Method for alleviating metal artifact in energy spectrum CT image domain

Abstract

The present invention belongs to the technical field of CT imaging. Provided is a method for alleviating a metal artifact in an energy spectrum CT image domain. During the imaging of a target (such as a patient) having a metal implant, metal artifacts can still be observed in virtual single-energy images obtained by means of existing devices, and a metal artifact is especially more obvious in a virtual single-energy image at a low energy (a low keV). In the method, a non-artifact region (or a low-artifact region) in a virtual single-energy image having an artifact is extracted and a base material is decomposed (such as a water-based decomposed image or a bone-based decomposed image), so that a relationship model between the base material and an artifact-free (or low-artifact) image (such as a high-energy virtual single-energy image) is constructed; then, pixel (voxel) CT values, corresponding to the artifact region in the foregoing artifact-contained image, in the artifact-free (or low-artifact) single-energy image are substituted into the relationship model, so that new base material decomposition is obtained; and then a virtual single-energy image under any energy after artifact reduction is synthesized. The method can effectively suppress metal artifacts in virtual single-energy images of existing CT devices, the manufacturers and types of which are not limited.

Description

A method for improving metal artifacts in spectral CT image domain Technical Field

[0001] The present invention belongs to the technical field of CT imaging, and in particular relates to a method for improving metal artifacts in a virtual monoenergetic image in a spectral CT image domain. Background Art

[0002] Metal artifact is a major artifact in clinical CT (Computed Tomography) scans, degrading image quality. Causes of metal artifact include beam hardening, scattering, and photon starvation. Existing metal artifact removal methods can be broadly categorized into several categories. High-voltage scanning can be used. Algorithms include the classic Normalized Metal Artifact Reduction (NMAR) method, which relies on projection and interpolation, and can be optimized using iterations to achieve better images. With the widespread application of deep learning in recent years, deep learning methods have also found application in metal artifact removal. Furthermore, virtual monoenergetic images from high-energy dual-energy CT can effectively reduce metal artifacts.

[0003] Currently, all dual-energy CT images with a single energy spectrum, also known as virtual monoenergetic images, have good metal artifact suppression effects at high energy. However, obvious metal artifacts are still prevalent in virtual monoenergetic images at low energy.

[0004] To address this phenomenon, various manufacturers have developed metal artifact reduction (MAR) algorithms for their CT products, further removing metal artifacts from virtual monoenergetic images. However, the core of the MAR algorithm is polynomial interpolation. If the metal shape is regular and the artifacts are small, the MAR algorithm is very effective in correcting metal artifacts. Conversely, for images with irregular metal shapes and relatively complex artifact distribution, MAR correction is less effective and may even introduce new artifacts. This is the case with CT scans after pedicle screw placement and cone beam computed tomography (CBCT) or CT scans after denture placement.

[0005] Summary of the Invention

[0006] To address the defects of the prior art, the present invention provides a method for improving metal artifacts in the spectral CT image domain, which effectively suppresses and improves metal artifacts in virtual monoenergetic images of any manufacturer based on the image domain.

[0007] The present invention adopts the following technical solutions to solve the above problems:

[0008] A method for improving metal artifacts in a spectral CT image domain, the method comprising the following steps:

[0009] Acquire a plurality of virtual monoenergetic images of arbitrary energy, and acquire a first image and a second image from the virtual monoenergetic images; the first image is a monoenergetic image with significant artifacts, and the second image is a monoenergetic image without artifacts or with minimal artifacts;

[0010] Extracting an artifact region and a non-artifact region of the first image, and obtaining an artifact region mask and a non-artifact region mask accordingly;

[0011] extracting an artifact region of the second image and a non-artifact region of the second image based on the artifact region mask and the non-artifact region mask;

[0012] Select two virtual monoenergetic images with energy differences to perform dual-energy decomposition to obtain at least two base material maps;

[0013] Comparing the non-artifact area mask, a corresponding artifact-free base image composed of pixels or voxels in the non-artifact area of ​​each base material image is obtained;

[0014] Based on the non-artifact area of ​​the second image and the artifact-free base images, constructing a relationship model between the non-artifact area of ​​the second image and the artifact-free base images;

[0015] Substituting the artifact region of the second image into the relationship model, obtaining a component map of the artifact region on each base material map, i.e., a corrected base material map;

[0016] The corrected base material map is synthesized to obtain an energy spectrum image with artifact correction at any energy.

[0017] In one embodiment, the image types of the energy spectrum image include a virtual monochromatic image, a material density image, an effective atomic number image, an electron density image, a virtual plain scan image, and an iodine image.

[0018] In one embodiment, extracting the artifact region and the non-artifact region of the first image and correspondingly obtaining the artifact region mask and the non-artifact region mask includes: performing artifact correction on the first image using an artifact correction algorithm and then obtaining the artifact region mask and the non-artifact region mask;

[0019] Alternatively, after extracting the artifact area and non-artifact area of ​​the first image, an artifact area mask and a non-artifact area mask are obtained; wherein, the extraction method of the artifact area and the non-artifact area of ​​the first image includes a metal artifact removal algorithm based on interpolation and reconstruction; the extraction method also includes a MAR algorithm based on deep learning and direct use of deep learning.

[0020] In one embodiment, when extracting the artifact area and non-artifact area of ​​the first image, the image threshold segmentation method is used to preset the metal threshold, and the first image is traversed point by point. The pixel points in the first image with a CT value greater than the metal threshold are set to 1, and the pixel points less than the threshold are set to 0. The metal area and the non-metal area are segmented according to the assigned pixel values, and the artifact area and the non-artifact area are extracted based on the metal area and the non-metal area.

[0021] In one embodiment, a threshold segmentation method is used to obtain an artifact area mask and a non-artifact area mask.

[0022] In one embodiment, the second image is obtained by selecting from a virtual monoenergetic image, selecting from an adjacent image layer, and reconstructing after coarse correction of artifacts.

[0023] In one embodiment, the relationship model includes a relationship model between CT values ​​and water / bone base images, a relationship model between CT values ​​and any base material, and a relationship model between any different types of energy spectrum images.

[0024] In one embodiment, the energy states of the virtual monoenergetic image include 50 keV, 70 keV, 80 keV, 100 keV, 120 keV, and 140 keV.

[0025] In one embodiment, the method can be adapted to use images containing metal or not containing metal in an image sequence and build a relationship model to achieve correction.

[0026] In one embodiment, the method is applicable to artifact improvement of two-dimensional images and three-dimensional images.

[0027] In one embodiment, the method can be used to construct a corresponding relationship model for each set of energy images, and to pre-construct a universal relationship model applicable to monoenergetic images and basis material components.

[0028] In one embodiment, the method for constructing the relationship model includes polynomial fitting, deep learning, and pattern recognition.

[0029] In one embodiment, the types of artifacts that the relational model in the method can be used to optimize include metal artifacts, bone artifacts, and water hardening artifacts.

[0030] In one embodiment, the relational model in the method can be applied to non-spectral images.

[0031] The beneficial effects of the present invention are:

[0032] 1. This paper has developed an algorithm that can effectively suppress and improve metal artifacts in virtual monoenergetic images from any manufacturer based on the image domain. This algorithm can achieve the same metal artifact suppression effect in low-energy monoenergetic images as in high-energy images.

[0033] 2. This algorithm does not introduce new artifacts. It is particularly effective in cases with complex artifacts and metal shapes, making up for the shortcomings of traditional MAR algorithms (but not limited to MAR). BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the specific embodiments of the present invention, the following will briefly introduce the drawings required for use in the description of the specific embodiments. Obviously, the drawings described below are 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.

[0035] FIG1 is a flow chart of an algorithm of the present invention;

[0036] Figure 2 is an example of the complete artifact image extraction process;

[0037] Figure 3 is a schematic diagram of the model construction steps;

[0038] Figure 4 is a virtual monoenergetic image of the phantom test results;

[0039] FIG5 is a virtual monoenergetic image of the test results of pedicle screw placement patient data;

[0040] FIG6 is a virtual monoenergetic image of the test results of oral denture implant patient data. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] It should be noted that the method for extracting artifacts in the examples of the present invention is MAR, but is not limited to MAR; in addition, the case of the present invention shows the establishment of a model between the base material and the CT value. Based on the existing results, the relationship model between the base materials can also be directly established, and the two are equivalent.

[0043] It should be noted that the method of using two single-energy images of the non-artifact area in the example of the present invention to create an artifact-free water / bone base map and establish a model can be extended to use other images in the image sequence (either images containing metal or not) and establish a model, and the two are equivalent.

[0044] It should be noted that the relationship model between CT values ​​and water / bone matrix images in the present invention can be extended to the relationship model between CT values ​​and any matrix pair images, as well as the relationship model between any other different types of energy spectrum images.

[0045] It should be noted that this method is applicable to artifact improvement of two-dimensional images and can also be directly extended to artifact improvement of three-dimensional images.

[0046] It should be noted that in addition to constructing a correction model for the current data for each set of input data, this method can also pre-construct a universal model relationship between monoenergetic images and basis material components. In this case, one correction model can be applied to all input data (including but not limited to constructing a universal model for different scanning sites, the presence or absence of contrast agents, and other different situations).

[0047] It should be noted that the model building or optimization methods in this method include but are not limited to polynomial fitting, deep learning (neural network), and pattern recognition.

[0048] It should be noted that the model in this method can be used to optimize artifact types, including but not limited to metal artifacts, bone artifacts, water hardening artifacts, etc.

[0049] It should be noted that the model used in this method is not limited to spectral images; similar modeling can also be performed on non-spectral images. Non-spectral images are images obtained by CT scans at a single energy level. These images are typically generated based on the attenuation of radiation to produce tissue density images, such as images showing water sclerosis or bone sclerosis. Spectral images are images obtained by scanning with X-rays at multiple energy levels. These images are generated based on the differences in tissue attenuation characteristics at different X-ray energies.

[0050] For ease of understanding, the specific meanings of the proper nouns used in this application are shown in the proper noun comparison table:

[0051] Typically, dual-energy CT can generate n virtual monoenergetic images at any energy between 40 and 140 keV. Here, it is assumed that the product outputs six monoenergetic images at 50 keV, 70 keV, 80 keV, 100 keV, 120 keV, and 140 keV. As shown in Figure 1, the present invention proposes a method for improving metal artifacts in the spectral CT image domain, comprising the following steps:

[0052] S1: Input n virtual monoenergetic images of arbitrary energy, select the monoenergetic image with significant artifacts (denoted as Image_Worst), i.e., the first image; identify the artifact areas and non-artifact areas in the above artifact image;

[0053] In this embodiment, as shown in Figure 3(a), Image_Worst is a 50keV monoenergetic image. Any artifact correction algorithm can be used to perform artifact correction on the image to obtain a mask of the artifact area, or the artifact area or non-artifact area of ​​the image can be directly identified. The method includes but is not limited to traditional metal artifact removal algorithms based on interpolation and reconstruction, such as MAR, fsMAR (frequency split metal artifact reduction), iMAR (iterative Metal Artifact Reduction), OMAR (Orthopedic Metal Artifact Reduction), sMAR (smart Metal Artifact Reduction), etc., and is not limited to MAR algorithms based on deep learning or directly using deep learning methods to identify artifact areas or non-artifact areas.

[0054] As shown in Figure 2, this is a method for extracting artifact areas: ① Image threshold segmentation: Set a metal threshold, such as HU (Hounsfield Unit) is 3000, and the HU value is also called the CT value. Traverse the image point by point, and set the pixel points with an image CT value greater than 3000 to 1, and the pixel points with a CT value less than 3000 to 0, thereby segmenting the metal area and the non-metal area; ② Mean smoothing of the metal image: The operator used in this case is a 7×7 mean filter operator, which is multiplied by the corresponding points of the image in a sliding window manner to obtain a smoothed, slightly enlarged metal area image; ③ The input image is forward projected using a projection algorithm, and the metal image is forward projected at the same time. In the projection data, the position of the metal projection data can be used to determine which positions in the projection of the original image are metal areas, and then polynomial interpolation is performed on these areas to obtain the corrected projection data; ④ The corrected projection data is processed Reconstruction will obtain a preliminary corrected image; ⑤ A 7×7 mean filter operator is used again to smooth the image; ⑥ Threshold segmentation: In this embodiment, the threshold range of soft tissue is set to 800-1200, the soft tissue area is extracted, and the average CT value of the soft tissue is calculated; ⑦ The soft tissue area excluding the metal area is filled to obtain the mean image; ⑧ The mean image is forward projected using the projection algorithm, and the difference between the projection data of the input image and the projection data of the mean image is recorded as Pjr_diff. This difference is again subjected to polynomial interpolation and recorded as Pjr_inter; ⑨ The result of Pjr_inter minus Pjr_diff is reconstructed using the reconstruction algorithm to obtain the final artifact area.

[0055] S2: Obtain the mask of the artifact area (denoted as ImageMask1) and the mask of the non-artifact area (denoted as ImageMask0).

[0056] In this embodiment, the algorithm selected is the threshold segmentation method. The artifact image is segmented and calculated using a threshold. The CT values ​​at corresponding locations are summed and reassigned to obtain an artifact region mask. Threshold segmentation methods include direct threshold segmentation and threshold segmentation methods with a specific function distribution. These threshold segmentation methods include histogram techniques, entropy algorithms, and adaptive threshold algorithms.

[0057] In this example, a threshold of -800 is selected to extract the air region (the air CT value is -1000). The image with significant artifacts (Figure 3(a)) is segmented, and points are traversed point by point. Values ​​greater than -800 are set to 1, and values ​​less than -800 are set to 0, resulting in Figure 3(d) . Using the same method, a threshold of 100 is set to extract the main artifact region. The artifact image (Figure 3(c)) is threshold segmented to obtain Figure 3(e). Finally, the CT values ​​of the corresponding positions in Figures 3(e) and 3(d) are added together, and points with a CT value of 2 are reset to 1. This yields the final artifact region mask, denoted as ImageMask1 (Figure 3(f)).

[0058] S3: Select an image without artifacts or with small artifacts (denoted as Image_Best), that is, the second image. According to the two masks obtained in S2, two new images are derived from Image_Best, one of which is an image consisting only of pixels or voxels of Image_Best in the area selected by the ImageMask1 mask (denoted as Image_Best_Artifact), that is, the artifact area of ​​the second image, and the other is an image consisting only of pixels or voxels of Image_Best in the area selected by the ImageMask0 mask (denoted as Image_Best_noArtifact), that is, the non-artifact area of ​​the second image.

[0059] It should be noted that the image with no artifacts or with small artifacts (Image_Best) can be selected from n virtual monoenergetic images of any energy in S1, or images reconstructed by other methods, including but not limited to other sources (such as different image layers, data after coarse correction of artifacts, etc.) or images reconstructed by other methods. Different image layers here refer to adjacent image layers. If the image being corrected is a two-dimensional image, and the actual human body data must be three-dimensional, image layers without metal artifacts or image layers with small artifacts can be found in the three-dimensional image. At this time, these image layers without artifacts or image layers with small artifacts can be used for modeling to correct the current artifact image layer. Coarse correction of artifacts can be done using conventional de-artifacting methods, such as MAR, interpolation-based projection data repair, iterative reconstruction correction, etc.

[0060] In this embodiment, Image_Best is selected from n virtual monoenergetic images of arbitrary energy in S1. As shown in FIG3(b), an algorithm is used to automatically identify images without artifacts or with small artifacts among the n input monoenergetic images, such as the TV (Total Variation) algorithm. In this case, the cost function of the theoretical TV function is used. The TV function is shown in Formula (1):

[0061] Among them, i, j represent the pixel index of the image, f i,j Represents the image pixel value at position (i, j), u i,j The TV value represents a pixel value. By traversing each pixel point of the image, the TV value of the current point can be calculated. Finally, the TV value of all points is summed to obtain the final TV value of the image.

[0062] The cost function values ​​of n images are calculated in sequence, and the image with the smallest cost function value is considered to be an image without artifacts or with small artifacts.

[0063] The identified image with no artifacts or small artifacts is multiplied by the corresponding position of the mask ImageMask0 of the non-artifact area to obtain the non-artifact area of ​​the current image (denoted as Image_Best_noArtifact). The image with no artifacts or small artifacts is multiplied by the mask ImageMask1 of the artifact area to obtain the artifact area of ​​the current image (denoted as Image_Best_Artifact).

[0064] S4: Select two virtual monoenergetic images with energy differences from S1 to perform dual-energy decomposition to obtain two or more base material images.

[0065] S5: Comparing with ImageMask0, obtaining an image composed of pixels or voxels in the non-artifact area of ​​two or more base material images.

[0066] It should be noted that there can be n types of decomposition base materials, but in order to explain the problem concisely, the embodiment shown in the present invention uses two decomposition base materials. The first base material is water-based and the second base material is bone-based. The density images of the water-based material and the bone-based material are multiplied by the mask of the non-artifact area to obtain a water-based image and a bone-based image without artifacts.

[0067] The artifact-free or artifact-less image (Image_Best) and the artifact-significant image (Image_Worst) identified in S3 are used for dual-energy decomposition to obtain the water-based image (denoted as imgWater) and the bone-based image (denoted as imgBone). The dual-energy decomposition process is as follows:

[0068] Among them, E1 and E2 represent two kinds of energy. and Corresponding to Image_Best and Image_Worst, ρ w represents the density of water, m w (E) represents the mass attenuation coefficient of water material under energy E, m b (E) represents the mass attenuation coefficient of the bone material at energy E. The mass attenuation coefficients of water and bone at different energies are both known and can be found on public websites. This equation can be used to solve the water-based and bone-based diagrams.

[0069] Finally, the water-based image and the bone-based image are multiplied by the mask ImageMask0 of the non-artifact area respectively, and the water-based image and the bone-based image without artifacts (denoted as imgWater_noArtifact and imgBone_noArtifact) will be obtained.

[0070] S6: Based on the Image_Best_noArtifact image in S3 and the image composed of pixels or voxels in the non-artifact area obtained in S5, a relationship model between the Image_Best_noArtifact image and the two or more base material maps is constructed.

[0071] Based on the CT value of the artifact-free image (Image_Best_noArtifact) and the artifact-free water-based and bone-based maps (imgWater_noArtifact and imgBone_noArtifact), we sequentially traverse every pixel of Image_Best_noArtifact, imgWater_noArtifact, and imgBone_noArtifact. The three images are all the same size. The pixel values ​​at the same position are taken to form two data pairs: CT value-water-based map value and CT value-bone-based map value. This generates multiple scattered points, corresponding to the scattered point distributions in Figures 3(g) and (h). Finally, a polynomial fit is performed on these scattered points, resulting in the fitting curves in Figures 3(g) and (h). This constructs a relationship model between CT value and water / bone-based maps.

[0072] It should be pointed out that the algorithm builds a model of the current data to be processed for a certain piece of data, so the model form is not unique.

[0073] S7: Substitute the Image_Best_Artifact image obtained in S3 into the relationship model established in S6 to obtain a component map of the artifact area on two or more base materials, that is, a corrected base material map.

[0074] Using the artifact region image obtained in S3, also known as Image_Best_Artifact, we traverse every pixel in the image and obtain the CT value. This is then fed into S6 to establish the curve fitting formulas in Figures 3(g) and 3(h). This yields the corresponding water-based and bone-based image values ​​for each CT value (denoted as imgWater_ArtifactCorrect and imgBone_ArtifactCorrect). Ultimately, imgWater_ArtifactCorrect plus imgWater_noArtifact is denoted as imgWater_Correct, and imgBone_ArtifactCorrect plus imgBone_noArtifact is denoted as imgBone_Correct.

[0075] S8: Using the base material images (imgWater_Correct and imgBone_Correct) obtained in S7, an energy spectrum image with artifact correction at any energy can be synthesized.

[0076] It should be noted that energy spectrum images include but are not limited to virtual monochrome images, material density images, effective atomic number images, electron density images, virtual plain scan images, iodine images and other energy spectrum images. Artifact manifestations can also improve the accuracy of non-image energy spectrum decomposition results such as energy spectrum curves and scatter plots.

[0077] In this embodiment, based on imgWater_Correct and imgBone_Correct, dual-energy decomposition is performed again to obtain a virtual monoenergetic image with artifact correction at any energy. The dual-energy decomposition formula is as follows:

[0078] The above algorithm verification was carried out, and the verification results are shown in Figures 4-6, where the red arrows represent artifacts existing in the image itself, the yellow arrows represent artifacts introduced by the correction algorithm (there are two types of correction algorithms, namely the traditional MAR algorithm and the algorithm proposed in this invention) relative to the original image, and the yellow box represents the changes in the image structure caused by the correction algorithm relative to the original image.

[0079] Verification Example 1: A self-developed CT performance phantom with a diameter of 200 mm and two titanium cylindrical phantoms with a diameter of 10 mm embedded inside was placed in a dual-energy CT scan from an internationally renowned manufacturer.

[0080] As shown in Figure 4, the first row of results shows virtual monoenergetic images at different energies provided by the manufacturer's software without MAR correction. It can be seen that at low energies, such as monoenergetic images below 70 keV, there is a significant artifact between the two metals, while the artifact correction effect of high-energy monoenergetic images is better. The second row of results shows the results of the manufacturer's software performing MAR correction. It can be seen that the overall image artifacts are suppressed to a certain extent, but many new artifacts are introduced. The correction effect of the 50 keV image is also not ideal. The third row shows the correction results of the algorithm described in this invention based on the first row of images. The overall correction effect is better than that of the manufacturer's MAR algorithm, and no new artifacts are introduced. The algorithm's artifact suppression effect is significant, demonstrating the effectiveness of the algorithm. The fourth row shows an enlarged image of the 70 keV image from the first three rows of results. It can be clearly seen that the proposed algorithm effectively improves the artifacts in the area indicated by the red arrow without introducing new artifacts.

[0081] Verification Example 2: A postoperative CT scan of a patient undergoing pedicle screw placement was performed using a dual-energy CT scan produced by an internationally renowned manufacturer.

[0082] As shown in Figure 5, similar to Figure 4, the first row is the manufacturer's original monoenergetic image, the second row is the manufacturer's monoenergetic image after MAR, the third row is the correction result of the current algorithm based on the first row of images, and the fourth row is the enlarged image of the 70keV image in the first three rows of results; the product MAR correction result obviously introduces dark band artifacts (as shown by the yellow arrow), the metal area is blurred, and it causes changes in tissue structure (as shown by the yellow box), while the current algorithm correction effectively improves the artifacts of images at all energies, and the metal boundary is clearer.

[0083] Verification Example 3: A postoperative CT scan of a patient undergoing oral denture implantation was performed using a dual-energy CT scan from an internationally renowned manufacturer.

[0084] As shown in Figure 6, similarly, the first row is the manufacturer's original monoenergetic image, the second row is the manufacturer's monoenergetic image after MAR, the third row is the correction result of the current algorithm based on the first row of images, and the fourth row is the enlarged image of the 70keV image in the first three rows of results; the product MAR correction result, although it improves the artifacts in the original image, it also introduces many new artifacts. The current correction algorithm is significantly better than the product's MAR result, does not introduce new artifacts, and also effectively improves the artifacts in the low keV image.

[0085] The present invention has been described in detail above through the embodiments. However, the contents described are only preferred embodiments of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for improving metal artifacts in spectral CT image domain, characterized by: The method comprises the following steps: Acquire a plurality of virtual monoenergetic images of arbitrary energy, and acquire a first image and a second image from the virtual monoenergetic images; the first image is a monoenergetic image with significant artifacts, and the second image is a monoenergetic image without artifacts or with relatively small artifacts; Extracting an artifact area and a non-artifact area of ​​the first image, and obtaining an artifact area mask and a non-artifact area mask accordingly; Extracting an artifact region of the second image and a non-artifact region of the second image based on the artifact region mask and the non-artifact region mask; Selecting two virtual monoenergetic images with energy differences to perform dual-energy decomposition to obtain at least two base material images; By comparing the non-artifact area mask, a corresponding artifact-free base image composed of pixels or voxels in the non-artifact area of ​​each base material image is obtained; Based on the non-artifact area of ​​the second image and the artifact-free base images, constructing a relationship model between the non-artifact area of ​​the second image and the artifact-free base images; Substituting the artifact region of the second image into the relationship model, obtaining a component map of the artifact region on each base material map, that is, a corrected base material map; The correction base material map is synthesized to obtain an energy spectrum image of artifact correction at any energy.

2. The method for improving metal artifacts in spectral CT image domain according to claim 1, characterized in that: The image types of the energy spectrum image include virtual monochrome image, material density image, effective atomic number image, electron density image, virtual plain scan image and iodine image.

3. The method for improving metal artifacts in spectral CT image domain according to claim 1, Features: The extracting the artifact area and the non-artifact area of ​​the first image and correspondingly obtaining the artifact area mask and the non-artifact area mask comprises: performing artifact correction on the first image using an artifact correction algorithm to obtain the artifact area mask and the non-artifact area mask; Alternatively, the artifact area mask and the non-artifact area mask are obtained after extracting the artifact area and the non-artifact area of ​​the first image; wherein the extraction method of the artifact area and the non-artifact area of ​​the first image includes a metal artifact removal algorithm based on interpolation and reconstruction; the extraction method also includes a MAR algorithm based on deep learning and directly using deep learning.

4. The method for improving metal artifacts in spectral CT image domain according to claim 3, characterized in that: When extracting the artifact area and the non-artifact area of ​​the first image, the image threshold segmentation method is used to preset the metal threshold, and the first image is traversed point by point. The pixel points in the first image whose CT values ​​are greater than the metal threshold are set to 1, and the pixel points whose CT values ​​are less than the threshold are set to 0. The metal area and the non-metal area are segmented according to the assigned values ​​of the pixel points, and the artifact area and the non-artifact area are extracted based on the metal area and the non-metal area.

5. The method for improving metal artifacts in spectral CT image domain according to claim 1, characterized in that: The artifact area mask and the non-artifact area mask are obtained by using a threshold segmentation method.

6. The method for improving metal artifacts in spectral CT image domain according to claim 1, characterized in that: The second image is obtained by selecting from the virtual monoenergetic image, selecting from an adjacent image layer, and reconstructing after coarse correction of artifacts.

7. The method for improving metal artifacts in spectral CT image domain according to claim 1, characterized in that: The relationship model includes a relationship model between CT value and water / bone base image, a relationship model between CT value and any base material, and a relationship model between any different types of energy spectrum images.

8. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The energy states of the virtual monoenergetic image include 50keV, 70keV, 80keV, 100keV, 120keV and 140keV.

9. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The method can be adapted to use images containing metal or not containing metal in an image sequence and establish the relationship model to achieve correction.

10. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The method is applicable to artifact improvement of two-dimensional images and three-dimensional images.

11. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The method can be used to construct the corresponding relationship model for each set of energy images, and to pre-construct a universal relationship model applicable to monoenergetic images and basis material components.

12. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The method for constructing the relationship model in the method includes at least one of polynomial fitting, deep learning, and pattern recognition.

13. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The artifact types that the relationship model in the method can be used to optimize include at least one of metal artifacts, bone artifacts and water hardening artifacts.

14. The method for improving metal artifacts in spectral CT image domain according to any one of claims 1 to 7, characterized in that: The relationship model in the method can be applicable to non-spectral images.

Images