Multi-source CT reconstruction method based on cone beam-parallel beam rearrangement

Abstract

The invention discloses a multi-source spiral CT (Computed Tomography) analysis and reconstruction method based on cone beam-parallel beam rearrangement, which comprises the following steps of: firstly, carrying out denoising processing on high-noise projection data, and effectively inhibiting noise and reserving edge information by adopting a strategy of combining bilateral filtering with residual image and bilateral filtering; then, rearranging the spiral cone beam projection data of the multi-source CT into three-dimensional parallel beam projection by using a cone beam-parallel beam rearrangement algorithm; then, parallel beam filtering is carried out on the rearranged projection data; and finally, carrying out cone angle weighted back projection on the filtered parallel beam projection to obtain an analytic reconstruction image of the multi-source CT. According to the method, a rapid analysis reconstruction algorithm for the multi-source CT is realized, the reconstruction problem of discontinuous scanning geometry of the multi-source CT is solved, the reconstruction efficiency is remarkably improved, and excessive calculation consumption is avoided. According to the method, image quality and details are guaranteed, artifacts and noise are suppressed, an efficient and convenient solution is provided for clinical application, and the method has high universality and is suitable for data processing requirements of different equipment manufacturers.

Description

Technical Field

[0001] The present invention relates to a multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement, and belongs to the technical field of computer image processing. Background Art

[0002] With the continuous development of computed tomography (CT) technology in recent years, CT equipment has played an increasingly important role in medical imaging. Although CT technology has made remarkable progress in the past few decades, with the increasing demand for higher imaging quality, CT technology is still in a stage of rapid development, especially in reducing radiation dose, reducing artifacts, and improving spatial and temporal resolutions.

[0003] For some clinically important application scenarios, such as lung imaging, cardiac imaging, and CT perfusion imaging, which have relatively high requirements for the temporal resolution of the equipment. If high-temporal-resolution data cannot be obtained in time, inevitable motion artifacts will appear in the reconstructed images, affecting the clinical diagnosis effect. Current CT equipment mainly uses a rotating gantry for scanning, where the X-ray source and detector are installed on the rotating gantry, and projection data at different angles are collected by rotating the gantry. Theoretically, high-temporal-resolution data can be achieved by accelerating the rotation speed of the gantry. However, when the rotation speed is too fast, a strong centrifugal force will be generated, which may exceed the load limit of mechanical equipment and damage the stability of the equipment. Therefore, the temporal resolution of current commercial CTs is still limited by the hardware bottleneck of the rotating gantry. Although current CT equipment has made certain optimizations in temporal resolution, the scanning speed of a single-source CT equipment has been optimized to 3 Hz. However, this temporal resolution is still insufficient for imaging moving organs (such as the heart and lungs), resulting in a decline in image quality. In addition, the movement of the X-ray source focus in the direction of motion may further exacerbate the problem of image blurring.

[0004] To address these issues, researchers have further developed multi-source CT. Multi-source CT technology mainly uses multiple X-ray sources and detectors to simultaneously acquire data from different angles, significantly improving the scanning speed and imaging quality, reducing motion artifacts, and decreasing the need for contrast agents. Compared with single-source CT, multi-source CT shows greater potential in aspects such as scanning speed, temporal resolution, and spectral imaging capabilities. For example, Dual-Source CT (DSCT) uses two X-ray sources and two detectors, usually arranged at a 90° angle. Dual-energy imaging can also be achieved through different X-ray tube voltages (such as 80 kVp and 140 kVp), which can be used for material separation (such as differentiating bone and soft tissue). Multi-source Inverse Geometry CT (MS-IGCT) uses distributed X-ray sources and a relatively small detector array. Each source in the transverse direction emits a narrow X-ray beam through a part of the field of view (FOV), enabling the system to achieve uniform spatial resolution within the FOV, while additional X-ray sources in the longitudinal direction can increase the volume coverage and reduce cone-beam artifacts.

[0005] Although multi-source CT has made certain breakthroughs in temporal resolution, it also faces other challenges. For example, the data volume collected by multi-source CT is extremely large, and the data transmission, storage, and reconstruction computational requirements are extremely high, requiring powerful computing resources. The traditional FDK algorithm cannot be directly used for multi-source CT, while the iterative reconstruction algorithm has a high computational cost and a long reconstruction time, affecting clinical efficiency. The X-ray sources of multi-source CT are discretely distributed, resulting in insufficient sampling in some directions, making it difficult to directly apply traditional CT reconstruction methods. Trajectory non-uniformity may lead to artifacts and affect image quality. Therefore, how to achieve reconstruction while improving temporal resolution and solving image quality and artifact problems remains the main challenge for multi-source CT technology. Summary of the Invention

[0006] To solve the problem that the discontinuous trajectory of multi-source CT makes it impossible to directly reconstruct using the traditional cone-beam FDK algorithm and the iterative algorithm consumes too much time, the present invention provides a multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement, achieving efficient analytical reconstruction of multi-source CT.

[0007] To achieve the above objective, the present invention provides the following technical solution. A multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement includes the following steps:

[0008] Step 1: Obtain the high-noise projection data P of multi-source CT obtained from clinical scans, and perform preliminary noise reduction on the high-noise projection data P using bilateral filtering to obtain the preliminary noise-reduced projection P1;

[0009] Step 2: Subtract the high-noise projection data P in Step 1 from the preliminary noise-reduced projection P1 to obtain the noise estimation projection P n, and then using the preliminary noise-reduced projection P1 as a guiding image to extract the residual structure in the noise estimation projection P n to obtain the residual structure projection P r ;

[0010] Step 3: Use the residual structure projection P r obtained in Step 2 to fuse the preliminary noise-reduced projection P1 obtained in Step 1, restore the lost structure, and obtain the final denoised projection image P denoise ;

[0011] Step 4: Use the cone-beam to parallel-beam rearrangement algorithm to rearrange the denoised projection image P denoise obtained in Step 3 to obtain the three-dimensional parallel-beam projection P rebin ;

[0012] Step 5: Use parallel-beam filtering to filter the sinogram P rebin obtained in Step 4 to obtain the filtered parallel-beam projection P filtered ;

[0013] Step 6: Use the filtered parallel-beam projection P filtered to perform back-projection, determine the back-projection rays according to the reconstruction point positions, and perform conical angle weighted integration on the projections. Finally, obtain the analytical reconstruction result X of the multi-source CT recon .

[0014] Furthermore, in Step 1, a bilateral filter is used to suppress the noise of the clinical high-noise projection data P

[0015] to obtain the preliminary noise-reduced projection P1, as shown in Equation 1:

[0016] P1 = Bifilter(P) (1)

[0017] Furthermore, in Step 2, the following specific process is included: The high-noise projection data P and the preliminary noise-reduced projection P1 are subtracted to obtain the noise estimation projection P removed by the bilateral filter n :

[0018] P n = P - P1 (2)

[0019] To avoid the over-smoothing of the projection data by the bilateral filter resulting in the weakening of the projection structure edges, using the preliminary noise-reduced projection P1 as a guiding image, use the joint bilateral filter to extract the structure in the noise estimation projection P n to obtain the residual structure projection P r :

[0020] P r = JointBifilter(P n, P1) (3)

[0021] Further, in step 3, project the residual structure P r and the preliminary noise reduction projection P1 are fused to obtain the denoising projection image P with final noise suppression denoise :

[0022] P denoise = P r + P1 (4)

[0023] Further, step 4 specifically includes the following process:

[0024] (1) First, the transformation relationship between the cone-beam projection coordinate system and the parallel-beam projection is shown in formulas 5 and 6, and the symbol meanings are as Figure 2 shown:

[0025] θ = α + β (5)

[0026] t = SOD·sin(β) (6)

[0027] where (θ, t, v) represents the coordinates in the parallel-beam geometry, (α, u, v) corresponds to the cone-beam geometry coordinates, SOD represents the distance from the ray source to the rotation center, β represents the angle between the interpolated ray and the line connecting the ray source and the rotation center in the cone-beam geometry, α represents the angle between the ray source and the x-axis, u represents the detector coordinate of the ray in the x-y plane, and v represents the detector coordinate perpendicular to the x-y plane.

[0028] (2) In addition, for the cone-beam to parallel-beam geometry, there are additional geometric transformation relationships, as shown in formulas 7 and 8:

[0029]

[0030] u = ODD·(γ + β) (8)

[0031] where ODD represents the distance from the rotation center to the detector, and γ represents the angle between the line connecting the detector intersection and the rotation center of the interpolated ray and the interpolated ray.

[0032] (3) The relationship between the final cone-beam geometry coordinates and the parallel-beam geometry coordinates can be simplified as shown in formulas 9 and 10:

[0033]

[0034] The position of each location in the rearranged parallel-beam sinogram is determined using formulas 9 and 10 to obtain the position under the corresponding cone-beam projection, and then the corresponding value is obtained using the linear interpolation algorithm, and finally the rearranged three-dimensional parallel-beam projection P rebin .

[0035] Further, step 5 specifically includes the following process: For the rearranged projection P rebin perform filtering, that is, in the (θ, t, v) coordinate system, filter in the t direction. This filtering process is implemented in the form of image domain convolution, and the filter kernel used is the Ramp kernel filter. After filtering, the filtered parallel beam projection P filtered is obtained:

[0036] P filtere d(θ, t, v) = ∫P rebin (θ, τ, v) · h ramp (t - τ)dτ (11)

[0037] where h ramp represents the Ramp filter kernel.

[0038] Further, step 6 specifically includes the following process: Use the filtered parallel beam projection P filtered to perform back-projection. For a certain reconstruction point (x, y, z), for the parallel beam projection with an angle of passing through this point in the parallel beam coordinate system, there is the following geometric relationship: There is the following geometric relationship:

[0039]

[0040] The corresponding can be calculated according to formula 6.

[0041] In the X - O - Y plane, the distance l from the ray source to the reconstruction point (x, y, z) can be obtained through the following geometric relationship:

[0042]

[0043] Similarly, in the X - O - Y plane, the distance l from the ray source passing through the reconstruction point (x, y, z) to the detector can be obtained through formula 14: e can be obtained through formula 14:

[0044]

[0045] Through the distances l and l, the longitudinal position of the pixel on the detector where the ray source passes through the reconstruction point (x, y, z) hits can be further determined: can be further determined:

[0046]

[0047] where z source represents the z - axis coordinate of the ray source corresponding to the ray represented by in the parallel beam geometric coordinates. Thus, we can determine the coordinate position of the back - projection ray in the rearranged parallel beam projection matrix.

[0048] To alleviate the precision error caused by large cone angle rays in analytical reconstruction, we designed a cone angle weighting coefficient w q to balance the influence of rays with different cone angles at the same angle on the same reconstructed pixel. Its calculation method is shown in Equation 16:

[0049]

[0050] where represents the ratio of the distance of the intersection point of the ray on the detector along the Z-axis to the maximum distance v of the detector along the Z-axis. When q is greater than the predetermined threshold ratio Q, the weight w max is used to reduce the contribution of large cone angle rays to the reconstruction result. The predetermined threshold ratio Q is set to 0.9. q Finally, the back-projection algorithm performs integration based on cone angle weighted sparsity on the parallel beam data passing through the target reconstruction point (x, y, z) at all angles:

[0051] In the above formula, i represents the serial number of the light source,

[0052]

[0053] is the projection angle, and its value range is [0, π]. By adding kπ (k is an integer), the projection angles of the entire helical scan are traversed; is the cone angle weighting of the inclined ray; represents the sum of the cone angle weighting coefficients of the parallel beams of all ray sources and all angles for the reconstructed pixel (x, y, z). represents the cone angle weighted coefficient sum of the parallel beams of the reconstructed pixel (x, y, z) at all angles under the i-th ray source. The integration and summation operations combine the multi-ray source and the multi-turn scan data of the same ray source. Each ray source uses the weighted and filtered projection data to achieve 3D reconstruction. is the filtered parallel beam projection at the angle of the i-th ray source. The integration and summation operations combine the multi-ray source and the multi-turn scan data of the same ray source. Each ray source uses the weighted and filtered projection data to achieve 3D reconstruction.

[0054] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the multi-source CT reconstruction method based on cone beam - parallel beam rearrangement described above.

[0055] A computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, they implement the multi-source CT reconstruction method based on cone beam - parallel beam rearrangement described above.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] 1. The present invention effectively suppresses noise while retaining edge information through the combination of bilateral filtering and joint bilateral filtering in the residual domain, preventing significant degradation of the spatial resolution of the reconstructed image due to over-smoothing in projection denoising.

[0058] 2. The present invention rearranges the three-dimensional cone-beam projection into a three-dimensional parallel-beam projection and uses a parallel-beam filtered back-projection algorithm based on cone-angle weighting to solve the potential image non-uniformity problem caused by direct reconstruction of the discontinuous scanning trajectory of a new multi-source CT, suppresses artifacts and noise to a certain extent, realizes efficient analytical reconstruction under the static CT geometry, and avoids the large amount of computational consumption required by iterative reconstruction algorithms.

[0059] 3. The reconstructed image obtained by the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement of the present invention can be used as the initial value of an iterative algorithm or a learning-based reconstruction algorithm, provides a stable prior image for further suppressing artifacts such as sparse angles and helices in multi-source CT reconstruction, and provides new ideas for subsequent multi-source CT reconstruction research.

[0060] 4. The present invention adopts a parallel-beam filtered back-projection algorithm based on cone-angle weighting. Compared with traditional iterative reconstruction methods, it avoids complex matrix operations and a large amount of computational resource consumption, significantly improves the computational speed of the reconstruction process, and meets the real-time requirements for processing large-scale medical image data.

[0061] 5. While ensuring high-quality image reconstruction, the method of the present invention is simple in structure, easy to implement, does not rely on complex regularization models or deep learning networks, reduces the complexity of algorithm parameter tuning, reduces the engineering implementation cost, and improves the operability in clinical applications.

[0062] 6. By means of cone-beam to parallel-beam rearrangement, the present invention improves the utilization rate of projection data, enables high-quality reconstructed images to be obtained even under sparse sampling conditions, reduces reconstruction artifacts, increases the utilization efficiency of projection information, and improves the image signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is the overall flowchart of the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement provided by the present invention.

[0064] Figure 2 is the schematic diagram of the geometric transformation of cone-beam to parallel-beam rearrangement of multi-source CT provided by the present invention.

[0065] Figure 3 is the reconstructed result diagram of the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement provided by the present invention on a head phantom.

[0066] In the figure: Figures (a)-(d) show four different tomograms of the head phantom reconstruction results. In this embodiment, 24 ray sources are arranged in a circular pattern, the rotation angle of each ray source per rotation is 0.5 degrees, the object-to-detector distance (ODD) is 434.75 mm, and the source-to-object distance (SOD) is 709 mm. Each ray source rotates through 284 angles, and the normalized pitch value of each ray source is 9.6735. Detailed implementation mode

[0067] The technical solution provided by the present invention will be described in detail below in conjunction with specific embodiments. It should be understood that the following specific implementation modes are only used to illustrate the present invention and not to limit the scope of the present invention.

[0068] Embodiment: The present invention provides a multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement. First, high-noise projection data of multi-source CT is obtained from clinical scans. Due to the limitation of the scanning dose, the reconstructed images have severe noise. Therefore, bilateral filtering is used to perform preliminary noise reduction on the high-noise projection data. Then, by calculating the difference between the high-noise projection and the preliminarily noise-reduced projection, a noise estimation projection is obtained. Subsequently, joint bilateral filtering is used to extract the residual structure in the noise estimation projection to obtain a residual structure projection. Next, the residual structure projection is fused with the preliminarily noise-reduced projection to restore the lost structure and obtain a final denoised projection image. Subsequently, the cone-beam to parallel-beam rearrangement algorithm is used to rearrange the denoised projection image, and the projection interpolation in the three-dimensional cone-beam coordinate system is transformed into the projection in the corresponding three-dimensional parallel-beam coordinate system. The rearranged projection is filtered by parallel-beam filtering, and finally, a multi-source CT analytical reconstruction image is obtained using the cone-angle weighted back-projection algorithm.

[0069] Specifically, a multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement. The overall process of the present invention is as shown in the appendix Figure 1 and includes the following steps:

[0070] Step 1: Obtain the high-noise projection data P of multi-source CT obtained from clinical scans, and use a bilateral filter to suppress the noise of the clinical high-noise projection data P to obtain a preliminarily noise-reduced projection P1, as shown in Equation 1:

[0071] P1 = Bifilter(P) (1)

[0072] Step 2: Subtract the high-noise projection data P from the preliminarily noise-reduced projection P1 to obtain the noise estimation projection P n :

[0073] P n = P - P1 (2)

[0074] To avoid the over - smoothing of the bilateral filter on the projection data, which may lead to the weakening of the projection structure edges, the preliminary denoised projection P1 is used as the guiding image, and the joint bilateral filter is used to extract the structure in the noise - estimated projection P n to obtain the residual structure projection P r :

[0075] P r = JointBifilter(P n , P1) (3)

[0076] Step 3: Using the residual structure projection P r obtained in Step 2 to fuse with the preliminary denoised projection P1 obtained in Step 1, restore the lost structure, and obtain the final denoised projection image P denoise :

[0077] P denoise = P r + P1 (4)

[0078] Step 4: Using the cone - beam to parallel - beam rearrangement algorithm to rearrange the denoised projection image P denoise obtained in Step 3:

[0079] (1) First, the transformation relationship between the cone - beam projection coordinate system and the parallel - beam projection is shown in Formulas 5 and 6, and the symbol meanings are as Figure 2 shown:

[0080] θ = α + β (5)

[0081] t = SOD·sin(β) (6)

[0082] where (θ, t, v) represents the coordinates in the parallel - beam geometry, (α, u, v) corresponds to the cone - beam geometry coordinates, SOD represents the distance from the ray source to the rotation center, β represents the angle between the interpolated ray and the line connecting the ray source and the rotation center in the cone - beam geometry, α represents the angle between the ray source and the x - axis, u represents the detector coordinate of the ray in the x - y plane, and v represents the detector coordinate perpendicular to the x - y plane.

[0083] (2) In addition, for the cone - beam to parallel - beam geometry, there are additional geometric transformation relationships, as shown in Formulas 7 and 8:

[0084]

[0085] u = ODD·(γ + β) (8)

[0086] where ODD represents the distance from the rotation center to the detector, and γ represents the angle between the line connecting the detector intersection of the interpolated ray and the rotation center and the interpolated ray.

[0087] (3) The relationship between the final cone-beam geometric coordinates and the parallel-beam geometric coordinates can be simplified as shown in Formulas 9 and 10:

[0088]

[0089] For each position in the rearranged parallel-beam sinogram, the position under the corresponding cone-beam projection is determined using Formulas 9 and 10. Subsequently, the corresponding value is obtained using the linear interpolation algorithm, and finally, the rearranged three-dimensional parallel-beam projection P is obtained. rebin .

[0090] Step 5: Filter the rearranged parallel-beam projection P rebin by using a Ramp filter, that is, filter in the t direction in the (θ, t, v) coordinate system. This filtering process is implemented in the form of image-domain convolution. After filtering, the filtered parallel-beam projection P is obtained: filtered :

[0091] P filtered (θ, t, v) = ∫P re bi n (θ, τ, v) · h ramp (t - τ) dτ (11)

[0092] where h ramp represents the Ramp filter kernel.

[0093] Step 6: Perform back-projection using the filtered parallel-beam projection P filtered . For a certain reconstruction point (x, y, z), for the parallel-beam projection with an angle of , the following geometric relationship exists for the point passed by the parallel-beam projection in the parallel-beam coordinate system: There is the following geometric relationship:

[0094]

[0095] The corresponding can be calculated according to Formula 6.

[0096] In the X-O-Y plane, the distance l from the ray source to the reconstruction point (x, y, z) can be obtained through the following geometric relationship:

[0097]

[0098] Similarly, in the X-O-Y plane, the distance l from the ray source passing through the reconstruction point (x, y, z) to the detector can be obtained through Formula 14: e

[0099]

[0100] Through the distance l e ​and l, the vertical position of the pixel where the ray source passes through the reconstruction point (x, y, z) and hits the detector It can be further determined that:

[0101]

[0102] where z source The parallel beam geometry is represented by The z-axis coordinate of the ray source corresponding to the ray represented, we can thus determine the coordinate position of the back-projected ray in the parallel beam projection matrix after rearrangement

[0103] In order to alleviate the accuracy error caused by large cone angle light to analytical reconstruction, we designed the cone angle weighting coefficient w q To balance the impact of light with different cone angles at the same angle on the same reconstructed pixel, the calculation method is shown in formula 16:

[0104]

[0105] in Indicates the distance between the intersection point of the light on the detector and the maximum distance v on the detector Z axis max When q is greater than the predetermined threshold ratio Q, the weight w q It is used to reduce the contribution of large cone-angle light to the reconstruction result, and the predetermined threshold ratio Q is set to 0.9.

[0106] The final back-projection algorithm performs cone-angle weighted sparse integration on the parallel beam data passing through the target reconstruction point (x, y, z) at all angles:

[0107]

[0108] In the above formula, i represents the serial number of the light source. is the projection angle, the value range is [0,π], and the projection angle of the entire spiral scan is traversed by adding kπ (k is an integer); Weighting for cone angle of oblique rays; Indicates that all ray sources, all The sum of the cone-angle weighting coefficients of the parallel bundles that reconstruct the pixel (x,y,z) at the angle. Indicates that the i-th ray source is The integration and summation operations combine the data from multiple ray sources and multiple scans from the same ray source. Each ray source is weighted and filtered through the projection data. to achieve three-dimensional reconstruction.

[0109] To verify the effectiveness of the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement disclosed in the present invention, the head phantom data obtained from clinical scans is used to verify the effectiveness of the disclosed method for the reconstruction problems in the current multi-source CT scenario. Attached Figure 3 Figure 3 shows the three-dimensional reconstruction results finally obtained by using the multi-source CT analytical reconstruction method of the present invention. The experimental results show that the method of the present invention can effectively reconstruct high-quality multi-source CT images. Especially when dealing with high-noise data, it can significantly suppress artifacts and retain detailed structures. This method successfully solves the problem of image reconstruction under the non-continuous scanning trajectory of multi-source CT, provides a new technical solution for the efficient reconstruction of multi-source CT images, has important clinical application value, and has profound significance for subsequent research related to multi-source CT.

[0110] The technical means disclosed in the solution of the present invention are not limited to the technical means disclosed in the above embodiments, and also include technical solutions composed of any combination of the above technical features. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement, characterized in that, Including the following steps: Step 1: Obtain the multi-source CT high-noise projection data P obtained from clinical scans, and perform preliminary noise reduction on the high-noise projection data P using bilateral filtering to obtain the preliminary noise-reduced projection P1; Step 2: Subtract the high-noise projection data P in Step 1 from the preliminary noise-reduced projection P1 to obtain a noise estimation projection P n , and then use the preliminary noise-reduced projection P1 as a guiding image to extract the residual structure in the noise estimation projection P n to obtain a residual structure projection P r ; Step 3: Project the residual structure P obtained in Step 2 r to fuse with the preliminary denoised projection P1 obtained in Step 1, restore the lost structure, and obtain the final denoised projection image P denoise ; Step 4: Use the cone-beam to parallel-beam rearrangement algorithm to rearrange the denoised projection image P obtained in Step 3 denoise to obtain a three-dimensional parallel-beam projection P rebin ; Step 5, filter the sinogram P obtained in Step 4 using parallel beam filtering rebin to obtain the filtered parallel beam projection P filtered ; Step 6: Using the filtered parallel beam projection P filtered perform back-projection, determine the back-projection rays according to the reconstructed point positions, and perform conical angle weighted integration on the projection, and finally obtain the analytical reconstruction result X of the multi-source CT recon .

2. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, wherein In the said Step 1, a bilateral filter is used to suppress the noise of the clinical high-noise projection data P to obtain the preliminary noise-reduced projection P1, as shown in Equation 1: P1 = Bifilter(P) (1).

3. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, wherein In step 2, the specific process is as follows: The high-noise projection data P and the preliminary denoised projection P1 are subtracted to obtain the noise estimation projection P removed by the bilateral filter n : P n = P - P1 (2), To avoid the over - smoothing of the projection data by bilateral filtering, which may lead to the weakening of the edges of the projection structure, the preliminary denoised projection P1 is used as the guiding image, and the joint bilateral filtering is used to extract the structure in the noise - estimated projection P n to obtain the residual structure projection P r : P r = JointBifilter(P n , P1) (3).

4. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, wherein The specific process of step 3 is as follows: project the residual structure P r and the preliminary noise reduction projection P1 are fused to obtain the denoising projection image P with final noise suppression denoise : P denoise = P r + P1(4).

5. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, wherein The specific process of the said Step 4 is as follows: (1) First, the transformation relationships between the cone-beam projection coordinate system and the parallel-beam projection are as shown in Equations 5 and 6: θ = α + β (5) t = SOD·sinβ (6) Wherein, (θ, t, v) represents the coordinates in the parallel-beam geometry, (α, u, v) corresponds to the cone-beam geometry coordinates, SOD represents the distance from the ray source to the rotation center, β represents the angle formed by the interpolated ray and the line connecting the ray source and the rotation center in the cone-beam geometry, α represents the angle between the ray source and the x-axis, u represents the detector coordinate of the ray in the x-y plane, and v represents the detector coordinate perpendicular to the x-y plane; (2) Assume that the detector is a cylindrical surface centered on the rotation center. For the cone-beam to parallel-beam geometry, the relationship between the horizontal direction coordinate u of the detector and t is as shown in Equations 7 and 8: u = ODD(γ + β) (8) Where ODD represents the distance from the rotation center to the detector, and γ represents the angle formed by the line connecting the intersection point of the interpolated ray on the detector and the rotation center and the interpolated ray; (3) Finally, the relationship between the cone-beam geometry coordinates and the parallel-beam geometry coordinates is simplified as shown in Equations 9 and 10: (4) Determine the position in the corresponding cone-beam projection for each position in the rearranged parallel-beam sinogram using Formulas 9 and 10, then obtain the corresponding value using the linear interpolation algorithm, and finally obtain the rearranged three-dimensional parallel-beam projection P rebin .

6. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, characterized in that Step 5 specifically includes the following process: For the re-arranged projection P rebin perform filtering, that is, in the (θ, t, v) coordinate system, perform filtering in the t direction. This filtering process is implemented in the form of image domain convolution. The filter kernel used is the Ramp kernel filter. After filtering, the filtered parallel beam projection P filtered is obtained: P filtered (θ,t,v) = ∫P rebin (θ,τ,v)·h ramp (t - τ)dτ (11) where h ramp represents the Ramp filter kernel.

7. The multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement according to claim 1, wherein, The specific process of step 6 is as follows: Using the filtered parallel beam projection P filtered to perform backprojection, with the angle being of the parallel beam projection passing through this point in the parallel beam coordinate system There is the following geometric relationship: corresponding Calculated according to Formula 6 In the X-O-Y plane, the distance l from the ray source to the reconstruction point (x, y, z) is obtained through the following geometric relationship: In the X-O-Y plane, assume that the detector is a cylindrical surface centered at the rotation center, and the distance l from the ray source passing through the reconstruction point (x, y, z) to the detector pixel e Obtained by formula 14: Through the distance l e and l, the longitudinal position of the pixel of the detector irradiated by the radiation source passing through the reconstruction point (x, y, z) is further determined as follows: where z source represents the z-axis coordinate of the ray source corresponding to the ray represented in the parallel beam geometry coordinates, so as to determine the coordinate position of the back-projected ray in the rearranged parallel beam projection matrix ​ Design cone angle weighting coefficient w q To balance the influence of light rays with different cone angles at the same angle on the same reconstructed pixel, its calculation method is shown in Equation 16: Among them represents the ratio of the distance of the intersection point where the light beam hits the detector on the Z-axis to the maximum distance v of the detector on the Z-axis. When q is greater than the predetermined threshold ratio Q, the weight w max is used to reduce the contribution of the large cone angle light beam to the reconstruction result, and the predetermined threshold ratio Q is set to 0.9; q ​ The final back-projection algorithm needs to consider the projection data of all light sources passing through the target reconstruction point (x, y, z) at all angles, and at the same time consider the weighting in the cone angle direction: In the above formula, i represents the serial number of the light source, is the projection angle, and its value range is [0, π]. By adding kπ (k is an integer), the projection angles of the entire helical scan are traversed; is the cone angle weighting of the inclined light; represents all ray sources, all the sum of the cone angle weighting coefficients of the parallel beam for reconstructing the pixel (x, y, z) at all angles, represents the filtered parallel beam projection of the i-th ray source at angle. The integration and summation operations combine the multi-ray source and the multi-turn scan data of the same ray source. Each ray source uses the weighted and filtered projection data to achieve three-dimensional reconstruction.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein: When the processor executes the said program, it implements the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement as described in any one of the above claims 1 to 7.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instruction is executed by the processor, it implements the multi-source CT reconstruction method based on cone-beam to parallel-beam rearrangement as described in any one of claims 1-7.

Images