Photon counting energy spectrum CT image reconstruction method and system based on feature learning

Abstract

The invention discloses a photon counting energy spectrum CT image reconstruction method and system based on feature learning, and belongs to the technical field of image processing, and the method comprises the steps: S1, collecting a photon counting energy spectrum CT image; s2, on the basis of the group sparse residual error of the CT image and the sparse feature of the image block, constructing a dual prior regular; s3, constructing a CT image reconstruction model based on the double prior regularities; and S4, performing numerical iteration solution on the CT image reconstruction model by adopting an alternating direction multiplier method to obtain a reconstructed CT image. According to the method, the problem of how to reconstruct the high-quality energy spectrum CT image from the X-ray energy spectrum CT projection data with the low signal-to-noise ratio and the limited scanning angle is solved, artifacts and noise occurring in photon counting energy spectrum CT scanning imaging at the limited angle are overcome, and the high-quality CT image with both image quality and radiation protection is reconstructed.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a photon counting spectral CT image reconstruction method and system based on feature learning. Background Art

[0002] Although spectral CT technology based on photon counting detectors has made great progress in recent years, it still has problems that need to be studied and solved. These include: electron pile-up effect, charge sharing phenomenon, and Compton scattering caused by the interaction between X-rays and matter. Photon counting detectors have other effects such as inconsistent corresponding responses, trapping effect, polarization effect, etc. Due to these effects of photon counting detectors, the signal-to-noise ratio of the projection data collected by CT systems using photon counting detectors is low, which will reduce the quality of image reconstruction. Whether in medical examination and diagnosis, target detection or computer fields, the impact of these artifacts and noise on imaging results cannot be ignored. The spectral CT images reconstructed by existing methods are severely degraded, such as severe noise in images of different channels, loss of some structural information in the image, and obvious artifacts.

[0003] In practical applications, a more flexible spectral CT scanning system is needed. Repeated CT scans are often required in clinical practice, impacting diagnostic needs. Furthermore, considering the age of patients (older patients often have renal insufficiency, and children are still developing organs) and the excessive radiation doses radiologists receive, optimizing X-ray scan dose becomes a critical issue, given the harmful effects of X-ray radiation on life and health. This leads to the ill-posed inverse problem of low-dose spectral CT image reconstruction.

[0004] The present invention investigates spectral CT scanning with equally spaced angles distributed along a circular trajectory. The corresponding photon-counting spectral CT reconstruction method is investigated for this scanning method. This scanning method results in a severe shortage of photons per energy channel, which reduces the signal-to-noise ratio of the projection data, resulting in significant image artifacts, noise, and other image quality issues.

[0005] Therefore, it is necessary to design a photon counting spectral CT image reconstruction method and system based on feature learning to address the above problems. Summary of the Invention

[0006] The purpose of the present invention is to address the problem of insufficient photon numbers in each narrow energy spectrum channel in the acquired energy spectrum CT image data, which leads to severe noise, missing structural information and obvious artifacts in the reconstructed image, thereby affecting the accuracy of material decomposition. The present invention provides a photon counting energy spectrum CT image reconstruction method based on feature learning, which overcomes the artifacts and noise that appear in the scanning imaging of photon counting energy spectrum CT and reconstructs high-quality CT images that take into account both image quality and radiation protection.

[0007] According to one aspect of the present specification, a method for reconstructing a photon counting spectral CT image based on feature learning is provided, comprising:

[0008] S1, collecting photon counting spectral CT images;

[0009] S2, constructing a dual prior regularizer based on the group sparse residual of CT images and the sparse features of image blocks;

[0010] S3. Based on the dual prior regularizer, a CT image reconstruction model is constructed, including: obtaining the low-rank characteristics of CT images of different energy channels through group sparse residuals and gradient image sparseness; performing sparsity constraints on the smoothness of the photon counting energy spectrum CT image space and the artifact characteristics to obtain the internal characteristics of the single energy channel image; based on the low-rank characteristics of CT images of different energy channels and the internal characteristics of the single energy channel image, the dual prior regularizer is used as a regularization term, and combined with the energy spectrum CT physical imaging model to obtain the CT image reconstruction model;

[0011] S4. Using the alternating direction multiplier method, the CT image reconstruction model is numerically iteratively solved to obtain a reconstructed CT image.

[0012] Furthermore, the S1 also includes: extracting projection data from the full-angle scanning data through computer simulation to generate CT images of different energy channels.

[0013] Furthermore, the S2 includes:

[0014] Based on the non-local self-similarity prior of CT images and self-supervised learning, the group sparsity coefficient of each original image group is estimated;

[0015] The group sparse coefficients of the corresponding degraded image group are applied for approximate estimation, and the group sparse residual constraint is used as a priori condition;

[0016] The nuclear norm and the zero norm of tight frame wavelet transform coefficients are used to replace the low-rank and sparse constraint functions respectively to construct a dual prior regularizer.

[0017] Furthermore, the S4 includes:

[0018] By analyzing the convexity and concavity of the CT image reconstruction model, the reconstruction problem is transformed into multiple sub-problems;

[0019] Each subproblem is numerically iterated and solved using the alternating direction multiplier method to obtain the reconstructed CT image.

[0020] Furthermore, after obtaining the reconstructed CT image, the method further includes:

[0021] Develop system software for photon counting spectral CT imaging based on a mathematical software platform and conduct experimental verification;

[0022] The imaging quality is studied based on the data scanned by the energy spectrum CT system, and real data verification is performed.

[0023] According to one aspect of the present specification, a photon counting spectral CT image reconstruction system based on feature learning is provided, comprising:

[0024] A data acquisition module, used for acquiring photon counting spectral CT images;

[0025] A regularizer construction module is used to construct a dual prior regularizer based on the group sparse residual of the CT image and the sparse features of the image block;

[0026] A CT image reconstruction model construction module is used to construct a CT image reconstruction model based on a dual prior regularizer, including: obtaining low-rank characteristics of CT images of different energy channels through group sparse residuals and gradient image sparseness; performing sparsity constraints on the smoothness of the photon counting energy spectrum CT image space and artifact characteristics to obtain internal characteristics of single energy channel images; based on the low-rank characteristics of CT images of different energy channels and the internal characteristics of single energy channel images, using the dual prior regularizer as a regularizer and combining it with the energy spectrum CT physical imaging model to obtain a CT image reconstruction model;

[0027] The CT image reconstruction module is used to adopt the alternating direction multiplier method to perform numerical iterative solution on the CT image reconstruction model to obtain the reconstructed CT image.

[0028] According to one aspect of the present specification, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the feature learning-based photon counting spectral CT image reconstruction method when executing the computer program.

[0029] According to one aspect of the present specification, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the photon counting spectral CT image reconstruction method based on feature learning are implemented.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The embodiments of the present invention address the serious noise problem in the reconstruction results of photon counting spectral CT images in different energy channels due to insufficient photon numbers. Targeted processing is performed on the noise in CT images of different energy channels, solving the problem of how to reconstruct high-quality spectral CT images from X-ray spectral CT projection data with low signal-to-noise ratio, thereby reducing radiation dose and improving clinical diagnostic quality.

[0032] 2. Compared with traditional spectral CT imaging technology, the embodiments of the present invention overcome the artifacts and noise that occur in photon counting spectral CT scanning imaging at restricted angles, reconstruct high-quality CT images that take into account both image quality and radiation protection, and improve the diagnostic level for subsequent application evaluation such as material decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 A schematic diagram of a technical route of an embodiment of the present invention; DETAILED DESCRIPTION

[0035] It should be noted that:

[0036] Full-angle scanning data refers to the projection data obtained by the imaging system performing 360° scanning and imaging around the scanning target.

[0037] The energy spectrum CT physical imaging model described in the specification of the present invention can be implemented using existing technologies, and the present invention will not elaborate on this.

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] The present invention utilizes a photon-counting spectral CT imaging system as a platform. Radiation scanning of the subject is performed by controlling the rotational angle range of the X-ray source, with the subject remaining stationary throughout the imaging process. Due to the narrowing of the spectral channels, the number of photons within each channel is very limited, making image reconstruction a highly ill-posed inverse problem. Therefore, to improve spectral CT imaging quality and material decomposition, a practical and effective spectral CT image reconstruction method is proposed.

[0040] Specifically, this paper aims to explore the high correlation between images of different energy channels and the internal features of single-energy channel images, construct regularizers that characterize these features, and couple them with the spectral CT physical imaging model to propose a new spectral CT image reconstruction model. A numerical solution for the model is provided, and the convergence of the solution is analyzed. The specific contents of the proposed research are as follows:

[0041] (1) Study the characteristics of energy channel and image space of spectral CT images: Since the reconstructed images of different energy channels of photon counting spectral CT correspond to the same object, they have the same structural characteristics. At the same time, since the materials of the structural images in the local area are mostly one or two components, the corresponding density attenuation coefficients are approximately linear. In order to effectively remove the artifacts and noise that appear in the reconstruction, the prior characteristics of spectral CT images are studied from two dimensions: energy channel and image space. On the one hand, the group sparse residual and gradient image sparse are jointly applied to the local blocks, and the nuclear norm is used to explore the low-rank information of the energy spectrum dimension of the local blocks to study the local low-rank sparse characteristics of the image blocks of different energy channel images. On the other hand, the smoothness of the image space and the characteristics of the artifacts are constrained by sparsity to study the internal characteristics of the single energy channel image. By utilizing the low rank of the local blocks of spectral CT images and the sparsity of the image space, artifacts and noise can be further suppressed and the edge structure and other features of the image can be retained.

[0042] (2) A photon counting spectral CT image reconstruction method based on feature learning is proposed: by introducing the concept of group sparse residual, the feature space of spectral CT images is characterized. On the one hand, the group sparse residual is combined with the gradient image sparseness to apply to local patches to study the low-rank characteristics of images of different energy channels; on the other hand, the sparsity constraints are imposed on the smoothness of the image space and the characteristics of the artifacts to study the internal characteristics of the single energy channel image. By using the image non-local self-similarity prior and the self-supervised learning scheme, the group sparse coefficient of each original image group is estimated, and then the group sparse coefficient of the corresponding degraded image group is applied to approximate the estimate, with the group sparse residual constraint as the prior condition; then, the nuclear norm and the zero norm of the tight frame wavelet transform coefficient are used to replace the low-rank and sparse constraint functions respectively, and a double prior constraint regularizer is constructed; finally, the alternating direction multiplier method is used to optimize the variables. In order to construct a new reconstruction model based on the physical imaging mechanism of spectral CT, a double-prior-constrained regularizer was used as the regularization term to construct an optimization model suitable for photon counting spectral CT image reconstruction. An effective numerical solution was designed using the alternating direction multiplier method, and the convergence of the solution was analyzed.

[0043] like Figure 1As shown, an embodiment of the present invention provides a photon counting spectral CT image reconstruction method based on feature learning, including: S1, collecting photon counting spectral CT images; S2, constructing a dual prior regularizer based on the group sparse residual of the CT image and the sparse features of the image block; S3, constructing a CT image reconstruction model based on the dual prior regularizer, including: obtaining the low-rank characteristics of CT images of different energy channels through group sparse residual and gradient image sparseness; performing sparsity constraints on the smoothness of the photon counting spectral CT image space and the artifact characteristics to obtain the internal features of the single energy channel image; based on the low-rank characteristics of the CT images of different energy channels and the internal features of the single energy channel image, using the dual prior regularizer as a regularization term, and combining it with the spectral CT physical imaging model to obtain a CT image reconstruction model; S4, using the alternating direction multiplier method to numerically iteratively solve the CT image reconstruction model to obtain a reconstructed CT image.

[0044] Specifically, the present invention explores image reconstruction techniques that can improve imaging quality and material decomposition by exploring the characteristics of different energy channels and single-energy images in spectral CT. Specifically, the following steps are included:

[0045] Step 1: Acquire photon-counting spectral CT image data. The computer simulation model utilizes the Shepp-Logan model, NCAT model, FORBILD model database, and the CT image database of the affiliated hospital's radiology imaging center. Projection data are extracted from full-angle scan data through computer simulation to generate photon-counting spectral CT image data for different energy spectrum segments. The hospital's CT image sequence database is also utilized to provide prior feature data support for subsequent research. Furthermore, collaboration with other research institutes is underway to acquire actual spectral CT projection data.

[0046] Step 2: Study the photon counting spectral CT image reconstruction method based on feature learning. Feature learning and regularizer design: Study the sparse features of the group sparse residual and image blocks, and design corresponding dual prior regularizers. By using the weighted average of non-local self-similar blocks to calculate the estimated value of the sparse coefficient of the original image group, the reconstruction accuracy is improved. Optimized reconstruction model with coupled feature learning: First, the image is subjected to low-rank sparse constraints to suppress noise while retaining the low-rank structure and details of the spectral CT image. Then, the noise in the CT images of different energy channels is processed in a targeted manner. Based on the low-rank structure and sparsity characteristics of photon counting spectral CT images, a dual prior regularizer is proposed. Combined with the spectral CT physical imaging model, an optimized reconstruction model is constructed, and a numerical solution is given. Finally, the model is solved: The convexity and concavity of the model are analyzed, the reconstruction optimization problem is transformed into multiple subproblems, its iterative solution method is studied, and the effectiveness and stability of the algorithm are verified.

[0047] Step 3: Feature learning-based reconstruction method. By introducing the concept of group sparse residual, the feature space of spectral CT images is characterized. Using the image non-local self-similarity prior and a self-supervised learning scheme, an estimate of the group sparse coefficient of each original image group is obtained. The group sparse coefficient of the corresponding degraded image group is then applied to approximate the estimate, using the group sparse residual constraint as a priori condition. The nuclear norm and the zero norm of the tight frame wavelet transform coefficients are used to replace the low-rank and sparse constraint functions, respectively, to construct a dual prior regularizer. Finally, the alternating direction multiplier method is used for variable optimization solution.

[0048] Step 4: Simulation Experiment. Develop system software for photon counting spectral CT imaging based on Microsoft Visual Studio 2010 and Matlab. Conduct experimental verification: Using a digital pleural mouse model as an example, analyze the reconstructed image's material decomposition accuracy, image structure recovery, artifact removal, and noise suppression performance.

[0049] Step 5: Clinical trials. Validation was achieved using real-world data. Using data from mice scanned by a spectral CT system developed by the Institute of High Energy Physics of the Chinese Academy of Sciences, imaging quality was studied, performance indicators such as the material decomposition accuracy of the reconstructed images were analyzed, and an adaptive parameter selection strategy was designed. CT engineers and clinical radiologists were invited to evaluate the reconstructed images, adjust and optimize algorithm parameters, and research parallel acceleration algorithms to meet practical application requirements, ensuring rigorous clinical evaluation and optimization.

[0050] Specifically, the embodiment of the present invention further provides that in step 1, simulated projection data is generated by computer simulation, or actual spectral CT projection data is obtained through a partner to obtain photon counting spectral CT image data.

[0051] Specifically, the embodiment of the present invention further provides a specific process of step 2:

[0052] A group sparse residual model is constructed, the group sparse coefficient of the original image is estimated using non-local priors, and a more accurate group sparse residual constraint is obtained by weighted averaging of similar blocks.

[0053] Group sparse residual constraint model to improve group sparse coefficient The accuracy of the model can be written as:

[0054] (1)

[0055] in, , The original image corresponds to the group sparse coefficient of each group, Represents the sparse coefficient of group i The Kth vector in , Represents the projection data under the i-th energy channel, reflecting the original CT image information affected by low photon number and noise, and is the key input in the data fidelity term. Di represents the sparse representation dictionary corresponding to the degraded image Yi. -norm is used to measure the error between matrices Yi and DiAi, p represents a value of 1 or 2, p=1 corresponds to the L1 norm, p=2 corresponds to the L2 norm, is a regularization parameter used to balance the weights of the data fidelity term and the regularization term. , the trade-off between noise suppression and image detail preservation can be controlled, ensuring that a reconstructed image with clear structure can be obtained under low-dose and limited-angle data.

[0056] For each group of m non-locally self-similar blocks, we can Each The weighted average of A good estimate of . As shown in formula (2):

[0057] (2) in, represent and The kth and jth vectors of is the weight.

[0058] Specifically, set With the target block and similar blocks The distance between them is inversely proportional to:

[0059] (3)

[0060] Where L is the normalization factor and h is a predefined constant.

[0061] Specifically, the embodiment of the present invention further provides a specific process of step 3:

[0062] In order to effectively suppress the serious noise problem in the reconstructed image caused by the limited number of photons in the narrow energy channel, while retaining the low-rank structure and details of the spectral CT image, based on the work in the previous step, a low-rank sparse constraint is applied to each image sub-block, and the constructed regularizer is as follows:

[0063] (4)

[0064] in, is the kth image block of the i-th energy channel image, and are the sparse coefficients of the degradation group and the sparse coefficients of the original group corresponding to the k-th image sub-block, i*j matrices, is the regularization parameter.

[0065] The noise in CT images of different energy channels is processed in a targeted manner, and the sparse representation of the image in a certain transformation domain is used as a priori feature to constrain the reconstruction optimization problem. Norm or coefficients under tight frame wavelet transform The norm is used to characterize the sparsity of single-channel energy CT images, and the corresponding regularizer is as follows:

[0066] (5)

[0067] The low-rank sparsity of images with different energy channels is combined with the sparsity of images with a single energy channel to construct a dual prior regularizer as follows:

[0068] (6)

[0069] The dual prior regularizer is coupled with the spectral CT physical imaging model to obtain the following reconstruction optimization model:

[0070] (7)

[0071] Where M represents the number of spectral CT channels, is the system projection matrix related to the photon counting spectral CT scanning structure, is the CT image to be reconstructed under the i-th energy channel, is the projection data under the i-th energy channel.

[0072] Specifically, an embodiment of the present invention further provides analyzing the convexity and concavity of the optimization model in step 4, converting the original optimization problem into multiple sub-problems, and solving them using a suitable iterative method; solving the sub-problems using methods such as the alternating direction method of multipliers (ADMM) or the split Bregman algorithm; and solving the above-mentioned optimization model through numerical iteration to obtain a reconstructed target image.

[0073] Specifically, the embodiment of the present invention also provides the use of simulation experiments in step 5 to evaluate the performance of the reconstruction method, including structure recovery, artifact removal, noise suppression and material decomposition accuracy; the proposed reconstruction method is evaluated using clinical experiments, and CT engineers or clinical imaging physicians are invited to conduct evaluations, and the algorithm parameters are adjusted and optimized according to the evaluation results.

[0074] The implementation of each embodiment of the present invention is based on programmed processing performed by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of each embodiment of the present invention are packaged into various modules. Based on this reality, and in addition to the aforementioned embodiments, an embodiment of the present invention provides a feature-learning-based photon counting spectral CT image reconstruction system. This system is used to implement a feature-learning-based photon counting spectral CT image reconstruction method described in the aforementioned method embodiments.

[0075] The system includes: a data acquisition module for acquiring photon counting energy spectrum CT images; a regularizer construction module for constructing a dual prior regularizer based on the group sparse residual of the CT image and the sparse features of the image block; a CT image reconstruction model construction module for constructing a CT image reconstruction model based on the dual prior regularizer, including: obtaining the low-rank characteristics of CT images of different energy channels through group sparse residual and gradient image sparseness; performing sparsity constraints on the smoothness of the photon counting energy spectrum CT image space and the artifact characteristics to obtain the internal features of the single energy channel image; based on the low-rank characteristics of the CT images of different energy channels and the internal features of the single energy channel image, using the dual prior regularizer as a regularizer and combining it with the energy spectrum CT physical imaging model to obtain the CT image reconstruction model; a CT image reconstruction module for using the alternating direction multiplier method to perform numerical iterative solution on the CT image reconstruction model to obtain the reconstructed CT image.

[0076] The photon counting spectral CT image reconstruction system based on feature learning provided by the embodiment of the present invention solves the serious noise problem of the reconstruction results caused by insufficient number of photons in different energy channels of the photon counting spectral CT image reconstruction. Several modules are used to perform targeted processing on the noise in the CT images of different energy channels, thereby solving the problem of how to reconstruct high-quality spectral CT images from X-ray spectral CT projection data with low signal-to-noise ratio and limited scanning angle, reducing the radiation dose while improving the quality of clinical diagnosis.

[0077] Based on the same inventive concept as the above-mentioned embodiment, an embodiment of the present invention also provides an electronic device, including a memory and a processor, the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a photon counting energy spectrum CT image reconstruction method based on feature learning as proposed in the above-mentioned embodiment.

[0078] An embodiment of the present invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this program overcomes the artifacts and noise associated with photon-counting spectral CT imaging at restricted angles, reconstructing high-quality CT images that balance image quality and radiation protection, thereby enhancing diagnostic capabilities for subsequent evaluation applications such as material decomposition.

[0079] The storage medium can be any non-volatile storage device such as a hard disk, solid-state drive, flash drive, optical disk, etc., which is used to store computer program code and necessary data files. The stored computer program includes: a data acquisition module, a regular sub-construction module, a CT image reconstruction model construction module, and a CT image reconstruction module.

[0080] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and is susceptible to numerous variations. Any simple modifications, equivalent variations, and modifications to the above specific embodiments based on the technical essence of the present invention shall be deemed to fall within the scope of protection of the present invention.

Claims

1. A photon counting spectral CT image reconstruction method based on feature learning, characterized in that: include: S1, collecting photon counting spectral CT images; S2, constructing a dual prior regularizer based on the group sparse residual of CT images and the sparse features of image blocks; S3. Based on the dual prior regularizer, a CT image reconstruction model is constructed, including: obtaining low-rank characteristics of CT images of different energy channels through group sparse residuals and gradient image sparseness; Sparsity constraints are imposed on the smoothness and artifact characteristics of the photon counting spectral CT image space to obtain the internal features of the single energy channel image. Based on the low-rank characteristics of CT images of different energy channels and the internal features of the single energy channel image, a dual prior regularizer is used as a regularizer, and combined with the spectral CT physical imaging model, a CT image reconstruction model is obtained. S4. Using the alternating direction multiplier method, the CT image reconstruction model is numerically iteratively solved to obtain a reconstructed CT image.

2. The photon counting spectral CT image reconstruction method based on feature learning according to claim 1, characterized in that: Said S1 further includes: extracting projection data from the full-angle scanning data by computer simulation to generate CT images of different energy channels.

3. The photon counting spectral CT image reconstruction method based on feature learning according to claim 1, characterized in that: Said S2 comprises: Based on the non-local self-similarity prior of CT images and self-supervised learning, the group sparsity coefficient of each original image group is estimated; The group sparse coefficients of the corresponding degraded image group are applied for approximate estimation, and the group sparse residual constraint is used as a priori condition; The nuclear norm and the zero norm of tight frame wavelet transform coefficients are used to replace the low-rank and sparse constraint functions respectively to construct a dual prior regularizer.

4. The photon counting spectral CT image reconstruction method based on feature learning according to claim 1, characterized in that: Said S4 comprises: By analyzing the convexity and concavity of the CT image reconstruction model, the reconstruction problem is transformed into multiple sub-problems; Each subproblem is numerically iterated and solved using the alternating direction multiplier method to obtain the reconstructed CT image.

5. The photon counting spectral CT image reconstruction method based on feature learning according to claim 1, characterized in that: After obtaining the reconstructed CT image, the following steps are also included: Develop system software for photon counting spectral CT imaging based on a mathematical software platform and conduct experimental verification; The imaging quality is studied based on the data scanned by the energy spectrum CT system, and real data verification is performed.

6. A photon counting spectral CT image reconstruction system based on feature learning, characterized in that: include: A data acquisition module, used for acquiring photon counting spectral CT images; A regularizer construction module is used to construct a dual prior regularizer based on the group sparse residual of the CT image and the sparse features of the image block; A CT image reconstruction model construction module is used to construct a CT image reconstruction model based on a dual prior regularizer, including: obtaining low-rank characteristics of CT images of different energy channels through group sparse residuals and gradient image sparseness; Sparsity constraints are imposed on the smoothness and artifact characteristics of the photon counting spectral CT image space to obtain the internal features of the single energy channel image. Based on the low-rank characteristics of CT images of different energy channels and the internal features of the single energy channel image, a dual prior regularizer is used as a regularizer, and combined with the spectral CT physical imaging model, a CT image reconstruction model is obtained. The CT image reconstruction module is used to adopt the alternating direction multiplier method to perform numerical iterative solution on the CT image reconstruction model to obtain the reconstructed CT image.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the photon counting spectral CT image reconstruction method based on feature learning according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the photon counting spectral CT image reconstruction method based on feature learning according to any one of claims 1 to 5 are implemented.

Images