Material Reconstruction Method and Apparatus for Multi-Energy Computed Tomography Imaging Systems

Abstract

This application relates to a method and apparatus for material reconstruction using a multi-energy computed tomography (MCT) imaging system. The method includes: scanning a target object using a MCT imaging system to acquire corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths; constructing an initial material density field model of the target object using the spatial geometric parameters to convert the three-dimensional continuous spatial coordinates into corresponding base material density; calculating corresponding predicted projection data using the base material density; optimizing the initial material density field model by combining the multi-energy projection data and the predicted projection data to obtain an actual material density field model; and reconstructing the base material density map of the target object at any resolution using the actual material density field model. This solves the technical problems in related technologies, such as reliance on discrete voxel grids, high computational complexity, and reconstruction quality limited by data quality, making it difficult to meet clinical needs.

Description

Technical Field

[0001] This application relates to the field of image data processing technology, and in particular to a method and apparatus for material reconstruction in a multi-energy computed tomography imaging system. Background Technology

[0002] When using a two-step method, the material decomposition and image reconstruction processes in multi-energy computed tomography (MCCT) imaging systems can be decoupled from each other, preventing them from interfering with each other. During execution, projection domain decomposition combined with filtered back projection (FBP) can be used. However, the projection domain decomposition process is extremely sensitive to noise. When the input is low signal-to-noise ratio projection data, the decomposition operation will further amplify the noise, seriously affecting the subsequent reconstruction accuracy. Moreover, projection domain decomposition combined with filtered back projection has strict requirements for the completeness of the projection data. In common clinical scenarios such as undersampling, it cannot effectively constrain the pathological nature of the reconstruction inverse problem, which can easily lead to obvious stripe artifacts in the reconstructed image and reduce image quality.

[0003] To address these issues, related technologies employ a one-step method for material reconstruction to avoid error propagation in intermediate steps. However, this method still fundamentally relies on explicit representations of discrete voxel meshes. This discretization design not only leads to a sharp increase in computational complexity, making the reconstruction process time-consuming and unsuitable for rapid clinical imaging needs, but also, in scenarios with poor data quality such as extremely low doses and sparse views, the reconstruction quality significantly degrades due to the pathological nature of the reconstruction inverse problem. This fails to meet the urgent clinical demand for efficient and high-precision energy dispersive spectral imaging and urgently requires improvement. Summary of the Invention

[0004] This application provides a method and apparatus for material reconstruction in a multi-energy computed tomography imaging system to solve the technical problems in related technologies, such as reliance on discrete voxel grids, high computational complexity, and reconstruction quality limited by data quality, which makes it difficult to meet clinical needs.

[0005] The first aspect of this application provides a method for material reconstruction using a multi-energy computed tomography (MCT) imaging system, comprising the following steps: scanning a target object using the MCT imaging system to obtain corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths; constructing an initial material density field model of the target object using the spatial geometric parameters, and using the initial material density field model to convert the three-dimensional continuous spatial coordinates into corresponding base material density; calculating corresponding predicted projection data using the base material density; optimizing the initial material density field model by combining the multi-energy projection data and the predicted projection data to obtain an actual material density field model, and reconstructing the base material density map of the target object at any resolution using the actual material density field model.

[0006] According to the above technical solution, the embodiments of this application can acquire multi-energy projection data, and through the established initial material density field model, perform the conversion between three-dimensional continuous spatial coordinates and base material density. It does not require decoupling material decomposition and image reconstruction, and directly performs mapping calculation on three-dimensional continuous spatial coordinates, thereby improving the signal-to-noise ratio of the reconstructed image, reducing the requirements for projection data, alleviating the problems caused by incomplete projection data, and eliminating the need for pre-dividing discrete voxel grids, breaking the limitations of discrete grids. It also uses the base material density to calculate and predict projection data, thereby combining multi-energy projection data and predicted projection data to optimize the initial material density field model, so as to reconstruct a base material density map of any resolution, which has higher applicability and reliability.

[0007] Optionally, in one embodiment of this application, the step of converting the three-dimensional continuous spatial coordinates into corresponding base matter densities using the initial matter density field model includes: using a position encoder in the initial matter density field model to convert any coordinate in the three-dimensional continuous spatial coordinates into a feature vector of the target dimension; and using a value prediction network in the initial matter density field model to map the feature vector to the base matter density at the corresponding location.

[0008] According to the above technical solution, in the embodiments of this application, the initial material density field model may include a position encoder and a value prediction network. The position encoder realizes the transformation of three-dimensional coordinates into high-dimensional feature vectors, transforming spatial location information into features that the model can recognize. The value prediction network realizes the mapping of features to the base material density, thereby improving the accuracy of subsequent density prediction.

[0009] Optionally, in one embodiment of this application, the step of converting arbitrary coordinates in the three-dimensional continuous spatial coordinates into feature vectors of the target dimension includes: extracting multiple grid vertices of the first network layer from the position encoder based on the arbitrary coordinates; mapping the multiple grid vertices to the corresponding indexes of a hash table using a preset hash function; querying the hash table using the indexes to obtain an independent feature vector corresponding to each grid vertex; fusing the independent feature vectors using trilinear interpolation to obtain a single-layer feature vector of the arbitrary coordinates in the first network layer; obtaining the single-layer feature vectors corresponding to the next network layer, until the single-layer feature vectors corresponding to all network layers are obtained; and concatenating all single-layer feature vectors to obtain the feature vector of the target dimension.

[0010] According to the above technical solution, the embodiments of this application combine multi-resolution grid hierarchy and hash coding to realize multi-scale feature extraction of spatial coordinates, improve the representation ability of feature vectors, and improve feature query efficiency and reduce computational complexity by quickly mapping indexes through hash functions. Then, discrete vertex features are transformed into continuous coordinate features through trilinear interpolation, which solves the resolution limitation caused by discrete grids.

[0011] Optionally, in one embodiment of this application, mapping the feature vector to the base material density at the corresponding position includes: using the value prediction network to map the feature vector to an initial value at the corresponding position; and scaling the initial value using a base material specificity coefficient to obtain the base material density.

[0012] According to the above technical solution, the embodiments of this application can use the base material specificity coefficient to scale the initial value in order to distinguish the physical properties of different base materials and avoid confusion of base material density.

[0013] Optionally, in one embodiment of this application, the step of calculating the corresponding predicted projection data using the matrix material density includes: obtaining a set of matrix material densities corresponding to multiple coordinates on any ray path; inputting the set of matrix material densities into a pre-constructed multi-energy computed tomography imaging physical forward model, so as to integrate the set of matrix material densities using the multi-energy computed tomography imaging physical forward model to obtain the predicted projection data of any ray path.

[0014] According to the above technical solution, the embodiments of this application can simulate the attenuation process of rays penetrating matter through a physical forward model, and convert the density of the base material into predicted projection data through integral calculation, thereby increasing the authenticity of the predicted projection data.

[0015] Optionally, in one embodiment of this application, the calculation expression of the multi-energy computed tomography imaging physical forward model is as follows:

[0016] in, Indicates predicted projection data, This represents the normalized X-ray energy spectrum distribution, where E represents photon energy and K represents the number of base material species. This indicates the number of points sampled along the ray. Indicates the coordinates of the point sampled along the ray. Indicates the corresponding sampling step size. Indicates the first The known linear decay coefficient of the seed material. This indicates the network output at the sampling point. The density of the k-th basic substance at location , These represent the parameters of a hidden neural representation network.

[0017] Optionally, in one embodiment of this application, the step of combining the multi-energy projection data and the predicted projection data to optimize the initial matter density field model to obtain the actual matter density field model includes: constructing a loss function using the multi-energy projection data and the predicted projection data; and optimizing the initial matter density field model using the loss function; wherein the calculation expression of the loss function is:

[0018] in, This refers to the multi-energy projection data.

[0019] According to the above technical solution, the embodiments of this application can quantify the error between the measured projection and the predicted projection through the loss function, providing a clear optimization target for model optimization.

[0020] A second aspect of this application provides a material reconstruction apparatus for a multi-energy computed tomography (MCT) imaging system, comprising: an acquisition module for scanning a target object using the MCT imaging system to acquire corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths; a conversion module for constructing an initial material density field model of the target object using the spatial geometric parameters, and converting the three-dimensional continuous spatial coordinates into corresponding base material density using the initial material density field model; a calculation module for calculating corresponding predicted projection data using the base material density; and a reconstruction module for combining the multi-energy projection data and the predicted projection data to optimize the initial material density field model, obtain an actual material density field model, and reconstruct the base material density map of the target object at any resolution using the actual material density field model.

[0021] Optionally, in one embodiment of this application, the conversion module includes: a first conversion unit, used to convert any coordinate in the three-dimensional continuous spatial coordinates into a feature vector of the target dimension using a position encoder in the initial material density field model; and a second conversion unit, used to map the feature vector to the base material density at the corresponding position using a value prediction network in the initial material density field model.

[0022] Optionally, in one embodiment of this application, the first transformation unit includes: an extraction subunit, configured to extract multiple grid vertices of a first network layer from the position encoder based on the arbitrary coordinates; a first mapping subunit, configured to map the multiple grid vertices to the corresponding indexes of a hash table using a preset hash function; a first acquisition subunit, configured to query the hash table using the indexes to obtain an independent feature vector corresponding to each grid vertex; a second acquisition subunit, configured to fuse the independent feature vectors using trilinear interpolation to obtain a single-layer feature vector of the arbitrary coordinates at the first network layer; a third acquisition subunit, configured to acquire the single-layer feature vector corresponding to the next network layer, until the single-layer feature vectors corresponding to all network layers are obtained; and a concatenation subunit, configured to concatenate all single-layer feature vectors to obtain the feature vector of the target dimension.

[0023] Optionally, in one embodiment of this application, the second transformation unit includes: a second mapping subunit, used to map the feature vector to an initial value at a corresponding position using the value prediction network; and a processing subunit, used to scale the initial value using a base material specificity coefficient to obtain the base material density.

[0024] Optionally, in one embodiment of this application, the calculation module includes: an acquisition unit, configured to acquire a set of matrix material densities corresponding to multiple coordinates on any ray path; and a calculation unit, configured to input the set of matrix material densities into a pre-constructed multi-energy computed tomography imaging physical forward model, so as to integrate the set of matrix material densities using the multi-energy computed tomography imaging physical forward model to obtain the predicted projection data of any ray path.

[0025] Optionally, in one embodiment of this application, the calculation expression of the multi-energy computed tomography imaging physical forward model is as follows:

[0026] in, Indicates predicted projection data, This represents the normalized X-ray energy spectrum distribution, where E represents photon energy and K represents the number of base material species. This indicates the number of points sampled along the ray. Indicates the coordinates of the point sampled along the ray. Indicates the corresponding sampling step size. Indicates the first The known linear decay coefficient of the seed material. This indicates the network output at the sampling point. The density of the k-th basic substance at location , These represent the parameters of a hidden neural representation network.

[0027] Optionally, in one embodiment of this application, the reconstruction module includes: a construction unit, configured to construct a loss function using the multi-energy projection data and the predicted projection data; and an optimization unit, configured to optimize the initial matter density field model using the loss function; wherein the calculation expression of the loss function is:

[0028] in, This refers to the multi-energy projection data.

[0029] A third aspect of this application 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 program to implement the material reconstruction method of the multi-energy computed tomography imaging system as described in the above embodiments.

[0030] A fourth aspect of this application provides a computer-readable storage medium storing computer instructions for causing the computer to perform a material reconstruction method for a multi-energy computed tomography imaging system as described in the above embodiments.

[0031] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, implements the material reconstruction method of the multi-energy computed tomography imaging system described above.

[0032] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0033] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a material reconstruction method for a multi-energy computed tomography imaging system according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the principle of a material reconstruction method for a multi-energy computed tomography imaging system according to an embodiment of this application; Figure 3 This is a comparison chart of the results of a multi-energy computed tomography imaging experiment based on kilovolt switching technology according to an embodiment of this application; Figure 4 This is a comparison chart of the results of a multi-energy computed tomography imaging experiment based on a dual-panel detector according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a material reconstruction device for a multi-energy computed tomography imaging system according to an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0034] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0035] Understandably, multi-energy computed tomography (MCT) imaging systems utilize multi-energy data to achieve material decomposition and quantitative analysis, effectively distinguishing tissue materials with different atomic numbers. However, spectral imaging is extremely sensitive to the quality of projection data. In practical applications, limitations imposed by radiation dose control, scanning speed requirements, and detector physical characteristics often result in problems such as high projection noise, sparse projection angles, or mismatch between high and low energy data.

[0036] At the level of specific algorithms for multi-energy CT reconstruction, existing methods are mainly divided into two-step and one-step methods.

[0037] The two-step method decouples the material decomposition and image reconstruction processes. Projective domain decomposition combined with filtered back projection (FBP) is a common implementation strategy. However, the projective domain decomposition process is sensitive to noise; low signal-to-noise ratio projected data will have its noise further amplified after decomposition. Furthermore, as an analytical reconstruction algorithm, projective domain decomposition combined with filtered back projection (FBP) has high requirements for data completeness and cannot effectively constrain the ill-conditioned nature of the inverse problem under undersampling conditions, easily producing fringe artifacts.

[0038] While one-step reconstruction avoids error propagation in intermediate steps, it still relies on the explicit representation of discrete voxel meshes. This discretization not only leads to extremely high computational complexity and long reconstruction time, but also results in a significant decrease in reconstruction quality when faced with extremely low-dose or sparse view data due to the characteristics of the ill-conditioned inverse problem.

[0039] Furthermore, while the hybrid reconstruction scheme that incorporates deep learning to accelerate computation improves the speed of computation, it fails to overcome the limitations of discrete grids on resolution and geometric flexibility, making it difficult to meet the urgent clinical demand for efficient and high-precision energy spectrum imaging.

[0040] To address the technical problems in the aforementioned related technologies, such as reliance on discrete voxel grids, high computational complexity, and decreased reconstruction quality and insufficient material decomposition accuracy under low-dose and undersampling conditions, which reduce the applicability and reliability of the system, the embodiments of this application can acquire multi-energy projection data, establish a continuous material density field model through a position encoder and value prediction network, and perform end-to-end physical-driven optimization by combining a physical forward model. This can break the limitations of discrete grids, improve the signal-to-noise ratio and material decomposition accuracy of the reconstructed image, and have stronger robustness under sparse view and low-dose conditions, thereby improving applicability and reliability.

[0041] Specifically, Figure 1 This is a schematic flowchart illustrating a material reconstruction method for a multi-energy computed tomography imaging system provided in an embodiment of this application.

[0042] like Figure 1 As shown, the material reconstruction method of this multi-energy computed tomography imaging system includes the following steps: In step S101, the target object is scanned using a multi-energy computed tomography imaging system to obtain the corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths.

[0043] Understandably, multi-energy computed tomography (MCT) imaging systems can acquire attenuation data of multiple X-ray energies in a single scan, obtaining multi-parameter molecular-level images. The material reconstruction of MCT imaging systems can simultaneously reconstruct the attenuation information of tissues under different X-ray energy bands.

[0044] Based on this, embodiments of this application can utilize a multi-energy computed tomography imaging system to emit X-rays of different energies and use a detector to simultaneously acquire signals after the X-rays of different energies pass through the target object, thereby obtaining multi-energy projection data.

[0045] When acquiring multi-energy projection data, embodiments of this application can also simultaneously record the spatial geometric parameters obtained by the multi-energy computed tomography imaging system, and the coordinates of all continuously distributed points of multiple ray paths are three-dimensional continuous spatial coordinates, wherein the three-dimensional continuous spatial coordinates can cover the entire area of ​​the target object scanned by the multi-energy computed tomography imaging system.

[0046] In step S102, an initial material density field model of the target object is constructed using spatial geometric parameters, so as to convert the three-dimensional continuous spatial coordinates into the corresponding basic material density using the initial material density field model.

[0047] Furthermore, such as Figure 2As shown, embodiments of this application can construct an initial matter density field model based on implicit neural representation to transform three-dimensional continuous spatial coordinates into corresponding basic matter densities.

[0048] In this embodiment, the spatial constraints of the initial material density field model can be determined by using spatial geometric parameters and a pre-set scanning field of view.

[0049] Optionally, in one embodiment of this application, the transformation of three-dimensional continuous spatial coordinates into corresponding base matter densities using an initial matter density field model includes: using a position encoder in the initial matter density field model to transform any coordinate in the three-dimensional continuous spatial coordinates into a feature vector of the target dimension; and using a value prediction network in the initial matter density field model to map the feature vector to the base matter density at the corresponding location.

[0050] Based on this, such as Figure 2 As shown, in this embodiment of the application, a position encoder can be used to map three-dimensional continuous spatial coordinates into high-dimensional feature vectors, and then a value prediction network can be used to map the high-dimensional feature vectors to output the matrix density at the spatial coordinates.

[0051] The value prediction network can be a multilayer perceptron or a residual network. Since the material density field model outputs multiple base material densities, multiple or a single multilayer perceptron or residual network can be used to generate different base material densities.

[0052] For example, for any input coordinates Position encoder Map the coordinates to a high-dimensional feature vector: , in, For position encoder Network parameters; For high-dimensional feature vectors Value prediction network Map the high-dimensional feature vector to the base material density at the corresponding position: , Value prediction network Network parameters; The overall matter density field model realizes the relationship from spatial location to... Mapping of the density of individual basic substances: .

[0053] Optionally, in one embodiment of this application, converting arbitrary coordinates in three-dimensional continuous spatial coordinates into feature vectors of the target dimension includes: extracting multiple grid vertices of the first network layer from the position encoder based on arbitrary coordinates; mapping the multiple grid vertices to the corresponding hash table indexes using a preset hash function; querying the hash table using the indexes to obtain the independent feature vectors corresponding to each grid vertex; fusing the independent feature vectors using trilinear interpolation to obtain single-layer feature vectors of arbitrary coordinates at the first network layer; obtaining single-layer feature vectors corresponding to the next network layer, until single-layer feature vectors corresponding to all network layers are obtained; and concatenating all single-layer feature vectors to obtain the feature vector of the target dimension.

[0054] In some embodiments, the position encoder may also be a multi-resolution hash encoder.

[0055] This application embodiment can be set There are several grid levels with different resolutions, each containing a hash table. ; For input coordinates In the Find the grid vertices surrounding the point in the layered grid, and use a hash function. Map vertex coordinates to hash table indices: ; The eigenvectors of the vertices are obtained by querying the hash table, and the coordinates at the x-th position are obtained by trilinear interpolation. Layer feature representation ; By concatenating the feature vectors from all levels, the final spatial encoding feature is obtained. .

[0056] Optionally, in one embodiment of this application, mapping the feature vector to the base material density at the corresponding position includes: using a value prediction network to map the feature vector to an initial value at the corresponding position; and using a base material specificity coefficient to scale the initial value to obtain the base material density.

[0057] In other embodiments, the present application embodiments may also introduce a base material specificity coefficient to scale the output of the value prediction network. The actual output of the value prediction network can be modeled as follows: , in, For the first The true density of the seed substrate For the first Specificity coefficient of seed material These are the weight parameters of the network.

[0058] Among them, specificity coefficient That is, the energy spectrum weighted average linear decay coefficient of the base material is used as its specificity coefficient.

[0059] In step S103, the corresponding predicted projection data is calculated using the density of the base material.

[0060] In actual implementation, the embodiments of this application can combine data such as the density of the base material and the linear attenuation coefficient of the base material to perform weighted summation on the data of each ray path to obtain the predicted projection data.

[0061] Optionally, in one embodiment of this application, the corresponding predicted projection data is calculated using the matrix material density, including: obtaining a set of matrix material densities corresponding to multiple coordinates on any ray path; inputting the set of matrix material densities into a pre-constructed multi-energy computed tomography (MCT) imaging physical forward model, so as to integrate the set of matrix material densities using the MCT imaging physical forward model to obtain the predicted projection data for any ray path. The calculation expression of the MCT imaging physical forward model is as follows:

[0062] in, Indicates predicted projection data, This represents the normalized X-ray energy spectrum distribution, where E represents photon energy and K represents the number of base material species. This indicates the number of points sampled along the ray. Indicates the coordinates of the point sampled along the ray. Indicates the corresponding sampling step size. Indicates the first The known linear decay coefficient of the seed material. This indicates the network output at the sampling point. The density of the k-th basic substance at location , These represent the parameters of a hidden neural representation network.

[0063] In multi-energy computed tomography imaging, the X-ray source emits a multi-energy spectrum beam. According to Beer-Lambert's law, for the ... A ray path, its measured projection value It can be represented as: , in, This represents the normalized energy-weighted X-ray energy spectrum distribution corresponding to the radiation. It is a location The energy is The linear attenuation coefficient.

[0064] To achieve the decomposition of matter, the embodiments of this application can model the linear decay coefficient as follows: Linear combination of seed materials (usually) Assume the linear decay coefficient of the base material is known. Then the attenuation coefficient at any position can be decomposed into: ,in Indicates position First Density distribution diagram of seed substrate.

[0065] like Figure 2 As shown in the embodiments of this application, the overall matter density field model realizes the relationship from spatial location to... Mapping of the density of individual basic substances: , Substituting the network output into the physical forward model, a model is established based on the network parameters. To projection data The end-to-end mapping. For a given ray path, the line integral is approximated by numerical integration: , in, For sampling along the ray One point, This represents the sampling step size for the corresponding sampling points.

[0066] in, The network output at the sampling point The first Seed material density. When calculating the projective integral, sampling along each ray can be performed using a combination of methods such as random sampling, equally spaced sampling, stratified sampling, and importance sampling. Different sampling methods will lead to different results. Different generation methods lead to different accuracy or efficiency in integral calculation.

[0067] In step S104, the initial material density field model is optimized by combining multi-energy projection data and predicted projection data to obtain the actual material density field model, and the base material density map of the target object at any resolution is reconstructed using the actual material density field model.

[0068] As one possible approach, embodiments of this application can optimize the parameters of the matter density field model by backpropagation with the goal of minimizing the projection consistency loss between the predicted projection data and the multi-energy projection data, thereby obtaining the actual matter density field model. In each iteration, embodiments of this application can randomly sample a batch of rays from the multi-energy projection data as batch data for gradient descent.

[0069] The actual matter density field model can model the matter density field to be reconstructed as a continuous function, rather than a traditional discrete voxel matrix. It can describe the attenuation process of X-rays interacting with matter and define how neural networks implicitly encode the spatial distribution inside an object, thereby achieving a differentiable mapping from three-dimensional continuous coordinates to multi-energy projection data. Based on the actual matter density field model, embodiments of this application can generate a base matter density map of arbitrary resolution.

[0070] Optionally, in one embodiment of this application, the initial matter density field model is optimized by combining multi-energy projection data and predicted projection data to obtain the actual matter density field model, including: constructing a loss function using multi-energy projection data and predicted projection data; and optimizing the initial matter density field model using the loss function; wherein the calculation expression of the loss function is:

[0071] in, This represents multi-energy projection data.

[0072] The embodiments of this application can transform the reconstruction process into a nonlinear optimization problem, minimizing the predicted projection data. Compared with actual measured multi-energy projection data Loss function: .

[0073] Combination Figures 2 to 4 As shown, the working principle of the material reconstruction method of the multi-energy computed tomography imaging system of this application is explained with reference to one embodiment.

[0074] For example, such as Figure 2 As shown, the matter density field model consists of two parts: a multi-resolution hash coding layer (position encoder) and two multilayer perceptrons (value mapping networks), aiming to achieve a high-precision mapping from three-dimensional spatial coordinates to two basic matter density values. The specific configuration is as follows: Set the number of levels in the hash grid. Dimension of each layer of feature vectors The hash table size T is set to... This is to balance resolution coverage with the probability of hash collisions. Regarding resolution parameters, the base resolution... Setting it to 16 ensures that the high-frequency edge information of the image can be covered and accurately represented. For any input coordinates... This layer extracts features from each level of the grid using trilinear interpolation and then concatenates them to form a layer with dimension 1. eigenvectors This serves as the input to the MLP (Multi-Layer Perceptron). Two MLPs are responsible for decoding the encoded feature vector into two base material density values, respectively.

[0075] To further improve network training speed and stability, embodiments of this application may use a base material specificity coefficient. Since differences in the order of magnitude of base material density values ​​may lead to training instability, this application does not directly model density, but instead models: .in The specificity coefficient is obtained by directly using the energy spectrum weighted average linear attenuation coefficient of this material: .

[0076] In the optimization process, this embodiment of the application may employ the Adam optimizer, with the initial learning rate set to... In each iteration, a batch of rays (e.g., Batch size = 2048) is randomly sampled from the projected data for stochastic gradient descent to introduce randomness and help escape local optima.

[0077] When calculating the integral, uniform sampling (e.g., 384 points) is performed along the ray, and an importance sampling mechanism is introduced to improve the accuracy of forward rendering.

[0078] Therefore, the embodiments of this application can break the fixed resolution limitation of discrete grids through continuous implicit representation, and adaptively fit signals of arbitrary spatial frequencies. Furthermore, since the optimization objective is directly based on the consistency of the projection data, no additional label data is required, resulting in higher physical interpretability.

[0079] For example, such as Figure 3 As shown, in a multi-energy computed tomography imaging simulation experiment based on kilovolt switching technology (with high-energy and low-energy X-ray tube voltages of 80kV and 140kV respectively), compared with the two-step method of projection domain material decomposition combined with filtered back projection (FBP) in related technologies, the embodiment of this application demonstrates superior reconstruction quality. The two-step method exhibits obvious fringe artifacts and noise amplification under sparse projection viewpoint conditions; while the embodiment of this application benefits from the strong fitting ability of multi-resolution hash coding to spatial high-frequency features, resulting in a sharper iodine map of the matrix material at tissue edges. Quantitative analysis shows that, at the same dose, the variance of the ROI (Region of Interest) region in the water-based material map of related technologies is significantly reduced from over 700 in the two-step method to less than 10.

[0080] For example, such as Figure 4As shown, in the actual multi-energy computed tomography imaging experimental platform verification using a dual-layer flat panel detector (120kV X-ray tube, dual-layer amorphous silicon flat panel, 2880×2880 pixel matrix), compared with the two-step method, the reconstructed matrix density map of this application embodiment has advantages in overall contrast and detail clarity. In high-gradient regions such as bone edges, the results of this application maintain sharper boundaries and less artifact interference; in low-contrast soft tissue regions, the image background is more uniform.

[0081] In summary, the embodiments of this application can realize the reconstruction and material decomposition of multi-energy computed tomography imaging by utilizing physical driving and implicit neural representation without relying on discrete voxel meshes. This effectively improves the signal-to-noise ratio and accuracy of the reconstructed images, enhances the precision of material decomposition, and improves the applicability and reliability of the system under low-dose and sparse view conditions.

[0082] The material reconstruction method for a multi-energy computed tomography imaging system proposed in this application can acquire multi-energy projection data. Through an established initial material density field model, it performs a conversion between three-dimensional continuous spatial coordinates and the base material density. This eliminates the need to decouple material decomposition from image reconstruction, directly mapping and calculating the three-dimensional continuous spatial coordinates. This improves the signal-to-noise ratio of the reconstructed image, reduces the requirements for projection data, alleviates problems caused by incomplete projection data, and eliminates the need for pre-dividing discrete voxel grids, breaking the limitations of discrete grids. Furthermore, it utilizes the base material density to calculate and predict projection data, thereby combining multi-energy projection data and predicted projection data to optimize the initial material density field model, reconstructing a base material density map at any resolution. This method offers higher applicability and reliability. Therefore, it solves the problems of related technologies that rely on discrete voxel grids, have high computational complexity, and whose reconstruction quality is limited by data quality, making it difficult to meet clinical needs.

[0083] Based on the same inventive concept as the above-described method embodiments, a material reconstruction apparatus for a multi-energy computed tomography imaging system according to an embodiment of this application is described with reference to the accompanying drawings.

[0084] Figure 5 This is a block diagram of the material reconstruction device of a multi-energy computed tomography imaging system according to an embodiment of this application.

[0085] like Figure 5 As shown, the material reconstruction device 10 of the multi-energy computed tomography imaging system includes: an acquisition module 100, a conversion module 200, a calculation module 300, and a reconstruction module 400.

[0086] Specifically, the acquisition module 100 is used to scan the target object using a multi-energy computed tomography imaging system to acquire the corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths.

[0087] The transformation module 200 is used to construct an initial material density field model of the target object using spatial geometric parameters, so as to transform the three-dimensional continuous spatial coordinates into the corresponding basic material density using the initial material density field model.

[0088] The calculation module 300 is used to calculate the corresponding predicted projection data using the density of the base material.

[0089] The reconstruction module 400 is used to combine multi-energy projection data and predicted projection data to optimize the initial matter density field model, obtain the actual matter density field model, and reconstruct the base matter density map of the target object at any resolution using the actual matter density field model.

[0090] Optionally, in one embodiment of this application, the conversion module 200 includes: a first conversion unit and a second conversion unit.

[0091] The first transformation unit is used to transform any coordinate in the three-dimensional continuous spatial coordinates into a feature vector of the target dimension using the position encoder in the initial material density field model.

[0092] The second transformation unit is used to map the feature vector to the base mass density at the corresponding position using the value prediction network in the initial mass density field model.

[0093] Optionally, in one embodiment of this application, the first conversion unit includes: an extraction subunit, a first mapping subunit, a first acquisition subunit, a second acquisition subunit, a third acquisition subunit, and a splicing subunit.

[0094] The extraction sub-unit is used to extract multiple mesh vertices of the first layer of the network from the position encoder based on arbitrary coordinates.

[0095] The first mapping subunit is used to map multiple grid vertices to the corresponding hash table indices using a preset hash function.

[0096] The first acquisition sub-unit is used to obtain the independent feature vector corresponding to each grid vertex by querying the hash table using the index.

[0097] The second acquisition subunit is used to fuse independent feature vectors using trilinear interpolation to obtain a single-layer feature vector with arbitrary coordinates at the first layer of the network.

[0098] The third acquisition subunit is used to acquire the single-layer feature vector corresponding to the next network layer, until the single-layer feature vectors corresponding to all network layers are obtained.

[0099] The splicing subunit is used to splice all single-layer feature vectors to obtain the feature vector of the target dimension.

[0100] Optionally, in one embodiment of this application, the second conversion unit includes a second mapping subunit and a processing subunit.

[0101] The second mapping subunit is used to map the feature vector to the initial value at the corresponding position using the value prediction network.

[0102] The processing subunit is used to scale the initial value using the base material specificity coefficient to obtain the base material density.

[0103] Optionally, in one embodiment of this application, the calculation module 300 includes an acquisition unit and a calculation unit.

[0104] The acquisition unit is used to acquire a set of base material densities corresponding to multiple coordinates on any ray path.

[0105] The computational unit is used to input a set of matrix material densities into a pre-built multi-energy computed tomography imaging physical forward model, so as to integrate the set of matrix material densities using the multi-energy computed tomography imaging physical forward model to obtain the predicted projection data of any ray path.

[0106] Optionally, in one embodiment of this application, the calculation expression of the physical forward model of multi-energy computed tomography imaging is as follows:

[0107] in, Indicates predicted projection data, This represents the normalized X-ray energy spectrum distribution, where E represents photon energy and K represents the number of base material species. This indicates the number of points sampled along the ray. Indicates the coordinates of the point sampled along the ray. Indicates the corresponding sampling step size. Indicates the first The known linear decay coefficient of the seed material. This indicates the network output at the sampling point. The density of the k-th basic substance at location , These represent the parameters of a hidden neural representation network.

[0108] Optionally, in one embodiment of this application, the reconstruction module 400 includes a construction unit and an optimization unit.

[0109] The building unit is used to construct a loss function using multi-energy projection data and predicted projection data.

[0110] The optimization unit is used to optimize the initial mass density field model using a loss function.

[0111] The loss function is calculated as follows:

[0112] in, This represents multi-energy projection data.

[0113] It should be noted that the foregoing explanation of the material reconstruction method embodiment of the multi-energy computed tomography imaging system also applies to the material reconstruction device of the multi-energy computed tomography imaging system in this embodiment, and will not be repeated here.

[0114] The material reconstruction device of the multi-energy computed tomography imaging system proposed in this application can acquire multi-energy projection data. Through an established initial material density field model, it performs the conversion between three-dimensional continuous spatial coordinates and the base material density. This eliminates the need to decouple material decomposition from image reconstruction, directly performing mapping calculations on the three-dimensional continuous spatial coordinates. This improves the signal-to-noise ratio of the reconstructed image, reduces the requirements for projection data, alleviates problems caused by incomplete projection data, and eliminates the need for pre-dividing discrete voxel grids, breaking the limitations of discrete grids. Furthermore, it uses the base material density to calculate and predict projection data, thereby combining multi-energy projection data and predicted projection data to optimize the initial material density field model, reconstructing a base material density map at any resolution. This results in higher applicability and reliability. Therefore, it solves the problems of related technologies that rely on discrete voxel grids, have high computational complexity, and whose reconstruction quality is limited by data quality, making it difficult to meet clinical needs.

[0115] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0116] When the processor 602 executes the program, it implements the material reconstruction method of the multi-energy computed tomography imaging system provided in the above embodiments.

[0117] Furthermore, electronic devices also include: Communication interface 603 is used for communication between memory 601 and processor 602.

[0118] The memory 601 is used to store computer programs that can run on the processor 602.

[0119] The memory 601 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0120] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0121] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0122] The processor 602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0123] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the material reconstruction method of the multi-energy computed tomography imaging system described above.

[0124] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the material reconstruction method of the multi-energy computed tomography imaging system provided in this embodiment of the invention.

[0125] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0126] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0127] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0128] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0129] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0130] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0131] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0132] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for material reconstruction in a multi-energy computed tomography imaging system, characterized in that, Includes the following steps: The target object is scanned using a multi-energy computed tomography imaging system to obtain the corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths; An initial material density field model of the target object is constructed using the spatial geometric parameters, and the three-dimensional continuous spatial coordinates are transformed into the corresponding basic material density using the initial material density field model. Using the density of the base material, the corresponding predicted projection data is calculated; By combining the multi-energy projection data and the predicted projection data, the initial matter density field model is optimized to obtain the actual matter density field model, and the actual matter density field model is used to reconstruct the base matter density map of the target object at any resolution.

2. The method according to claim 1, characterized in that, The process of converting the three-dimensional continuous spatial coordinates into the corresponding basic matter density using the initial matter density field model includes: Using the position encoder in the initial material density field model, any coordinate in the three-dimensional continuous spatial coordinates is transformed into a feature vector of the target dimension; Using the value prediction network in the initial material density field model, the feature vector is mapped to the base material density at the corresponding location.

3. The method according to claim 2, characterized in that, The step of converting any coordinate in the three-dimensional continuous spatial coordinates into a feature vector of the target dimension includes: Based on the arbitrary coordinates, extract multiple grid vertices of the first network layer from the position encoder; Using a preset hash function, multiple mesh vertices are mapped to corresponding hash table indices; The independent feature vector corresponding to each grid vertex is obtained by querying the hash table using the index; By fusing the independent feature vectors using trilinear interpolation, a single-layer feature vector at the first network layer is obtained for the arbitrary coordinates. Obtain the single-layer feature vector corresponding to the next network layer, until the single-layer feature vectors corresponding to all network layers are obtained; By concatenating all single-layer feature vectors, the feature vector of the target dimension is obtained.

4. The method according to claim 2, characterized in that, The step of mapping the feature vector to the base material density at the corresponding position includes: The value prediction network is used to map the feature vector to the initial value at the corresponding position; The initial value is scaled using a base material specificity coefficient to obtain the base material density.

5. The method according to claim 1, characterized in that, The calculation of the corresponding predicted projection data using the density of the base material includes: Obtain a set of base material densities corresponding to multiple coordinates along any ray path; A set of the matrix material densities is input into a pre-constructed multi-energy computed tomography imaging physical forward model, and the multi-energy computed tomography imaging physical forward model is used to integrate the set of the matrix material densities to obtain the predicted projection data of any of the ray paths.

6. The method according to claim 5, characterized in that, The calculation expression for the physical forward model of multi-energy computed tomography imaging is as follows: in, Indicates predicted projection data, This represents the normalized X-ray energy spectrum distribution, where E represents photon energy and K represents the number of base material species. This indicates the number of points sampled along the ray. Indicates the coordinates of the point sampled along the ray. Indicates the corresponding sampling step size. Indicates the first The known linear decay coefficient of the seed material. This indicates the network output at the sampling point. The density of the k-th basic substance at location, These represent the parameters of a hidden neural representation network.

7. The method according to claim 1, characterized in that, The step of combining the multi-energy projection data and the predicted projection data to optimize the initial matter density field model to obtain the actual matter density field model includes: A loss function is constructed using the multi-energy projection data and the predicted projection data; The initial mass density field model is optimized using the loss function. The calculation expression for the loss function is as follows: in, This refers to the multi-energy projection data.

8. A material reconstruction device for a multi-energy computed tomography imaging system, characterized in that, include: The acquisition module is used to scan the target object using a multi-energy computed tomography imaging system to acquire the corresponding multi-energy projection data, spatial geometric parameters, and three-dimensional continuous spatial coordinates of multiple ray paths. The conversion module is used to construct an initial material density field model of the target object using the spatial geometric parameters, so as to convert the three-dimensional continuous spatial coordinates into the corresponding basic material density using the initial material density field model; The calculation module is used to calculate the corresponding predicted projection data using the density of the base material; The reconstruction module is used to combine the multi-energy projection data and the predicted projection data to optimize the initial matter density field model, obtain the actual matter density field model, and use the actual matter density field model to reconstruct the base matter density map of the target object at any resolution.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, the processor executing the program to implement the material reconstruction method of a multi-energy computed tomography imaging system as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the material reconstruction method of the multi-energy computed tomography imaging system as described in any one of claims 1-7.

Images