A CT image reconstruction method based on a three-dimensional diffusion model and a related device

Abstract

The application belongs to the field of CT image processing, and particularly relates to a CT image reconstruction method based on a three-dimensional diffusion model and a related device, which comprises the following steps: obtaining three-dimensional CT voxel data by using a medical CT scanner, and preprocessing the three-dimensional CT voxel data; extracting a medical object surface point cloud from the preprocessed three-dimensional CT voxel data by using a voxel-level reconstruction algorithm, and generating three-dimensional sparse point cloud data; processing the three-dimensional sparse point cloud data by using a 3D Gaussian splatting technology, converting the three-dimensional sparse point cloud data into three-dimensional Gaussian distribution data by using a pre-trained three-dimensional Gaussian kernel; gradually adding Gaussian noise to the three-dimensional Gaussian distribution data by using a three-dimensional diffusion model, and gradually removing the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data; and projecting the reconstructed three-dimensional Gaussian distribution data onto a two-dimensional plane by using a rendering technology to generate a two-dimensional CT reconstruction image. The application combines the advantages of the 3D Gaussian splatting technology and the three-dimensional diffusion model, can realize efficient, real-time and denoising medical image reconstruction under low-dose CT imaging conditions, and effectively solves the white floating artifact problem that may occur in the reconstruction process.

Description

Technical Field

[0001] This invention belongs to the field of medical sparse view CT reconstruction technology, specifically relating to a CT image reconstruction method and related apparatus based on a three-dimensional diffusion model. Background Technology

[0002] In 3D CT scans, an object or human body is irradiated with X-rays from different angles. A detector receives the X-rays that have passed through the object and converts them into digital signals. These digital signals are recorded to form raw 3D voxel data. 3D voxel data is the smallest data unit in 3D space, similar to a pixel in a 2D image. Using specific reconstruction algorithms, this 3D voxel data can be converted into a 2D CT image.

[0003] Currently, 3D CT reconstruction technology is an important tool for analyzing internal structures and diagnosis. Traditional 3D CT reconstruction methods, such as algebraic reconstruction and filtered backprojection, are susceptible to noise under low-dose CT, leading to artifacts and reconstruction errors. Furthermore, these methods struggle to generate high-quality images under sparse view conditions. In recent years, deep learning methods such as convolutional neural networks have been introduced into CT reconstruction, as exemplified by the scheme in Chinese patent CN116824146A. While these schemes improve image quality and noise suppression, their application is limited by their dependence on large amounts of labeled data, complex training processes, and insufficient generalization ability. Generative adversarial networks (GANs) have advantages in generating high-definition images, but their training is unstable and prone to producing unrealistic artifacts. Neural implicit representation methods such as NeRF are effective in simulating complex geometry, but their high computational cost and long training time make them unsuitable for real-time imaging.

[0004] Currently, some technicians are using 3D Gaussian splashing technology to solve the problem of 3D image reconstruction. The core idea of ​​3D Gaussian splashing is to represent a 3D scene as a set of 3D Gaussian functions. Each 3D Gaussian function has its specific parameters, including 3D position, covariance matrix, color, and transparency. These Gaussian functions are distributed in 3D space and together constitute a continuous and realistic 3D scene. However, 3D Gaussian splashing technology cannot directly reconstruct 3D CT voxel data, and it is difficult to remove noise and white floating object artifacts in CT images. Summary of the Invention

[0005] In view of this, the present invention proposes a CT image reconstruction method and related device based on a three-dimensional diffusion model. The method uses a voxel-level reconstruction algorithm to reconstruct three-dimensional CT voxel data, and then uses 3D Gaussian splashing technology to convert the three-dimensional sparse point cloud data into high-quality three-dimensional Gaussian distribution data. By combining the three-dimensional diffusion model to process the three-dimensional Gaussian distribution data, noise and white floating object artifacts in CT images can be gradually removed. Under sparse view conditions, noise is significantly suppressed and artifacts are reduced, the original image details are restored, and the structural information of medical images is effectively preserved.

[0006] In a first aspect, the present invention provides a CT image reconstruction method based on a three-dimensional diffusion model, the method comprising:

[0007] Three-dimensional CT voxel data are acquired using a medical CT scanner, and the three-dimensional CT voxel data are preprocessed.

[0008] A voxel-level reconstruction algorithm is used to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data to generate 3D sparse point cloud data.

[0009] 3D Gaussian splashing technology is used to process 3D sparse point cloud data, and pre-trained 3D Gaussian kernels are used to convert the data into 3D Gaussian distribution data.

[0010] Gaussian noise is gradually added to the three-dimensional Gaussian distribution data using a three-dimensional diffusion model, and then Gaussian noise is gradually removed to generate reconstructed three-dimensional Gaussian distribution data.

[0011] Rendering techniques are used to project the reconstructed 3D Gaussian distribution data onto a 2D plane to generate a 2D CT reconstructed image.

[0012] In a second aspect, the present invention also proposes a CT image reconstruction device based on a three-dimensional diffusion model, the device comprising:

[0013] The data acquisition module uses a medical CT scanner to acquire three-dimensional CT voxel data and preprocesses the three-dimensional CT voxel data.

[0014] The voxel reconstruction module uses a voxel-level reconstruction algorithm to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data and generate 3D sparse point cloud data.

[0015] The point cloud reconstruction module uses 3D Gaussian splashing technology to process 3D sparse point cloud data and uses a pre-trained 3D Gaussian kernel to convert it into 3D Gaussian distribution data.

[0016] The diffusion reconstruction module uses a three-dimensional diffusion model to gradually add Gaussian noise to the three-dimensional Gaussian distribution data and gradually remove the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data.

[0017] The rendering and reconstruction module uses rendering technology to project the reconstructed 3D Gaussian distribution data onto a 2D plane to generate a 2D CT reconstructed image.

[0018] In a third aspect, the present invention also provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect of the present invention.

[0019] In a fourth aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0020] Compared with the prior art, the advantages of the present invention are as follows:

[0021] This invention overcomes many limitations of existing CT reconstruction techniques. First, it employs frequency regularization to constrain the positional encoding of the Gaussian kernel, effectively preventing overfitting during model training and thus improving the generalization ability of the reconstruction results. Second, it utilizes a three-dimensional diffusion model for denoising, progressively removing noise and white floater artifacts from CT images, restoring original image details, and effectively preserving the structural information of medical images. This avoids the over-smoothing phenomenon that may occur in traditional denoising methods, retaining details and edge information, and enhancing the diagnostic value of the images. Finally, combined with 3D Gaussian splatting technology, this invention achieves efficient three-dimensional data mapping and image reconstruction, significantly improving computational efficiency while ensuring image quality. Ultimately, this invention possesses real-time processing capabilities, making it particularly suitable for rapid reconstruction under low-dose scanning conditions, meeting the dual requirements of real-time performance and high quality in medical imaging. Attached Figure Description

[0022] Figure 1 This is a flowchart of the CT image reconstruction method based on a three-dimensional diffusion model according to an embodiment of the present invention;

[0023] Figure 2 This is a flowchart of a preferred embodiment of the CT image reconstruction method based on a three-dimensional diffusion model according to the present invention;

[0024] Figure 3 This is a schematic diagram of the CT image reconstruction structure based on a three-dimensional diffusion model according to an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of an electronic device structure according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] In some embodiments, this embodiment provides a CT image reconstruction method based on a three-dimensional diffusion model, such as... Figure 1 As shown, the method includes the following:

[0028] S101. A medical CT scanner is used to acquire three-dimensional CT voxel data, and the three-dimensional CT voxel data is preprocessed. In some embodiments, the three-dimensional CT voxel data is acquired by a medical CT scanner and then subjected to smooth interpolation and noise reduction processing to improve data quality.

[0029] First, a series of two-dimensional chest slice images are acquired using a medical CT scanner. By setting appropriate scanning parameters, the chest is scanned from multiple angles, resulting in approximately 20 to 30 chest CT images at different angles. Each image represents a cross-section or slice of the chest, and these two-dimensional slice data are combined into a three-dimensional chest CT voxel dataset through a stacking operation. Second, filtering techniques (such as Gaussian filtering and median filtering) are applied to remove noise from the images. Noise removal improves image clarity and detail, and reduces errors in subsequent processing.

[0030] S201. Voxel-level reconstruction algorithm is used to extract the surface point cloud of medical object from the preprocessed 3D CT voxel data to generate 3D sparse point cloud data.

[0031] In some embodiments, this embodiment uses the Marching Cubes algorithm to extract point clouds from the surface of medical objects from preprocessed CT data, generating sparse point cloud data to provide three-dimensional geometric information.

[0032] After step S101, three-dimensional CT voxel data of the chest, after smoothing interpolation and denoising, were obtained. This voxel dataset contains information on various structures inside the chest, including the lungs, pleural cavity, heart, and blood vessels.

[0033] First, the Marching Cubes algorithm is used to perform surface detection in the 3D voxel data. For each voxel, the algorithm analyzes other voxels in its neighborhood to find surface boundaries that meet a set threshold T. For soft tissue in chest CT scan data, the threshold range is set to 200. When a voxel's value reaches or exceeds T, it is identified as a voxel on the surface.

[0034] A sign change between adjacent CT voxel values ​​indicates the existence of an isosurface between the two CT voxels. Linear interpolation is used to calculate the intersection coordinates to determine the precise intersection point between the isosurface and the CT voxel boundary.

[0035] Based on the state of each CT voxel, the connection method of the vertices is determined with reference to the lookup table of the Marching Cubes algorithm. A continuous mesh is formed by connecting triangles to reconstruct the surface structure of the chest. In chest CT data, the mesh for the lungs and pleural cavity surface is mainly composed of larger triangles, while in more detailed structures such as vascular branches and small cavities, the mesh consists of more and denser triangles. This triangular mesh connection method fully presents the three-dimensional structure of the chest organ surface.

[0036] S301. The 3D Gaussian splashing technique is used to process the three-dimensional sparse point cloud data, and the pre-trained three-dimensional Gaussian kernel is used to convert it into three-dimensional Gaussian distribution data.

[0037] In some embodiments, the 3DGS method is used to train a Gaussian kernel from 3D point cloud data to represent a real 3D scene, and frequency regularization is applied to the Gaussian kernel position encoding.

[0038] In step S201, sparse point cloud data of the medical object, containing 3D surface information of the chest, was extracted from the chest CT data. Next, using the 3DGS method, this sparse point cloud data is transformed into a Gaussian kernel representation, thereby constructing a continuous distribution model of the 3D scene. The Gaussian kernel takes the form:

[0039]

[0040] in, Let μ represent the Gaussian kernel at position P, μ = (x, y, z) be the position of the point, Σ be the covariance matrix, and |Σ| be the determinant of the covariance matrix.

[0041] This Gaussian kernel representation method can construct a three-dimensional scene with continuous properties on the sparse representation of chest CT point cloud, effectively describing the geometric shape and color features of chest anatomy.

[0042] In some embodiments, this embodiment applies frequency regularization to the Gaussian kernel positional encoding to enhance the model's ability to represent high-frequency detail information. While maintaining overall smoothness, it enhances the capture of details, enabling the model to better reproduce the fine structures in chest CT data. Frequency regularization adjusts the positional encoding by introducing high-frequency components, specifically using the following formula:

[0043]

[0044] Where f is the regularization frequency factor, which controls the frequency level; λ is the regularization weight, which controls the smoothness; ‖P‖ represents the magnitude of the position vector, which is used to adjust the regularization effect at different positions. Here, P = (x,y,z), and correspondingly, ‖(x,y,z)‖ represents the magnitude of the position vector of point (x,y,z), which measures the distance of the point in space.

[0045] After completing the frequency regularization of the positional encoding, this embodiment can use a gradient descent optimization algorithm to train the Gaussian kernel parameters to optimally represent the three-dimensional structure of the chest. Specifically, by continuously adjusting the center position and covariance matrix of each Gaussian kernel, the error between the Gaussian kernel representation and the actual chest CT structure is minimized.

[0046] During training, real point cloud data from chest CT scans is used as the target, and a loss function L is defined to represent the difference between the reconstructed result and the real scene. The loss function is defined as a combination of the Gaussian kernel and the geometric and color errors of the point cloud data:

[0047]

[0048] Among them, P i,model and P i,target C represents the location of the Gaussian kernel generated by the model and the location of the target point cloud, respectively; i,model and C i,target These represent the Gaussian kernel color generated by the model and the target point cloud color, respectively; α is the weight of the color error, which controls the relative importance of the geometric error and the color error.

[0049] S401. Using a three-dimensional diffusion model, Gaussian noise is gradually added to the three-dimensional Gaussian distribution data, and Gaussian noise is gradually removed to generate reconstructed three-dimensional Gaussian distribution data.

[0050] In some embodiments, Gaussian noise is added to features such as Gaussian kernel position, covariance, and color, and noise is gradually removed using a three-dimensional diffusion model to restore the original sparse point cloud.

[0051] In some embodiments, during the forward noise addition process, Gaussian noise is independently added to the position encoding vector, covariance matrix and color feature value of the three-dimensional Gaussian distribution data to generate three-dimensional noisy point cloud data.

[0052] In the reverse denoising process, Gaussian noise is removed independently from the position encoding vector, covariance matrix, and color feature values ​​of the 3D noisy point cloud data to generate reconstructed 3D Gaussian distribution data.

[0053] In step S301, the 3D sparse point cloud data of the chest CT dataset was represented as 3D Gaussian distribution data using the 3DGS method, generating a 3D model with features such as position, covariance, and color. Gaussian noise was then added to each feature of the 3D Gaussian kernel to simulate noise and verify the model's denoising capability.

[0054] For each location point of the Gaussian kernel, independent and identically distributed Gaussian noise is added at that location. The noise equation is as follows:

[0055]

[0056] Where, η P This is the location noise term, with zero mean and variance of . Gaussian distribution.

[0057] For the covariance matrix of each Gaussian kernel, Gaussian noise is added to the matrix to simulate uncertainty:

[0058]

[0059] Where, η Σ It is a covariance noise matrix with zero mean and variance. , where I is the identity matrix.

[0060] For each Gaussian kernel's color feature, Gaussian noise is added to the color value:

[0061]

[0062] Where, η C It is a color noise term with zero mean and variance of . The Gaussian distribution is applied to the red, green, and blue channels respectively.

[0063] In some embodiments, a three-dimensional diffusion model is used to progressively denoise the Gaussian kernel features after noise has been added. The three-dimensional diffusion model gradually removes Gaussian noise and restores the original three-dimensional Gaussian distribution data through forward noise addition and backward denoising steps.

[0064] During the training phase of the diffusion model, the forward process generates training samples containing different noise levels by progressively adding noise to the three-dimensional Gaussian distributed data at each time step t. The noise addition formula for this process is:

[0065]

[0066] Where, α t Controlling the noise intensity at each time step, σ 2 It is the noise variance. This represents the Gaussian kernel with added noise at time step t.

[0067] In the denoising stage, a trained 3D diffusion model is used to progressively denoise the noisy Gaussian kernel features. At each time step t, the model predicts the denoised image based on the current noisy image and gradually removes the noise during the backdiffusion process. The denoising formula for the backdiffusion process is:

[0068]

[0069] Among them, f θ It is a pre-trained diffusion model. It is the noisy Gaussian kernel feature at step t. These are features generated after denoising, gradually approximating the original sparse point cloud.

[0070] In some embodiments, noise at each time step is gradually removed through iterative steps of a backdiffusion process. The denoised Gaussian kernel features obtained in each step are used as input for the next step to continue denoising. After multiple iterations, Gaussian kernel features that are close to the original sparse point cloud are finally recovered. The resulting point cloud data has high-quality 3D geometric and color information, suitable for subsequent 3D reconstruction and medical analysis.

[0071] S501. Using rendering technology, the reconstructed three-dimensional Gaussian distribution data is projected onto a two-dimensional plane to generate a two-dimensional CT reconstructed image.

[0072] In some embodiments, this embodiment utilizes rendering technology to map a three-dimensional Gaussian kernel onto a two-dimensional plane, completing the conversion from three-dimensional data to a two-dimensional image.

[0073] In step S401, denoising of the sparse point cloud in the chest CT data was performed using a 3D diffusion model to obtain high-quality sparse point cloud data. Next, 3DGS technology was used to map this 3D point cloud data onto a 2D plane to generate CT image slices.

[0074] When projecting a 3D Gaussian kernel onto a 2D image plane, multiple Gaussian kernels influence the same 2D pixel location, thus requiring a weighted average of these influences. A weighted average is used to mix the intensities of multiple Gaussian kernels at the same location. The weighting coefficient of each Gaussian kernel is related to its own intensity density.

[0075]

[0076] Among them, w i The weights are related to the density intensity of the Gaussian kernel. Related; This represents the contribution of the Gaussian kernel at the projection position.

[0077] In some embodiments, after all three-dimensional Gaussian kernels are projected onto a two-dimensional plane, different z-axis positions are selected as slice depths, multiple slice depths are set, and planes at different depths are projected to obtain a series of CT image slices.

[0078] In the application of chest CT data, these slices represent different anatomical layers, such as slice views of the lungs, pleural cavity, and heart. The image pixel values ​​on each slice plane are generated by a three-dimensional Gaussian kernel projection of the corresponding depth, ultimately resulting in a two-dimensional CT reconstructed image.

[0079] In some preferred embodiments, this embodiment provides a CT image reconstruction method based on a three-dimensional diffusion model, such as... Figure 2 As shown, the method includes the following:

[0080] S111. Acquire three-dimensional CT voxel data using a medical CT scanner, and preprocess the three-dimensional CT voxel data.

[0081] S211. A voxel-level reconstruction algorithm is used to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data to generate 3D sparse point cloud data.

[0082] S311. Use 3D Gaussian splashing technology to process three-dimensional sparse point cloud data, and use pre-trained three-dimensional Gaussian kernels to convert it into three-dimensional Gaussian distribution data.

[0083] S411. Using a three-dimensional diffusion model, Gaussian noise is gradually added to the three-dimensional Gaussian distribution data, and Gaussian noise is gradually removed to generate reconstructed three-dimensional Gaussian distribution data.

[0084] S511. Using rendering technology, the reconstructed three-dimensional Gaussian distribution data is projected onto a two-dimensional plane to generate a two-dimensional CT reconstruction image;

[0085] S611. Gaussian noise is gradually added to the two-dimensional CT reconstructed image using a two-dimensional diffusion model, and then Gaussian noise is gradually removed to generate an updated two-dimensional CT reconstructed image.

[0086] The specific implementation of steps S211-S511 can be referred to in steps S201-S501, which will not be repeated in this embodiment. Step S611 will be explained in detail below.

[0087] In this embodiment, a two-dimensional diffusion model is used to assist in denoising two-dimensional CT images to generate high-quality medical CT images for medical diagnosis and analysis.

[0088] In steps S501 and S511, 3D Gaussian splatting technology is used to map a three-dimensional Gaussian kernel onto a two-dimensional plane, generating two-dimensional CT image slices of different depths. Due to uncertainties in the Gaussian kernel mapping process and various factors in medical imaging, these CT images may contain Gaussian noise, artifacts, scanning errors, and other noise. When the image quality of the two-dimensional CT reconstructed image generated in S501 or S511 is less than a preset threshold, such as the contrast of the two-dimensional CT reconstructed image, if the contrast exceeds 80%, we consider the generated two-dimensional CT reconstructed image to meet the requirements; if the contrast is less than 80%, we consider the generated two-dimensional CT reconstructed image to need to be further denoised using a two-dimensional diffusion model. Therefore, the generated two-dimensional CT image slices are used as denoising input and fed into the two-dimensional diffusion model for denoising processing.

[0089] A two-dimensional diffusion model is used to denoise input noisy CT image slices to gradually restore image details and clarity. The diffusion model is trained to identify and remove noise through forward noise addition and reverse denoising processes.

[0090] During the training phase of the diffusion model, the forward process generates training samples with different noise levels by progressively adding noise to the noise-free image, enabling the model to learn denoising properties. In the denoising phase, the trained 2D diffusion model is used to progressively recover a clear image from the noisy image. Through the progressive denoising operation of the backward diffusion process, the model can gradually eliminate noise and restore image details, preserving edge details and structural information in chest CT images.

[0091] Noise in the image is gradually removed through multiple iterations of the back-diffusion process, with the denoising result of each step serving as input for the next. After multiple iterations, the final denoised image I is obtained. clean Approximately original, noise-free medical CT images.

[0092] After denoising, the recovered chest CT images no longer contain artifacts and noise, and retain clear edges and structural information, providing reliable imaging support for clinical medicine.

[0093] Please see Figure 3 , Figure 3 This is a schematic diagram of the structural composition of the CT image reconstruction device based on a three-dimensional diffusion model according to an embodiment of the present invention. Figure 3 As shown, a CT image reconstruction device based on a three-dimensional diffusion model is disclosed. The device includes: a data acquisition module 121, which uses a medical CT scanner to acquire three-dimensional CT voxel data and preprocesses the three-dimensional CT voxel data.

[0094] Voxel reconstruction module 221 uses a voxel-level reconstruction algorithm to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data and generate 3D sparse point cloud data.

[0095] The point cloud reconstruction module 321 uses 3D Gaussian splashing technology to process three-dimensional sparse point cloud data and uses a pre-trained three-dimensional Gaussian kernel to convert it into three-dimensional Gaussian distribution data.

[0096] The diffusion reconstruction module 421 uses a three-dimensional diffusion model to gradually add Gaussian noise to the three-dimensional Gaussian distribution data and gradually remove Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data.

[0097] The rendering and reconstruction module 521 uses rendering technology to project the reconstructed three-dimensional Gaussian distribution data onto a two-dimensional plane to generate a two-dimensional CT reconstruction image.

[0098] This invention also provides an electronic device, such as... Figure 4 As shown. The electronic device includes components such as a processor 402, a memory 403, an input unit 404, and a display unit 405. Those skilled in the art will understand that... Figure 4 The illustrated device structures are not intended to limit all devices and may include more or fewer components than shown, or combine certain components. Memory 403 can be used to store application program 401 and various functional modules. Processor 402 runs application program 401 stored in memory 403, thereby performing various functional applications and data processing of the device. Memory can be internal memory or external memory, or both. Internal memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, or random access memory. External memory may include hard disks, floppy disks, ZIP disks, USB flash drives, magnetic tapes, etc. The memory disclosed in this invention includes, but is not limited to, these types of memory. The memory disclosed in this invention is only an example and not a limitation.

[0099] This invention provides a computer-readable storage medium storing an application program. When executed by a processor, this program implements the CT image reconstruction method based on a three-dimensional diffusion model, as described in any of the above embodiments. The computer-readable storage medium includes, but is not limited to, any type of disk (including floppy disks, hard disks, optical disks, CD-ROMs, and magneto-optical disks), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic cards, or optical cards. In other words, the storage device includes any medium that stores or transmits information in a readable form by a device (e.g., a computer, a mobile phone), and can be a read-only memory, a disk, or an optical disk, etc.

[0100] This invention also provides a computer application program that runs on a computer and is used to execute a CT image reconstruction method based on a three-dimensional diffusion model according to any of the above embodiments. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0101] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include ROM, RAM, disk, or optical disk, etc.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A CT image reconstruction method based on a three-dimensional diffusion model, characterized in that, Includes the following steps: Three-dimensional CT voxel data are acquired using a medical CT scanner, and the three-dimensional CT voxel data are preprocessed. A voxel-level reconstruction algorithm is used to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data to generate 3D sparse point cloud data. 3D Gaussian splashing technology is used to process 3D sparse point cloud data, and a pre-trained 3D Gaussian kernel is used to convert it into 3D Gaussian distribution data; the position encoding vector of the 3D Gaussian kernel is represented as: in, Indicates the three-dimensional Gaussian kernel in Position encoding vector, This is a regularization frequency factor used to control the frequency level. These are regularization weights used to control the degree of smoothness; The modulus of the positional encoding vector is used to adjust the regularization effect at different positions; A three-dimensional diffusion model is used to progressively add Gaussian noise to the three-dimensional Gaussian distribution data, and then progressively remove the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data; including: In the forward noise addition process, Gaussian noise is independently added to the position encoding vector, covariance matrix and color feature value of the three-dimensional Gaussian distribution data to generate three-dimensional noisy point cloud data. In the reverse denoising process, Gaussian noise is independently removed from the position encoding vector, covariance matrix and color feature value of the 3D noisy point cloud data to generate reconstructed 3D Gaussian distribution data. Rendering techniques are used to project the reconstructed 3D Gaussian distribution data onto a 2D plane to generate a 2D CT reconstructed image.

2. The CT image reconstruction method based on a three-dimensional diffusion model according to claim 1, characterized in that, The position encoding vector of the three-dimensional Gaussian kernel is processed by frequency regularization.

3. The CT image reconstruction method based on a three-dimensional diffusion model according to claim 1, characterized in that, The regularization frequency factor is a high-frequency factor, which controls the adjustment of the position encoding vector by the high-frequency components.

4. The CT image reconstruction method based on a three-dimensional diffusion model according to claim 1, characterized in that, The step of independently adding Gaussian noise to the position encoding vector, covariance matrix, and color eigenvalues ​​of the three-dimensional Gaussian distribution data includes: For the location encoding vector of the three-dimensional Gaussian distribution data, add independent first Gaussian noise at each location; For the covariance matrix of three-dimensional Gaussian distributed data, add independent second Gaussian noise to each matrix; For the color features of the three-dimensional Gaussian distribution data, an independent third Gaussian noise is added to each color channel; The first Gaussian noise, the second Gaussian noise, and the third Gaussian noise are all distributed in the same way.

5. The CT image reconstruction method based on a three-dimensional diffusion model according to claim 1, characterized in that, After using rendering techniques to project the reconstructed 3D Gaussian distribution data onto a 2D plane to generate a 2D CT reconstructed image, the process also includes: If the image quality of the two-dimensional CT reconstructed image does not exceed a preset threshold, a two-dimensional diffusion model is used to gradually add Gaussian noise to the two-dimensional CT reconstructed image and gradually remove the Gaussian noise to generate an updated two-dimensional CT reconstructed image.

6. A CT image reconstruction device based on a three-dimensional diffusion model, characterized in that, The device includes: The data acquisition module uses a medical CT scanner to acquire three-dimensional CT voxel data and preprocesses the three-dimensional CT voxel data. The voxel reconstruction module uses a voxel-level reconstruction algorithm to extract the surface point cloud of the medical object from the preprocessed 3D CT voxel data and generate 3D sparse point cloud data. The point cloud reconstruction module employs 3D Gaussian splashing technology to process 3D sparse point cloud data, converting it into 3D Gaussian distribution data using a pre-trained 3D Gaussian kernel; the position encoding vector of the 3D Gaussian kernel is represented as: in, Indicates the three-dimensional Gaussian kernel in Position encoding vector, This is a regularization frequency factor used to control the frequency level. These are regularization weights used to control the degree of smoothness; The modulus of the positional encoding vector is used to adjust the regularization effect at different positions; The diffusion reconstruction module uses a three-dimensional diffusion model to progressively add Gaussian noise to the three-dimensional Gaussian distribution data and then progressively removes the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data; including: In the forward noise addition process, Gaussian noise is independently added to the position encoding vector, covariance matrix and color feature value of the three-dimensional Gaussian distribution data to generate three-dimensional noisy point cloud data. In the reverse denoising process, Gaussian noise is independently removed from the position encoding vector, covariance matrix and color feature value of the 3D noisy point cloud data to generate reconstructed 3D Gaussian distribution data. The rendering and reconstruction module uses rendering technology to project the reconstructed 3D Gaussian distribution data onto a 2D plane to generate a 2D CT reconstructed image.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images