Energy spectrum image processing method and device, equipment and storage medium

Abstract

The invention discloses an energy spectrum image processing method and device, equipment and a storage medium. The method comprises the following steps: determining an initial substance-based image meeting an initial quantitative precision condition; in the process of executing target image iteration reconstruction on the initial substance-based image, determining a current iteration result corresponding to the current diffusion step number, and obtaining current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, performing noise reduction processing on the current iteration result based on the current noise data so as to update the current iteration result, the initial diffusion step number in the target image iteration reconstruction process being determined based on an inverse diffusion initial point, and the inverse diffusion initial point being determined based on a data acquisition condition corresponding to the initial substance-based image; and under the condition that the iterative reconstruction of the target image is finished, taking the updated current iterative result as a target substance-based image. According to the embodiment of the invention, the substance decomposition quantitative result of the energy spectrum image obtained under various data acquisition conditions can be ensured to have relatively high precision.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular, to a spectral image processing method, apparatus, device, and storage medium. Background Art

[0002] The prior art usually performs spectral image reconstruction by modeling the physical process of data acquisition. This spectral image reconstruction method improves the quantitative accuracy of material decomposition by introducing a physical process model of data acquisition. However, the ill-conditioning degree of the material decomposition problem corresponding to the existing non-linear multi-energy X-ray measurement model is relatively high.

[0003] Therefore, a spectral image reconstruction method is needed to improve the quantitative accuracy of material decomposition in spectral images. Summary of the Invention

[0004] The present invention provides a spectral image processing method, apparatus, device, and storage medium to solve the problem that the prior art cannot ensure a high accuracy of material decomposition in spectral images.

[0005] According to one aspect of the present invention, a spectral image processing method is provided, including:

[0006] Determining an initial material basis image that meets the initial quantitative accuracy condition;

[0007] During the process of performing target image iterative reconstruction on the initial material basis image, determining a current iteration result corresponding to the current diffusion step, obtaining current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and performing noise reduction processing on the current iteration result based on the current noise data to update the current iteration result, where the starting diffusion step in the target image iterative reconstruction process is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition conditions corresponding to the initial material basis image;

[0008] When the target image iterative reconstruction ends, using the updated current iteration result as the target material basis image.

[0009] According to another aspect of the present invention, a spectral image processing apparatus is provided, including:

[0010] An image determination module, configured to determine an initial material basis image that meets the initial quantitative accuracy condition;

[0011] A reconstruction denoising module, configured to determine a current iteration result corresponding to a current diffusion step during performing target image iterative reconstruction on the initial material basis image, obtain current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and perform noise reduction processing on the current iteration result based on the current noise data to update the current iteration result, wherein a starting diffusion step in the target image iterative reconstruction process is determined based on an inverse diffusion starting point, and the inverse diffusion starting point is determined based on data acquisition conditions corresponding to the initial material basis image;

[0012] A result module, configured to use the updated current iteration result as the target material basis image when the target image iterative reconstruction ends.

[0013] According to another aspect of the present invention, there is provided an electronic device, which includes:

[0014] At least one processor; and

[0015] A memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is enabled to execute the energy spectrum image processing method according to any embodiment of the present invention.

[0017] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for implementing the energy spectrum image processing method according to any embodiment of the present invention when executed by a processor.

[0018] In the technical solution of the energy spectrum image processing method provided by the embodiments of the present invention, during performing target image iterative reconstruction on the initial material basis image, a current iteration result corresponding to a current diffusion step is determined, the current iteration result is input into a pre-trained diffusion noise model to obtain current noise data corresponding to the current diffusion step, and noise reduction processing is performed on the current iteration result based on the current noise data to update the current iteration result, achieving the technical effect of performing regularization constraint denoising on image iterative reconstruction based on a pre-trained diffusion noise model, so that the final target material basis image has high quantitative accuracy and very natural noise texture, fully meeting the clinical requirements for the quality of energy spectrum images; moreover, since the starting diffusion step in the target image iterative reconstruction process is determined based on an inverse diffusion starting point, and the inverse diffusion starting point is determined based on data acquisition conditions corresponding to the initial material basis image, the pre-trained diffusion noise model can process initial material basis images corresponding to various data acquisition conditions, meeting the energy spectrum image processing requirements of different scenarios.

[0019] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become readily understood from the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0021] Figure 1 is a flowchart of an energy spectrum image processing method provided according to an embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of an energy spectrum image acquisition technique provided according to an embodiment of the present invention;

[0023] Figure 3 is a schematic diagram of the results of different energy spectrum image processing methods provided according to an embodiment of the present invention;

[0024] Figure 4 is another flowchart of an energy spectrum image processing method provided according to an embodiment of the present invention;

[0025] Figure 5A is a schematic diagram of a first basis image provided according to an embodiment of the present invention;

[0026] Figure 5B is a schematic diagram of a second basis image provided according to an embodiment of the present invention;

[0027] Figure 6 is a flowchart of a noise reduction method provided according to an embodiment of the present invention;

[0028] Figure 7A is a schematic diagram of the structure of a diffusion noise model provided according to an embodiment of the present invention;

[0029] Figure 7B is a schematic diagram of the structure of a residual network provided according to an embodiment of the present invention;

[0030] Figure 8A is a schematic diagram of the structure of an energy spectrum image processing apparatus provided according to an embodiment of the present invention;

[0031] Figure 8B is another schematic diagram of the structure of an energy spectrum image processing apparatus provided according to an embodiment of the present invention;

[0032] Figure 9It is a schematic structural diagram of the C-arm CT imaging system provided by an embodiment of the present invention;

[0033] Figure 10A It is a schematic structural diagram of the diagnostic CT imaging system provided by an embodiment of the present invention;

[0034] Figure 10B It is a schematic structural diagram of another diagnostic CT imaging system provided by an embodiment of the present invention;

[0035] Figure 11 It is a schematic structural diagram of an electronic device for implementing the energy spectrum image processing method of an embodiment of the present invention. Detailed implementation manners

[0036] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0038] Figure 1 It is a flowchart of the energy spectrum image processing method provided by an embodiment of the present invention. This embodiment is applicable to the situation where the quantitative accuracy of material decomposition of the material-based image is improved by adopting a pre-trained diffusion noise model to constrain and reduce noise during the image iterative reconstruction process. This method can be executed by an energy spectrum image processing device, which can be implemented in the form of hardware and / or software, and the energy spectrum image processing device can be configured in the processor of an electronic device. As Figure 1 shown, the method includes:

[0039] S110. Determine an initial material-based image that meets the initial quantitative accuracy condition.

[0040] The initial material-based image that meets the initial quantitative accuracy condition refers to the material-based image that can meet the preliminary quantitative accuracy requirements.

[0041] Optionally, determine the material-based image to be processed corresponding to the projection data based on image domain decomposition; perform initial image iterative reconstruction on the material-based image to be processed using a model-based iterative reconstruction algorithm to obtain an initial material-based image that meets the initial quantitative accuracy condition.

[0042] Among them, any of the scanning methods such as the Figure 2 dual-source dual-detector method, fast tube voltage switching method, beam filtration method, photon counting method, and double-layer detector method shown can be used to obtain the projection data.

[0043] Specifically, after the material-based image to be processed is determined, perform initial image iterative reconstruction on the material-based image to be processed using the following formula (1). The number of iterations can be set to 50 times, 60 times, 70 times, etc. to obtain an initial material-based image that meets the initial quantitative accuracy condition. It can be understood that the number of iterative reconstructions can be determined based on the initial quantitative accuracy condition.

[0044] Among them, formula (1) is as follows:

[0045]

[0046] During the initial image iterative reconstruction process, t is the number of iterations. C j,t represents the current iteration result or the material-based image to be processed with pixel identifier j and iteration number t. Specifically, if the iteration number t is the initial iteration number, then C j,t is the material-based image to be processed, otherwise C j,t is the current iteration result corresponding to the iteration number t. Use formula (1) to perform one image iterative reconstruction on the material-based image to be processed or the current iteration result corresponding to the iteration number t to obtain the corresponding iteration result. The iteration number corresponding to this iteration result is Take this iteration result as the image corresponding to the iteration number t - 1 and make it enter the next image iterative reconstruction process.

[0047] In formula (1), the first derivative of Q(C j ; C j,t ) with respect to C j is a vector. When iterating to C j,t :

[0048]

[0049]

[0050] Among them, Indicates the number of photons obtained by the forward projection of C when iterating to t.

[0051] At Q(C j ; C j,t ), the second derivative corresponding to C j is a Hessian matrix. When iterating to C j,t :

[0052]

[0053]

[0054] Among them, the derivation process of formula (1) is as follows: Based on Bayes' formula, the following formula can be obtained:

[0055]

[0056] Among them, C = {c jk'} N×K represents the equivalent decomposition coefficient image of K material bases, and each equivalent decomposition coefficient image has N pixels. N represents the projection data collected (the number of photons of each ray detected by the detector). represents the projection data obtained by performing forward projection calculation on the material base image. is the data fidelity term. is the regularization term.

[0057] The Poisson model is used to model the data fidelity term, specifically as follows:

[0058]

[0059] Among them, i represents the index of the X-ray, k represents the index of the energy spectrum, and the maximum value is K. Then minimizing is to minimize According to the principle of transfer optimization, it is converted into a quadratic objective function Q(C; C t ), and this objective function is specifically as follows:

[0060]

[0061] Among them, c jk' represents the equivalent coefficient of the jth pixel and the k'th material base; ω ikε represents the "empty scan" photon number at the i'th ray, the k'th energy spectrum, and the ε'th energy level; b k'ε represents the attenuation coefficient of the k'th material base at the ε'th energy level; {a ij} represents the system matrix; γ ij represents a ij / ∑ j a ij ; qikε,t (·) is a quadratic scalar function with a vector parameter; t iε (C t ) = exp(-∑ k' b k'ε ∑ j a ij c jk' ).

[0062] By using the Newton iteration method to solve the above objective function for each pixel, the above formula (1) can be obtained.

[0063] S120. During the process of performing target image iterative reconstruction on the initial material basis image, determine the current iterative result corresponding to the current diffusion step number, obtain the current noise data corresponding to the current iterative result based on a pre-trained diffusion noise model, and perform noise reduction processing on the current iterative result based on the current noise data to update the current iterative result, where the starting diffusion step number in the target image iterative reconstruction process is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition conditions corresponding to the initial material basis image.

[0064] The diffusion noise model can be understood as a diffusion model based on noise control. Specifically, by controlling the inverse diffusion starting point in the target image iterative reconstruction process, the model denoising process is controlled.

[0065] Perform target image iterative reconstruction on the initial material basis image using formula (1). At this time, C j,t represents the current iterative result or the initial material basis image with the pixel identifier j and the diffusion step number t. Specifically, if the diffusion step number t is the starting diffusion step number, then C j,t is the initial material basis image that meets the initial quantitative accuracy condition, otherwise C j,t is the current iterative result corresponding to the diffusion step number t. It can be understood that performing one image iterative reconstruction on the initial material basis image or the current iterative result corresponding to the diffusion step number t using formula (1), the iterative result corresponding to this iterative result has a diffusion step number of

[0066] In one embodiment, the target image iterative reconstruction process can be specifically implemented as:

[0067] Step a1. Determine the starting diffusion step number corresponding to the initial material basis image, and perform image reconstruction corresponding to the starting diffusion step number on this initial material basis image to obtain the current iterative result.

[0068] Step a2. Input the current iterative result and the diffusion step number corresponding to the current iterative result into a pre-trained diffusion noise model to obtain the current noise data corresponding to the current iterative result.

[0069] Step a3: Denoise the current iteration result based on the current noise data, and use the denoised result as the updated current iteration result.

[0070] Step a4: Determine the current diffusion step by subtracting one from the number of steps. If the updated current diffusion step does not meet the set termination condition, perform image reconstruction corresponding to the current diffusion step on the current iteration result to obtain the current iteration result, and return to execute Step a2; if the updated current diffusion step meets the set termination condition, stop the image iterative reconstruction.

[0071] Exemplarily, if the current diffusion step corresponding to the current iteration result is t, then use formula (1) to perform iterative reconstruction on it to update the current iteration result, and the diffusion step corresponding to the updated iterative reconstruction result is Use the updated current iteration result and this Input the pre-trained diffusion noise model, and control the model to determine the current noise data corresponding to the updated current iteration result based on the model parameters corresponding to this ; then use the current noise data to denoise the updated current iteration result to update the current iteration result again, then set the current diffusion step to t - 1, and repeat the above process.

[0072] Among them, the set termination condition is that the current diffusion step is 0, that is, when it is detected that the updated current diffusion step is zero, terminate the above-mentioned target image iterative reconstruction process.

[0073] In this embodiment, the output of the pre-trained diffusion noise model is denoted as g Θ* (C t , t), Θ* is the network parameter of the pre-trained diffusion noise model, and this network parameter corresponds to the diffusion step. The correspondence between the network parameter and the diffusion step achieves the effect of performing targeted regularization denoising processing on each iteration result in the image iterative reconstruction through the pre-trained diffusion noise model, improving the accuracy of denoising the current iteration result, that is, the accuracy of the updated current iteration result, and the improvement of the accuracy of the updated current iteration result helps to improve the accuracy of the next image reconstruction, thereby improving the accuracy of the entire image iterative reconstruction.

[0074] In one embodiment, input the current iteration result and the diffusion step corresponding to the current iteration result into the pre-trained diffusion noise model to obtain the current noise data. Since the diffusion step corresponding to the current iteration result is also input while inputting the current iteration result, the model can determine the current noise data corresponding to the current iteration result based on the diffusion step corresponding to the current iteration result, which can improve the accuracy of noise data determination.

[0075] In order to further improve the denoising effect, in this embodiment, based on the data acquisition conditions corresponding to the initial material basis image, the inverse diffusion starting point in the iterative reconstruction process of the target image is determined, and the starting diffusion step number is determined based on this inverse diffusion starting point. By establishing the correspondence between the starting diffusion step number and the data acquisition conditions corresponding to the initial material basis image, the accuracy of determining the starting diffusion step number is improved, so that the technical solution of the energy spectrum image processing method provided by the embodiments of the present invention can flexibly process the energy spectrum data obtained based on various data acquisition conditions.

[0076] S130. When the iterative reconstruction of the target image ends, use the updated current iteration result as the target material basis image.

[0077] After the image iterative reconstruction is completed, use the latest updated current iteration result as the target material image.

[0078] Figure 3 It is a schematic diagram of the results of different energy spectrum image processing methods provided by the embodiments of the present invention. From Figure 3 It can be seen that compared with the true values of the water-based image and the iodine-based image, the error images of the water-based image and the iodine-based image determined by the image domain material decomposition method are the largest; the error images of the water-based image and the iodine-based image determined by the model-based iterative reconstruction method are the second largest, and the error images of the water-based image and the iodine-based image determined by the energy spectrum image processing method described in the embodiments of the present invention are the smallest. Among them, the true values of the water-based image and the iodine-based image are determined based on the first projection data obtained under high scan dose conditions. The processing objects of the image domain material decomposition method, the model-based iterative reconstruction method, and the energy spectrum image processing method described in the embodiments of the present invention are the same second projection data collected under low scan dose conditions. The acquisition conditions of the first projection data and the second projection data only differ in the scan dose.

[0079] In one embodiment, the system matrix is stored in a pre-sparse storage manner. Among them, the system matrix {a ij} represents the length of the i-th X-ray passing through the j-th pixel. Each X-ray only passes through a part of the pixels in the entire image, that is, only corresponds to a part of the pixels in the entire image, so the entire system matrix is very sparse. Moreover, for the same scanning system, the system matrix is unchanged during each scan. Therefore, adopting the pre-sparse storage method of the system matrix can improve the iteration speed of formula (1), thereby improving the speed of the entire energy spectrum image processing.

[0080] In one embodiment, the CUDA (Compute Unified Device Architecture) programming method is adopted to improve the speed of energy spectrum image processing. The GPU (Graphics Processing Unit) can greatly enhance the image processing speed, and CUDA programming allows developers to use NVIDIA GPUs for general computing. Therefore, adopting CUDA programming can greatly improve the speed of energy spectrum image processing and the accuracy of processing results.

[0081] In one embodiment, the parallel computing method is adopted to improve the speed of energy spectrum image processing. The three-dimensional image of a target object (patient) includes at least two two-dimensional slices, and the reconstruction of each two-dimensional slice is independent of each other. Therefore, the parallel image iterative reconstruction method is adopted to improve the speed of energy spectrum image processing.

[0082] In one embodiment, the method of upgrading and expanding the graphics card is adopted to improve the speed of energy spectrum image processing. To cope with parallel image iterative reconstruction, a graphics card with a fast running speed is selected for energy spectrum image processing. For a three-dimensional image including 400 slices, the reconstruction of one slice (with a resolution of 512×512) requires approximately 1G of video memory. At this time, 6 Nvidia H100 (80G) graphics cards can be selected for configuration.

[0083] After adopting the above speed improvement and optimization strategy, for the projection data collected under conventional data acquisition conditions, if it corresponds to 400 slices and the starting diffusion step is set to 250, the reconstruction time of all slices can be reduced to 95s. It should be noted that, under the condition of keeping other scanning conditions unchanged, the lower the scanning dose, the larger the starting diffusion step, and the longer the corresponding image processing time.

[0084] In the technical solution of the energy spectrum image processing method provided by the embodiments of the present invention, during the process of performing target image iterative reconstruction on the initial material basis image, the current iterative result corresponding to the current diffusion step is determined, and the current iterative result is input into a pre-trained diffusion noise model to obtain the current noise data corresponding to the current diffusion step. The current iterative result is denoised based on the current noise data to update the current iterative result, achieving the technical effect of regularizing and denoising the image iterative reconstruction based on the pre-trained diffusion noise model, so that the final target material basis image has high quantitative accuracy and very natural noise texture, fully meeting the clinical requirements for the quality of energy spectrum images. Moreover, since the starting diffusion step in the target image iterative reconstruction process is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition conditions corresponding to the initial material basis image, the pre-trained diffusion noise model can process the initial material basis images corresponding to various data acquisition conditions, meeting the energy spectrum image processing requirements of different scenarios.

[0085] Figure 4 It is another flowchart of the energy spectrum image processing method provided by the embodiments of the present invention. In this embodiment, the step of determining the starting diffusion step is added on the basis of the foregoing embodiment. As Figure 4 shown, the method includes:

[0086] S2101. Determine the initial material basis image that meets the initial quantitative accuracy condition.

[0087] S2102. Determine the target data acquisition conditions corresponding to the initial material basis image.

[0088] Among them, the target acquisition conditions include but are not limited to at least one of scanning dose, tube current, tube voltage, X-ray filtering material, and butterfly filter.

[0089] S2103. According to the corresponding relationship between the pre-created data acquisition conditions and the noise covariance matrix, determine the noise covariance matrix corresponding to the target data acquisition conditions.

[0090] Read the pre-created first correspondence file, which records at least two data acquisition conditions and the noise covariance matrix corresponding to each data acquisition condition, and read the noise covariance matrix corresponding to the target data acquisition conditions from the first correspondence file.

[0091] In one embodiment, the noise covariance matrix corresponding to each data acquisition condition is determined through the following steps:

[0092] Step b1: For each data acquisition condition in the data acquisition condition directory, normalize the first basis image and the second basis image corresponding to the current data acquisition condition. The first basis image and the second basis image are two basis images corresponding to the projection data of the calibration phantom, and these two basis images are determined based on image domain decomposition.

[0093] The first basis image and the second basis image of the calibration phantom are obtained as follows: For each scan protocol in the scan protocol directory, control the spectral imaging system to scan the calibration phantom based on the current scan protocol to obtain projection data, and determine the first basis image and the second basis image according to the projection data. Among them, the calibration phantom is the largest phantom that the spectral imaging system can scan, such as a water phantom with a diameter of 40 cm. Different scan protocols correspond to different data scan conditions.

[0094] Exemplarily, the first basis image of the calibration phantom is a water basis image (see Figure 5A ), and the second basis image of the calibration phantom is an iodine basis image (see Figure 5B ); Normalize the first basis image with a window of [800, 1200] mg / m; Normalize the second basis image with a window of [0, 15] mg / m.

[0095] Step b2: Select at least two uniform image patches from the first basis image and the second basis image respectively, and the positions of the at least two uniform image patches on the first basis image correspond to the positions of the at least two uniform image patches on the second basis image respectively.

[0096] Exemplarily, according to the same uniform image patch selection rule, use the same selection box to select n uniform image patches from the first basis image and the second basis image respectively; The position of the uniform image patch labeled 1 on the first basis image is the same as the position of the uniform image patch labeled 1 on the second basis image; The position of the uniform image patch labeled 2 on the first basis image is the same as the position of the uniform image patch labeled 2 on the second basis image, and so on. The position of the uniform image patch labeled n on the first basis image is the same as the position of the uniform image patch labeled n on the second basis image.

[0097] Step b3: Calculate the first noise covariance matrix between two corresponding uniform image patches on the first basis image and the second basis image to obtain at least two first noise covariance matrices.

[0098] Calculate the first noise covariance matrix between the uniform image blocks labeled 1 on the first basis image and the uniform image blocks labeled 1 on the second basis image, calculate the first noise covariance matrix between the uniform image blocks labeled 2 on the first basis image and the uniform image blocks labeled 2 on the second basis image, and so on, calculate the first noise covariance matrix between the uniform image blocks labeled n on the first basis image and the uniform image blocks labeled 2 on the second basis image.

[0099] Step b4: For each first noise covariance matrix among the at least two first noise covariance matrices, use the ratio of the current first noise covariance matrix to the maximum value of the current first noise covariance matrix as the second noise covariance matrix.

[0100] After each first noise covariance matrix is determined, determine the ratio of each data in the first noise covariance matrix to the maximum value in the first noise covariance matrix, and use this ratio as the second noise covariance matrix. The specific calculation formula is as follows:

[0101]

[0102] where S is the second noise covariance matrix, is the mean of the first noise covariance matrix.

[0103] Step b5: Use the mean of all the second noise covariance matrices as the noise covariance matrix corresponding to the current data acquisition condition.

[0104] In one embodiment, calculate the mean of all the second noise covariance matrices, and use this mean as the noise covariance matrix corresponding to the current data acquisition condition. This way of determining the covariance is simple and direct.

[0105] In one embodiment, remove the outliers from all the second noise covariance matrices, determine the mean of all the remaining second noise covariance matrices, and use this mean as the noise covariance matrix corresponding to the current data acquisition condition. This embodiment can ensure that the noise covariance matrix corresponding to the current data acquisition condition has high accuracy.

[0106] S2104: Determine the inverse diffusion starting point estimator according to the noise covariance matrix, and determine the starting diffusion step number in the iterative reconstruction process of the target image according to the inverse diffusion starting point estimator.

[0107] The reverse diffusion starting point (RSP) in the reverse diffusion denoising process using a trained diffusion noise model is also related to the noise covariance matrix. For example, the smaller the scanning dose, the higher the noise, the larger the noise covariance matrix, and the larger the noise covariance matrix, the larger the reverse diffusion starting point, and vice versa.

[0108] The starting point of inverse diffusion should be selected such that the signal-to-noise ratio of the image is approximately To this end, in this embodiment, an estimator of the starting point of inverse diffusion is determined based on the noise covariance matrix corresponding to the target data acquisition conditions, and can be specifically calculated by the following formula:

[0109]

[0110] During the iterative reconstruction process of the target image, the data fidelity term will introduce noise. Therefore, the amount of noise to be removed is more than that in the original image. So the starting point of inverse diffusion needs to be appropriately increased based on the estimator of the starting point of inverse diffusion. For example, according to different scan doses, 30 steps or 20 steps are increased, etc. If 30 steps are increased, then at this time where RSP is the starting point of inverse diffusion, is the estimator of the starting point of inverse diffusion, 30 is the correction amount of the starting point of inverse diffusion, is the mean value of the first noise covariance matrix.

[0111] In one embodiment, after the estimator of the starting point of inverse diffusion is determined, the diffusion step number corresponding to the estimator of the starting point of inverse diffusion is directly used as the starting diffusion step number. This method for determining the starting diffusion step number is simple and direct.

[0112] In one embodiment, after the estimator of the starting point of inverse diffusion is determined, the starting diffusion step number in the iterative reconstruction process of the target image is determined based on the target correction amount of the starting point of inverse diffusion. In this embodiment, for different estimators of the starting point of inverse diffusion or data acquisition conditions, the target correction amount remains unchanged. This method for determining the starting diffusion step number is also relatively simple.

[0113] In one embodiment, according to the correspondence relationship pre-created between the data acquisition conditions and the correction amount of the starting point of inverse diffusion, the correction amount of the starting point of inverse diffusion corresponding to the target data acquisition conditions is determined; the starting point of inverse diffusion is determined according to the estimator of the starting point of inverse diffusion and the correction amount of the starting point of inverse diffusion; the starting diffusion step number in the iterative reconstruction process of the target image is determined according to the starting point of inverse diffusion. This embodiment can match different correction amounts of the starting point of inverse diffusion for different data acquisition conditions, and the correction effect of the starting point of inverse diffusion is better, and the denoising effect of the trained diffusion noise model is better.

[0114] Specifically, read a pre-created second correspondence file, which records at least two data acquisition conditions and the inverse diffusion starting point correction amount corresponding to each data acquisition condition. From the second correspondence file, determine the inverse diffusion starting point correction amount corresponding to the target data acquisition condition. Optionally, the first correspondence file and the second correspondence file are one correspondence file, that is, one correspondence file records at least two data acquisition conditions and the noise covariance matrix and the inverse diffusion starting point correction amount corresponding to each data acquisition condition.

[0115] In one embodiment, the correspondence between the data acquisition conditions and the inverse diffusion starting point correction amount is determined based on a large amount of experimental data. The establishment of this correspondence greatly improves the accuracy of determining the inverse diffusion starting point and the accuracy of the starting diffusion step number determined based on the inverse diffusion starting point.

[0116] S220. During the process of performing target image iterative reconstruction on the initial material basis image, determine the current iteration result corresponding to the current diffusion step number, obtain the current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and perform noise reduction processing on the current iteration result based on the current noise data to update the current iteration result.

[0117] S230. When the target image iterative reconstruction ends, use the updated current iteration result as the target material basis image.

[0118] The embodiment of the present invention determines the inverse diffusion starting point based on the data acquisition conditions of the initial material basis image, and determines the starting diffusion step number for target image iterative reconstruction according to the inverse diffusion starting point, achieving the technical effect of matching an accurate starting diffusion step number for the initial material basis image, improving the accuracy of determining the starting diffusion step number, and thus improving the image quality of the denoised material basis image.

[0119] Figure 6 It is a flowchart of the noise reduction method provided by the embodiment of the present invention. This embodiment is used to refine the noise reduction processing flow in the foregoing embodiment. As Figure 6 shown, the method includes:

[0120] S3201. Determine the noise weighting coefficient corresponding to the current noise data. The noise weighting coefficient is determined based on the estimated diffusion step number, and the estimated diffusion step number is determined based on the set diffusion parameters.

[0121] Determining the noise weighting coefficient corresponding to the current noise data is essentially determining the noise weighting coefficient of the diffusion step number corresponding to the current noise data. In one embodiment, the noise weighting coefficient is determined through the following steps:

[0122]

[0123] Among them, is the diffusion step estimator, which can be determined by the following formula:

[0124]

[0125] β max and β min are the set diffusion parameters, which are 0.02 and 0.0001 respectively. T is the set total diffusion steps, which can be set to 1000.

[0126] S3202. Perform noise reduction processing on the current iteration result based on the current noise data and the noise weighting coefficient to update the current iteration result.

[0127] Determine the product of the current noise data and the noise weighting coefficient, and use this product to update the current noise data; determine the difference between the current iteration result and the updated current noise data, and use this difference as the updated current iteration result.

[0128] In one embodiment, the above denoising process is implemented by the following formula:

[0129]

[0130] As can be seen from the foregoing embodiments, is the current iteration result. Use this formula to perform noise reduction processing on the current iteration result. The denoised current iteration result, that is, the diffusion steps corresponding to the updated current iteration result is t - 1, and t - 1 is exactly the diffusion steps corresponding to the next image iteration process.

[0131] In the embodiment of the present invention, by using the noise weighting coefficient corresponding to the current noise data, the accuracy of the noise data corresponding to the current iteration result is improved, thereby improving the accuracy of denoising the current iteration result and the accuracy of image iterative reconstruction, that is, improving the image quality of the target substance base image.

[0132] Figure 7A is the flowchart of the energy spectrum image processing method provided by the embodiment of the present invention. This embodiment details the diffusion noise model in the foregoing embodiments.

[0133] As Figure 7A shown, the diffusion noise model is a symmetric structure, including an encoder on the left, a decoder on the right, and a bottleneck layer for connecting the encoder and the decoder.

[0134] The encoder includes four residual networks and pooling layers connected to the ends of each residual network. The residual networks are used to determine the target residual results of the received feature maps, and the pooling layers are used to downsample the received feature maps. The kernel size of each pooling layer is K = 2, and the stride is S = 2. After four times of residual calculation and pooling, the current iteration result of 32×32 becomes an image of 256×256.

[0135] The decoder includes upsampling layers and residual networks connected to the ends of each upsampling layer. The upsampling layers are used to perform an upsampling operation with a stride of 2 and a size of 3×3 on the received feature maps. The upsampling result and the last encoding result of the corresponding layer are concatenated together to obtain a concatenated result, and the concatenated result is fed into the corresponding residual network. Each residual network performs two convolution operations on the received concatenated result. The last residual network feeds the last convolution result into a normalization network, which includes a GN (Group Normalization) layer, a GELU activation function, and a 3×3 convolution kernel.

[0136] The bottleneck layer connects the encoder and the decoder through two residual networks.

[0137] Among them, as Figure 7B shown, the residual network includes two inputs. One input is the embedded time information, such as the diffusion step number, and the other input is the received feature map. The embedded time information is processed through a GELU activation function and a linear layer to obtain a time result. The received feature map is processed by a residual network to obtain a residual result; the combined result of the time result and the residual result is determined; the combined result is processed by a residual network to obtain an intermediate residual result; the target residual result is determined according to the intermediate residual result and the received feature map. This target residual result is the output data of the residual network.

[0138] The diffusion noise model in this embodiment uses the following loss function for model training, specifically:

[0139]

[0140] Among them,

[0141]

[0142] C s,t is the initial substance base image in the s-th group of samples in the sample set, is the diffusion step number estimator, Δ is the random noise sampled by the Monte Carlo method, and it follows a Gaussian distribution with a mean of 0 and a covariance of S. Among them, the determination method of the covariance S can refer to the noise covariance matrix determination method described in the foregoing embodiment, and this embodiment will not elaborate here.

[0143] Embodiments of the present invention describe in detail the structure and loss function of a diffusion noise model. The combined use of the model with the loss function improves the noise prediction accuracy and generalization ability of the pre-trained diffusion noise model.

[0144] Figure 8A FIG. is a schematic structural diagram of an energy spectrum image processing device provided by an embodiment of the present invention. As Figure 8A shown, the device includes:

[0145] An image determination module 51, configured to determine an initial substance-based image that meets the initial quantitative accuracy condition;

[0146] A reconstruction denoising module 52, configured to determine a current iteration result corresponding to the current diffusion step during the process of performing target image iterative reconstruction on the initial substance-based image, obtain current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and perform denoising processing on the current iteration result based on the current noise data to update the current iteration result, where the starting diffusion step of the inverse diffusion starting point in the target image iterative reconstruction process is determined, and the inverse diffusion starting point is determined based on the data acquisition conditions corresponding to the initial substance-based image;

[0147] A result module 53, configured to use the updated current iteration result as a target substance-based image when the target image iterative reconstruction ends.

[0148] In one embodiment, the reconstruction denoising module 52 is specifically configured to:

[0149] Input the current iteration result and the diffusion step corresponding to the current iteration result into a pre-trained diffusion noise model to obtain current noise data.

[0150] In one embodiment, the reconstruction denoising module 52 is specifically configured to:

[0151] Determine a noise weighting coefficient corresponding to the current noise data, where the noise weighting coefficient is determined based on a diffusion step estimator, and the diffusion step estimator is determined based on set diffusion parameters;

[0152] Perform denoising processing on the current iteration result based on the current noise data and the noise weighting coefficient to update the current iteration result.

[0153] In one embodiment, the loss function for training the noise model is:

[0154]

[0155] where C s,t is the substance-based image at time t in the s-th group of samples, is the estimated value of the diffusion step, and Δ is the random noise sampled by the Monte Carlo method. This random noise follows a Gaussian distribution with a mean of 0 and a covariance of S, where S is the noise covariance matrix, and the noise covariance matrix is determined based on the data acquisition conditions corresponding to the corresponding samples.

[0156] In one embodiment, as Figure 8B shown, the device further includes a starting diffusion step module 54, and the starting diffusion step module 54 includes:

[0157] A data acquisition condition unit for determining the target data acquisition conditions corresponding to the initial substance-based image;

[0158] A noise covariance matrix unit for determining the noise covariance matrix corresponding to the target data acquisition conditions according to the correspondence between the pre-created data acquisition conditions and the noise covariance matrix;

[0159] A starting diffusion step unit determines the inverse diffusion starting point estimator according to the noise covariance matrix, and determines the starting diffusion step in the iterative reconstruction process of the target image according to the inverse diffusion starting point estimator.

[0160] In one embodiment, the noise covariance matrix unit is further used for:

[0161] Determining the inverse diffusion starting point correction amount corresponding to the target data acquisition conditions according to the correspondence between the pre-created data acquisition conditions and the inverse diffusion starting point correction amount;

[0162] The starting diffusion step unit is specifically used for:

[0163] Determining the inverse diffusion starting point according to the inverse diffusion starting point estimator and the inverse diffusion starting point correction amount;

[0164] Determining the starting diffusion step in the iterative reconstruction process of the target image according to the inverse diffusion starting point.

[0165] In one embodiment, the noise covariance matrix unit is used for:

[0166] For each data acquisition condition in the data acquisition condition catalog, normalize the first basis image and the second basis image corresponding to the current data acquisition condition respectively. The first basis image and the second basis image are two basis images corresponding to the projection data of the calibration phantom, and the two basis images are determined based on the image domain decomposition.

[0167] Select at least two uniform image patches from the first basis image and the second basis image respectively, and the positions of the at least two uniform image patches on the first basis image correspond to the positions of the at least two uniform image patches on the second basis image respectively;

[0168] Calculate a first noise covariance matrix between two uniform image patches with a corresponding relationship on the first basis image and the second basis image to obtain at least two first noise covariance matrices;

[0169] For each first noise covariance matrix among the at least two first noise covariance matrices, determine a ratio of the current first noise covariance matrix to the maximum value of the current first noise covariance matrix as a second noise covariance matrix;

[0170] Take the mean of all the second noise covariance matrices as the noise covariance matrix corresponding to the current data acquisition condition.

[0171] In one embodiment, the image determination module is specifically configured to:

[0172] Determine a to-be-processed substance basis image corresponding to projection data based on image domain decomposition;

[0173] Perform iterative reconstruction on the to-be-processed substance basis image a set number of times using a model-based iterative reconstruction algorithm to obtain an initial substance basis image that meets the initial quantitative accuracy condition.

[0174] In the technical solution of the spectral image processing apparatus provided by the embodiments of the present invention, during the process of performing target image iterative reconstruction on the initial substance basis image, a current iteration result corresponding to the current diffusion step is determined, the current iteration result is input into a pre-trained diffusion noise model to obtain current noise data corresponding to the current diffusion step, and the current iteration result is denoised based on the current noise data to update the current iteration result, achieving the technical effect of regularizing and denoising the image iterative reconstruction based on the pre-trained diffusion noise model, such that the final substance basis image has a high quantitative accuracy and very natural noise texture, fully meeting the clinical requirements for the quality of spectral images; moreover, since the starting diffusion step in the target image iterative reconstruction process is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition condition corresponding to the initial basis image, the pre-trained diffusion noise model can process the initial substance basis images corresponding to various data acquisition conditions, meeting the spectral image processing requirements of different scenarios.

[0175] The spectral image processing apparatus provided by the embodiments of the present invention can execute the spectral image processing method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution of the method.

[0176] Figure 9It is a schematic structural diagram of a C-arm CT imaging system provided by another embodiment of the present invention. The system includes a gantry 1211, a detector 1212, a couch 1214, an X-ray tube 1215, a C-arm drive shaft 1216, a rotating shaft 1217, and a base 1219. The X-ray tube 1215 and the detector 1212 are equipped at both ends of the C-shaped gantry 1211, and the central connection line between the two is perpendicular to its rotation axis 1218. The C-shaped gantry 1211 rotates around the rotation axis 1218, so as to capture the image data of the patient 1213 on the couch at different projection angles. The X-ray tube 1215 is controlled by an X-ray generator 123 for its current, voltage, exposure time, etc. The projection data collected by the detector 1212 is transmitted to a computer device by a communication system 126. The gantry 1211 is connected to the C-arm drive shaft 1216, and its power is provided by the rotating shaft 1217. The base 1219 is responsible for bearing the weight. The C-arm control unit 121 controls the rotation speed, angle, position, etc. of the gantry 1211. The main shaft control unit 122 is connected to the base 1219 and provides power support for the entire C-arm system. The X-ray generator 123 controls the current, voltage, and exposure time of the X-ray tube 1215. The data acquisition system 124 coordinates the gantry 1211, the detector 1212, and the X-ray generator 1215, and collects the acquired data. The couch control system 125 controls the position and movement speed of the couch 1214 to achieve different scanning trajectories for the patient 1213. The communication system 126 is connected to the C-arm control unit 121, the main shaft control unit 122, the X-ray generator 124, the data acquisition system 124, and the couch control system 125, and transmits the collected projection data to the memory of the computer device 2.

[0177] Figure 10A and Figure 10B shows a schematic structural diagram of another CT imaging system. This CT imaging system is a diagnostic CT. Compared with the C-arm CT, its gantry 1211 is annular, the detector 1212 and the X-ray tube 1215 are both arranged on the gantry and are relatively distributed. The couch 1214 enters and exits the aperture of the gantry under the control of the couch controller 125, and the gantry drives the detector 1212 and the X-ray tube 1215 to move around the couch 1214.

[0178] Figure 11 shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0179] As Figure 11As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0180] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0181] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the energy spectrum image processing method.

[0182] In some embodiments, the energy spectrum image processing method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the energy spectrum image processing method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the energy spectrum image processing method in any other appropriate way (e.g., by means of firmware).

[0183] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0184] The computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, a special purpose computer, or other programmable data processing device, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer programs can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine, or entirely on the remote machine or server.

[0185] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0186] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0187] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0188] The computing system can include a client and a server. The client and the server are generally far from each other and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0189] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0190] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An energy spectrum image processing method, characterized in that, it includes: Determine an initial material basis image that meets the initial quantitative accuracy conditions; During the process of performing target image iterative reconstruction on the initial material basis image, determine the current iteration result corresponding to the current diffusion step, obtain the current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and perform noise reduction processing on the current iteration result based on the current noise data to update the current iteration result, where the starting diffusion step in the target image iterative reconstruction process is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition conditions corresponding to the initial material basis image; When the target image iterative reconstruction ends, use the updated current iteration result as the target material basis image.

2. The method according to claim 1, characterized in that, The obtaining of the current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model includes: Input the current iteration result and the diffusion step corresponding to the current iteration result into a pre-trained diffusion noise model to obtain the current noise data.

3. The method according to claim 1, characterized in that, The performing of noise reduction processing on the current iteration result based on the current noise data to update the current iteration result includes: Determine a noise weighting coefficient corresponding to the current noise data, the noise weighting coefficient is determined based on a diffusion step estimator, and the diffusion step estimator is determined based on set diffusion parameters; Perform noise reduction processing on the current iteration result based on the current noise data and the noise weighting coefficient to update the current iteration result.

4. The method according to claim 1, characterized in that, The loss function for training the diffusion noise model is: Among them, C s,t is the material base image at time t in the s-th group of samples, is the diffusion step estimator, Δ is the random noise sampled by the Monte Carlo method, and the random noise follows a Gaussian distribution with a mean of 0 and a covariance of S, where S is the noise covariance matrix, and the noise covariance matrix is determined based on the corresponding data acquisition conditions of the corresponding samples.

5. The method according to claim 1, characterized in that, The starting diffusion step in the target image iterative reconstruction process is determined through the following steps: Determine the target data acquisition conditions corresponding to the initial material basis image; According to the corresponding relationship between the pre-created data acquisition conditions and the noise covariance matrix, determine the noise covariance matrix corresponding to the target data acquisition conditions; Determine an inverse diffusion starting point estimator according to the noise covariance matrix, and determine the starting diffusion step in the target image iterative reconstruction process according to the inverse diffusion starting point estimator.

6. The method according to claim 5, characterized in that, When determining the noise covariance matrix corresponding to the target data acquisition conditions according to the corresponding relationship between the pre-created data acquisition conditions and the covariance matrix, it further includes: Determine the inverse diffusion starting point correction amount corresponding to the target data acquisition conditions according to the corresponding relationship between the pre-created data acquisition conditions and the inverse diffusion starting point correction amount; The determining of the starting diffusion step in the target image iterative reconstruction process according to the inverse diffusion starting point estimator includes: Determine the inverse diffusion starting point according to the inverse diffusion starting point estimator and the inverse diffusion starting point correction amount; Determine the starting diffusion step number in the iterative reconstruction process of the target image according to the inverse diffusion starting point.

7. The method according to claim 5, wherein, determine the noise covariance matrix corresponding to each data acquisition condition through the following steps: For each group of conditions in the data acquisition condition directory, normalize the first basis image and the second basis image corresponding to the current data acquisition condition respectively. The first basis image and the second basis image are two basis images corresponding to the projection data of the calibration phantom, and the two basis images are determined based on image domain decomposition; Select at least two uniform image patches from the first basis image and the second basis image respectively, and the positions of the at least two uniform image patches on the first basis image correspond to the positions of the at least two uniform image patches on the second basis image respectively; Calculate the first noise covariance matrix between the two uniform image patches with corresponding relationships on the first basis image and the second basis image to obtain at least two first noise covariance matrices; For each first noise covariance matrix in the at least two first noise covariance matrices, use the ratio of the current first noise covariance matrix to the maximum value of the current first noise covariance matrix as the second noise covariance matrix; Take the mean value of all the second noise covariance matrices as the noise covariance matrix corresponding to the current data acquisition condition.

8. The method according to claim 1, wherein, the determination of the initial substance basis image that meets the initial quantitative accuracy condition includes: Determine the substance basis image to be processed corresponding to the projection data based on image domain decomposition; Perform initial image iterative reconstruction on the substance basis image to be processed using a model-based iterative reconstruction algorithm to obtain an initial substance basis image that meets the initial quantitative accuracy condition.

9. An energy spectrum image processing device, wherein, comprising: An image determination module for determining an initial substance basis image that meets the initial quantitative accuracy condition; A reconstruction denoising module for determining the current iteration result corresponding to the current diffusion step number during the iterative reconstruction of the target image for the initial substance basis image, obtaining the current noise data corresponding to the current iteration result based on a pre-trained diffusion noise model, and performing denoising processing on the current iteration result based on the current noise data to update the current iteration result. Among them, the starting diffusion step number in the iterative reconstruction process of the target image is determined based on the inverse diffusion starting point, and the inverse diffusion starting point is determined based on the data acquisition condition corresponding to the initial substance basis image; A result module for taking the updated current iteration result as the target substance basis image when the iterative reconstruction of the target image ends.

10. An electronic device, wherein, the electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the energy spectrum image processing method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions for implementing the energy spectrum image processing method according to any one of claims 1-8 when the instructions are executed by a processor.

Images