completed projection data from the projection domain to the image domain to obtain the initial restored image data. The 3D spiral reconstruction operator is represented by equation (6). ...Equation (6); in, To initially recover image data, For 3D spiral reconstruction operators, The 3D forward projection operator transforms the initial restored image data from the image domain to the projection domain, as expressed by equation (7); ...Equation (7); in, For 3D front projection operator, Projection data obtained by 3D front projection of the initial restored image data. This is for data that needs to be backpropagated from the image domain to the projection domain using gradients.

8. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 7, characterized in that: In S7 or A7, the image domain deep learning model utilizes sparse angled CT images and projected completed image data for detail restoration and residual artifact recovery. It employs a multi-scale algorithm to extract global image information and utilizes a channel attention module to extract information between different layers to obtain the final reconstructed image. The image domain deep learning model is expressed by equation (8): ...Equation (8); in, For image domain deep learning models, These are the learnable parameters for deep learning models in the image domain. For the final reconstructed image.

9. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 8, characterized in that: The loss function described in A8 is represented by equations (9)-(13): ...Equation (9); ...Equation (10); ...Equation (11); ...Equation (12); ...Equation (13); in, For the total loss function, For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights For loss The corresponding weights This represents the total number of angles corresponding to the full-angle spiral projection data. This represents the number of images corresponding to a full-angle spiral CT image. To reconstruct the operator.

10. The sparse spiral CT image reconstruction method based on a differentiable spiral reconstruction operator according to claim 9, characterized in that: In A9, the update of the image domain deep learning model is represented by equation (14): ...Equation (14); In A9, the update of the projection completion deep learning model is represented by equations (15) and (16): ...Equation (14); ...Equation (15); in, The loss between the current image and the full-angle spiral CT image, and in middle for ,exist middle for ； The actual helical scanning geometry parameters in S1 or A1 are the distance from the X-ray source to the detector and the center of rotation, the detector size, the detector element size, the number of exposure angles per revolution, the collimator width, the pitch, and the starting scan angle.

Images