Method, device, computer equipment and storage medium for removing artifacts from scanned images

Abstract

The present invention relates to the field of medical image processing technology, and discloses a method, apparatus, computer device, and storage medium for removing artifacts from scanned images. The method comprises: obtaining a scanned image to be corrected that has ring artifacts; converting the ring artifacts in the scanned image to be corrected into strip artifacts to obtain a first image with strip artifacts; using a preset unsupervised network model to eliminate the strip artifacts in the first image to generate a second image; and performing artifact removal based on the first and second images to obtain a target scanned image. By implementing the technical solution of the present invention, the unsupervised network model in deep learning is fully utilized to solve the serious problem of ring artifacts in scanned images, thereby achieving effective removal of ring artifacts.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method, device, computer equipment and storage medium for removing artifacts from scanned images. Background Art

[0002] Computed tomography (CT) images often exhibit numerous ring artifacts, which significantly reduce the quality of reconstructed images and may lead to misdiagnosis later on. Therefore, removing ring artifacts from CT images is crucial.

[0003] Currently, methods for removing ring artifacts from CT images fall into three main categories: hardware-based methods, projection-domain sinusoidal correction methods, and image-domain post-processing methods using polar coordinate transformation. However, hardware-based methods rely on specialized hardware design, which is complex and difficult to guarantee artifact removal effectiveness; projection-domain sinusoidal correction methods rely on the original projection data and can affect image detail, resulting in poor reconstructed image quality; and image-domain post-processing methods using polar coordinate transformation have a certain error rate and are difficult to effectively remove artifacts from images. Consequently, none of these methods can effectively remove ring artifacts from CT images. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method, apparatus, computer device, and storage medium for removing artifacts from scanned images to solve the problem that artifacts in CT images are difficult to remove effectively.

[0005] In a first aspect, an embodiment of the present invention provides a method for removing artifacts from a scanned image, the method comprising: obtaining a scanned image to be corrected with ring artifacts; converting the ring artifacts in the scanned image to be corrected into strip artifacts to obtain a first image with strip artifacts; using a preset unsupervised network model to eliminate the strip artifacts in the first image to generate a second image; and performing artifact removal based on the first image and the second image to obtain a target scanned image.

[0006] The method for removing artifacts from scanned images provided by an embodiment of the present invention converts the ring artifacts in the scanned image to be corrected into strip artifacts, and uses a preset unsupervised network model to eliminate the strip artifacts. This enables high-quality images to be generated through comparative learning of the preset unsupervised network model, and effectively eliminates the ring artifacts in the scanned image. The preset unsupervised network model in this method does not require paired data, which reduces the difficulty of collecting a large amount of annotated training data and improves the training efficiency of the preset unsupervised network model. Moreover, this method does not require additional prior information or hardware modifications, and fully utilizes the unsupervised network model in deep learning to solve the serious problem of ring artifacts in scanned images, thereby achieving effective removal of ring artifacts.

[0007] In combination with the first aspect, in one embodiment, the ring artifacts in the scanned image to be corrected are converted into strip artifacts to obtain a first image with strip artifacts, including: obtaining the first position of each pixel point in the scanned image to be corrected in a Cartesian coordinate system; converting the first position corresponding to each pixel point to a polar coordinate system to obtain the second position of each pixel point in the polar coordinate system; based on the second position, obtaining the first image with strip artifacts.

[0008] The method for removing artifacts from scanned images provided by an embodiment of the present invention is such that, since strip artifacts are easier to remove than ring artifacts, the ring artifacts in the scanned image to be corrected are converted into strip artifacts through coordinate system transformation, thereby facilitating better removal of artifacts in the scanned image and improving the artifact removal effect.

[0009] In combination with the first aspect or its corresponding embodiment, in one embodiment, the method further includes: detecting and acquiring multiple unconverted pixel points generated during the coordinate transformation process; and generating multiple unconverted pixel points using a bilinear interpolation algorithm.

[0010] The method for removing artifacts from scanned images provided by an embodiment of the present invention may cause multiple pixel points in the Cartesian coordinate system to fail to be successfully converted due to the different coordinate parameters between the polar coordinate system and the Cartesian coordinate system. At this time, a bilinear interpolation algorithm is used to generate multiple unconverted pixel points to avoid sudden or jagged artifacts or image distortion in the polar coordinate system, thereby ensuring the generation of a smooth and continuous image in the polar coordinate system, so that subsequent image processing and analysis can be more accurate and reliable.

[0011] In combination with the first aspect, in one embodiment, artifact removal is performed based on the first image and the second image to obtain a target scanned image, including: performing difference processing based on the first image and the second image to obtain a strip artifact image; performing inverse transformation processing on the strip artifact image to generate a ring artifact image; performing difference processing based on the scanned image to be corrected and the ring artifact image to obtain the target scanned image.

[0012] In the artifact removal method for scanned images provided by the present invention, the second image is an image with stripe artifacts removed. The stripe artifact image is obtained by subtracting the first image with stripe artifacts from the second image. This image is then converted into a ring artifact image through coordinate transformation. The target scanned image with ring artifacts removed is then subtracted from the scanned image to be corrected. This method maximizes the preservation of image detail information, thereby ensuring the quality of the subsequently reconstructed image.

[0013] In combination with the first aspect or its corresponding embodiment, in one embodiment, a striped artifact image is obtained based on a difference processing between the first image and the second image, including: subtracting the first image from the second image to generate a residual image; and filtering the residual image to obtain a striped artifact image.

[0014] The artifact removal method for a scanned image provided by an embodiment of the present invention performs difference processing on a first image with strip artifacts and a second image to obtain a residual image with image details and strip artifacts. The residual image is filtered to remove the image details it carries to obtain a strip artifact image, thereby facilitating the acquisition of a ring artifact that does not carry image details. Therefore, when the ring artifacts are subsequently eliminated, the image detail information can be maintained, while the anatomical integrity and spatial resolution of the original scanned image can be maintained.

[0015] In combination with the first aspect or its corresponding embodiment, in one embodiment, the strip artifact image is inversely transformed to generate a ring artifact image, including: obtaining the third position of each pixel point in the strip artifact image in the polar coordinate system; converting the third position corresponding to each pixel point to the Cartesian coordinate system to obtain the fourth position of each pixel point in the Cartesian coordinate system; and generating a ring artifact image based on each pixel point and its corresponding fourth position.

[0016] The method for removing artifacts from scanned images provided by an embodiment of the present invention converts a strip artifact image into a ring artifact image through coordinate system transformation, which facilitates subtraction processing between the strip artifact image and the scanned image to be corrected, thereby effectively removing the ring artifacts in the scanned image, thereby achieving effective removal of the ring artifacts.

[0017] In combination with the first aspect, in one embodiment, the method further includes: obtaining a first voxel value of the scanned image to be corrected and a second voxel value of the target scanned image; and performing voxel value correction on the target scanned image based on the difference between the first voxel value and the second voxel value.

[0018] The method for removing artifacts from scanned images provided by an embodiment of the present invention corrects the voxel values of the target scanned image so that the voxel distribution of the corrected target scanned image is consistent with that of the original scanned image to be corrected, thereby making the removal effect of the ring artifact more obvious.

[0019] In combination with the first aspect, in one embodiment, the preset unsupervised network model includes a generator and a discriminator, and the training method of the preset unsupervised network model includes: inputting a slice of the first image into the generator to obtain a composite image; inputting the composite image into the discriminator to obtain a recognition result for the composite image; based on the composite image and the recognition result, using a backpropagation algorithm to iterate the generator and the discriminator until the discriminator determines that the composite image output by the generator is an image without strip artifacts.

[0020] The method for removing artifacts from scanned images provided by an embodiment of the present invention continuously iterates the generator and the discriminator, so that the synthetic image output by the generator continuously approaches the image without strip artifacts. At the same time, the identification ability of the discriminator is continuously enhanced, thereby improving the quality and credibility of the synthetic image.

[0021] In combination with the first aspect or its corresponding embodiment, in one embodiment, the generator includes multiple downsampling units, multiple residual units and multiple upsampling units, and a slice of the first image is input into the generator to obtain a composite image, including: performing feature extraction on the slice of the first image to obtain a feature map of the first dimension; inputting the feature map into multiple downsampling units to obtain a downsampling feature map of the second dimension; inputting the downsampling feature map into multiple residual units to output a residual feature map; inputting the residual feature map into multiple upsampling units to obtain a composite image, and the dimension of the composite image is equal to the dimension of the first image.

[0022] The artifact removal method for scanned images provided in an embodiment of the present invention combines multiple downsampling units, multiple residual units, and multiple upsampling units to obtain a synthetic image, thereby eliminating the problems of difficulty and degradation in network model training and ensuring the training effect of the unsupervised network model.

[0023] In combination with the first aspect or its corresponding embodiment, in one embodiment, the discriminator includes multiple downsampling units, and the composite image is input into the discriminator to obtain a recognition result for the composite image, including: cropping the composite image into multiple image blocks of preset dimensions; inputting each image block into multiple downsampling units in turn, and outputting the discrimination result corresponding to each image block through the last downsampling unit; and performing mean processing on each discrimination result to obtain a recognition result.

[0024] The method for removing artifacts from a scanned image provided by an embodiment of the present invention crops a synthesized image into multiple image blocks and uses multiple downsampling units to identify each image block to obtain a recognition result corresponding to each image block, thereby reducing the architectural parameters of the discriminator and making the training of the discriminator more convenient.

[0025] In combination with the first aspect or its corresponding embodiment, in one embodiment, the method further includes: determining the loss function of the preset unsupervised network model based on the synthetic image output by the preset unsupervised network model; and optimizing the preset unsupervised network model based on the loss function.

[0026] The method for removing artifacts from scanned images provided in an embodiment of the present invention optimizes a preset unsupervised network model through a determined loss function, thereby promoting the image generated by the preset unsupervised network model to be more visually similar to the actual scanned image, making the distribution patterns of the two as consistent as possible, and maximizing the artifact removal effect of the scanned image.

[0027] In second aspect, an embodiment of the present invention provides an artifact removal device for a scanned image, the device comprising: an image acquisition module for acquiring a scanned image to be corrected with ring artifacts; an image conversion module for converting the ring artifacts in the scanned image to be corrected into strip artifacts to obtain a first image with strip artifacts; an artifact elimination module for eliminating the strip artifacts in the first image based on a preset unsupervised network model to generate a second image; and a correction module for performing artifact removal based on the first image and the second image to obtain a target scanned image.

[0028] In a third aspect, an embodiment of the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, computer instructions being stored in the memory, and the processor executing the method for removing artifacts from a scanned image according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0029] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for removing artifacts from a scanned image according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. 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.

[0031] Figure 1 is a flowchart of a method for removing artifacts from a scanned image according to some embodiments of the present invention;

[0032] Figure 2 is a flowchart of another method for removing artifacts from a scanned image according to some embodiments of the present invention;

[0033] Figure 3 is a flowchart of a method for training a preset unsupervised network model according to some embodiments of the present invention;

[0034] Figure 4is a flow chart of an artifact removal method according to some embodiments of the present invention;

[0035] Figure 5 is a histogram comparing standard deviations of selected ROIs in CT images before and after correction in some embodiments of the present invention;

[0036] Figure 6 is a histogram comparing standard deviations of selected ROIs from simulated CT data before and after correction in some embodiments of the present invention;

[0037] Figure 7 is a schematic diagram of artifact removal results on a real CT image according to some embodiments of the present invention;

[0038] Figure 8 is a graph showing changes in CT values at two lines in a CT image according to some embodiments of the present invention;

[0039] Figure 9 is a schematic diagram of uniform axial CBCT slices according to some embodiments of the present invention;

[0040] Figure 10 is a structural block diagram of a device for removing artifacts from a scanned image according to an embodiment of the present invention;

[0041] Figure 11 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0043] In recent years, many researchers have proposed many methods to eliminate ring artifacts in CT images. These methods can be roughly divided into three categories: 1) hardware-based methods; 2) correction of sinusoids in the projection domain; and 3) post-processing methods using polar coordinate transformation in the image domain.

[0044] Hardware-based methods require special hardware design, which means that the detector array needs to be moved to obtain the different responses of each detector element during data acquisition. The characteristics of all detector elements are then averaged to reduce artifacts, but they cannot effectively eliminate artifacts completely and the design is complex.

[0045] Correction of the projection domain sinusoidal pattern is a preprocessing method for image reconstruction. Related art preprocessing methods rely on the original projection data, which limits their further application. Furthermore, low-pass filtering in these preprocessing methods removes image details, reducing the quality of the reconstructed image.

[0046] In the related art, the methods for removing ring artifacts based on polar coordinate transformation post-processing methods mainly include: independent component analysis-based methods; radial basis function neural network (RBFNN)-based methods; and hybrid ring artifact removal algorithms based on convolutional neural network (CNN). However, the independent component analysis-based methods are prone to losing image detail information; the RBFNN-based methods require manual identification of enhanced streak artifacts. For artifacts without obvious features, such as messy strips or gaps in the artifacts, manual identification methods may have a certain error rate; and the CNN-based hybrid ring artifact removal algorithm is difficult to remove all artifacts, and minor artifacts are usually retained.

[0047] Based on this, the technical solution of the present invention addresses the shortcomings of the related art and proposes a method for removing artifacts from scanned images based on unsupervised learning. Specifically, this method converts ring artifacts in the scanned image to be corrected into stripe artifacts, which are then removed using a pre-defined unsupervised network model. This method, through contrastive learning using the pre-defined unsupervised network model, generates high-quality images and effectively removes ring artifacts from scanned images. The pre-defined unsupervised network model in this method does not require paired data, reducing the difficulty of collecting large amounts of annotated training data and improving the training efficiency of the pre-defined unsupervised network model. Furthermore, without requiring additional prior information or hardware modifications, the method fully utilizes the unsupervised network model in deep learning to address the severe problem of ring artifacts in scanned images and effectively remove them. Compared to pre-processing methods before image reconstruction, this method preserves image detail while maintaining the anatomical integrity, detail, and spatial resolution of the original scanned image. Furthermore, the method was tested and evaluated using simulated CT data with ring artifacts as an internal test dataset and using patient brain CT data as an external test dataset, enhancing its robustness. Furthermore, this method provides technical support for realizing adaptive radiotherapy and improving the treatment effect of cancer patients.

[0048] According to an embodiment of the present invention, an embodiment of a method for removing artifacts from a scanned image is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0049] In this embodiment, a method for removing artifacts from a scanned image is provided, which can be used in computer equipment. Figure 1 FIG. 1 is a flow chart of a method for removing artifacts from a scanned image according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0050] Step S101: Acquire a scanned image to be corrected with ring artifacts.

[0051] The scanned image to be corrected is a computed tomography (CT) image obtained by scanning a human body part using a scanning device, and the scanned image to be corrected contains a ring artifact. The scanning device is communicatively connected to a computer device, and the scanning device can transmit the scanned image to be corrected to the computer device. Correspondingly, the computer device can receive the scanned image to be corrected containing the ring artifact.

[0052] It should be noted that a scanned image to be corrected with a ring artifact can also be generated through simulation. Specifically, a computer device can simulate real scan data to generate a scanned image, and a simulation tool (such as MATLAB software) can be used to add a simulated ring artifact to the real scanned image. The simulated ring artifact is generated as follows: a Radon transform is performed on each layer in each scanned image to generate forward projection data; a random gain perturbation is added to the forward projection data to generate the perturbed forward projection data; and an inverse Radon transform is performed on the randomly perturbed forward projection data, which generates multiple concentric arc-shaped artifacts in the scanned image.

[0053] Here, the scanned image to be corrected is generated by simulating the scanned data and the ring artifact, so as to facilitate internal testing of artifact elimination using the simulated data to be corrected to verify the artifact elimination effect.

[0054] Step S102 : converting the ring artifacts in the scanned image to be corrected into strip artifacts to obtain a first image with strip artifacts.

[0055] Strip artifacts are easier to remove than ring artifacts. After obtaining the scanned image to be corrected, polar coordinate transformation is performed on the image to be corrected to transform the ring artifacts into strip artifacts, thereby generating a first image with strip artifacts.

[0056] Step S103: Using a preset unsupervised network model to eliminate stripe artifacts in the first image to generate a second image.

[0057] The preset unsupervised network model is a pre-trained neural network model for eliminating stripe artifacts, for example, a generative adversarial network based on dual contrast learning (DCLGAN). The training process of the preset unsupervised network model will be described in detail in the following embodiments and will not be repeated here.

[0058] The trained unsupervised network model is deployed on a computer device. The first image is input into the preset unsupervised network model, and the preset unsupervised network model removes the stripe artifacts in the first image to generate a second image. That is, the second image contains image details but does not contain stripe artifacts.

[0059] Step S104: performing artifact removal based on the first image and the second image to obtain a target scan image.

[0060] The first image contains stripe artifacts and image details, while the second image contains image details. By combining the first and second images for image processing, a stripe artifact image can be obtained. Subsequently, the stripe artifact image is converted into a ring artifact image through coordinate transformation. This ring artifact image is combined with the scanned image to be corrected for image processing. This eliminates the ring artifacts in the scanned image to be corrected, resulting in an artifact-free target scanned image.

[0061] The method for removing artifacts from scanned images provided in this embodiment converts ring artifacts in the scanned image to be corrected into strip artifacts, and then uses a preset unsupervised network model to eliminate the strip artifacts. This allows high-quality images to be generated through comparative learning of the preset unsupervised network model, effectively eliminating ring artifacts in the scanned image. The preset unsupervised network model in this method does not require paired data, which reduces the difficulty of collecting large amounts of annotated training data and improves the training efficiency of the preset unsupervised network model. Furthermore, without the need for additional prior information or hardware modifications, the unsupervised network model in deep learning is fully utilized to solve the serious problem of ring artifacts in scanned images, achieving effective removal of ring artifacts.

[0062] In this embodiment, a method for removing artifacts from a scanned image is provided, which can be used in computer equipment. Figure 2 FIG. 1 is a flow chart of a method for removing artifacts from a scanned image according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0063] Step S201: Obtain a scanned image with ring artifacts to be corrected. Detailed descriptions refer to the corresponding descriptions of the above embodiments, which will not be repeated here.

[0064] Step S202 : converting the ring artifacts in the scanned image to be corrected into stripe artifacts to obtain a first image with stripe artifacts.

[0065] Specifically, the above step S202 may include:

[0066] Step S2021: Obtain the first position of each pixel point in the scanned image to be corrected in the Cartesian coordinate system.

[0067] The first position is the coordinate position of each pixel in the scanned image to be corrected in the Cartesian coordinate system. By constructing a Cartesian coordinate system and placing the scanned image to be corrected in the Cartesian coordinate system, the first position of each pixel in the Cartesian coordinate system can be obtained.

[0068] Step S2022: Convert the first position corresponding to each pixel point into a polar coordinate system to obtain the second position of each pixel point in the polar coordinate system.

[0069] The second position is the coordinate position of each pixel point in the scanned image to be corrected in the polar coordinate system. According to the coordinate conversion rules between the Cartesian coordinate system and the polar coordinate system, the scanned image to be corrected can be converted from the Cartesian coordinate system to the polar coordinate system.

[0070] Specifically, the conversion rules between the Cartesian coordinate system and the polar coordinate system are as follows:

[0071]

[0072] Wherein, (x, y) is the coordinate position in the Cartesian coordinate system, and its corresponding coordinate position in the polar coordinate system is (θ, ρ); M is the size of the scanned image to be corrected, specifically including the width and height of the scanned image to be corrected.

[0073] Step S2023: Obtain a first image with stripe artifacts based on the second position.

[0074] A converted artifact image is generated according to the second position corresponding to each pixel point, that is, the ring artifact is converted into a strip artifact, and a first image with the strip artifact is generated.

[0075] In some optional embodiments, the above step S202 may further include:

[0076] Step a1: Detect and obtain a plurality of unconverted pixel points generated during the coordinate conversion process.

[0077] Step a2: Generate multiple unconverted pixel points using a bilinear interpolation algorithm.

[0078] When converting the scanned image to be corrected from a Cartesian coordinate system to a polar coordinate system, the discrete nature of sampling may result in sampling discontinuities in the polar coordinate system. This is because the number of pixels at different angles and distances in the polar coordinate system may be different, resulting in some pixels having corresponding values in the Cartesian coordinate system but not in the polar coordinate system.

[0079] Here, a bilinear interpolation algorithm is used to generate pixels that cannot be converted due to sampling discontinuities. Specifically, the bilinear interpolation algorithm uses the values of surrounding known pixels to perform a weighted average to estimate the pixel values corresponding to each unconverted pixel. This method can generate a smooth and continuous first image in a polar coordinate system, which is crucial for subsequent tasks such as removing streaking and other artifacts, accurately calculating image features, and performing image segmentation.

[0080] The above-mentioned bilinear interpolation algorithm is used to generate multiple unconverted pixel points to ensure the generation of a smooth and continuous image in the polar coordinate system, avoiding sudden or jagged artifacts or image distortion, so that subsequent image processing and analysis can be more accurate and reliable.

[0081] Step S203: Using a preset unsupervised network model to eliminate stripe artifacts in the first image, and generate a second image. Detailed descriptions refer to the corresponding descriptions of the above embodiments, which will not be repeated here.

[0082] Step S204: performing artifact removal based on the first image and the second image to obtain a target scan image.

[0083] Specifically, the above step S204 may include:

[0084] Step S2041 : performing a difference process based on the first image and the second image to obtain a stripe artifact image.

[0085] The first image is an image with stripe artifacts and image details, and the second image has image details but no stripe artifacts. Subtracting the first image from the second image can eliminate the image details in the first image and obtain an image without stripe artifacts.

[0086] In some optional embodiments, the above step S2041 may include:

[0087] Step b1: Subtract the first image from the second image to generate a residual image.

[0088] Step b2: filtering the residual image to obtain a striped artifact image.

[0089] A first image with streak artifacts is subtracted from a second image without streak artifacts to obtain a residual image with image details and streak artifacts. Subsequently, a predetermined filtering method (e.g., mean filtering, median filtering, etc.) is used to remove image details in the residual image to obtain a corresponding streak artifact-free image.

[0090] The first image with strip artifacts is subtracted from the second image to obtain a residual image with image details and strip artifacts. The residual image is filtered to remove the image details it carries, thereby obtaining a strip artifact image, which facilitates obtaining a ring artifact that does not carry image details. Therefore, when the ring artifact is subsequently eliminated, the image detail information can be maintained, and the anatomical integrity and spatial resolution of the original scan image can be maintained.

[0091] Step S2042 : performing inverse transformation on the stripe artifact image to generate a ring artifact image.

[0092] As described above, the ring-shaped artifacts can be converted into strip-shaped artifacts by performing coordinate transformation. Here, the strip-shaped artifact image can be converted into the ring-shaped artifact image by performing inverse coordinate transformation on the strip-shaped artifact image.

[0093] In some optional embodiments, the above step S2042 may include:

[0094] Step c1: obtaining a third position of each pixel point in the stripe artifact image in the polar coordinate system.

[0095] Step c2: converting the third position corresponding to each pixel point into a Cartesian coordinate system to obtain a fourth position of each pixel point in the Cartesian coordinate system.

[0096] Step c3: generating a ring artifact image based on each pixel point and its corresponding fourth position.

[0097] The third position is the coordinate position of each pixel in the stripe artifact image in the polar coordinate system. The fourth position is the coordinate position of each pixel in the stripe artifact image converted to the Cartesian coordinate system. According to the coordinate conversion rules between the Cartesian coordinate system and the polar coordinate system, the stripe artifact image can be converted from the polar coordinate system to the Cartesian coordinate system. The specific conversion rules are as follows:

[0098]

[0099] Wherein, (θ1, ρ1) is the coordinate position in the polar coordinate system; (x1, y1) is the coordinate position corresponding to (θ1, ρ1) in the Cartesian coordinate system; N is the size of the strip artifact image, specifically including the width and height of the strip artifact image.

[0100] According to the above conversion rule, the fourth position of each pixel point in the Cartesian coordinate system can be obtained. Therefore, by performing image reconstruction based on the fourth position corresponding to each pixel point, the strip artifact image can be converted into a ring artifact image.

[0101] The strip artifact image is converted into a ring artifact image through coordinate system transformation, which is convenient for performing difference processing between the strip artifact image and the scanned image to be corrected, so as to effectively remove the ring artifact in the scanned image, thereby achieving effective removal of the ring artifact.

[0102] Step S2043 : performing a difference process based on the scanned image to be corrected and the ring artifact image to obtain a target scanned image.

[0103] By subtracting the ring artifact image from the scanned image to be corrected with the ring artifact image, the ring artifact image in the scanned image to be corrected can be eliminated to obtain a target scanned image without artifacts.

[0104] Step S205 : obtaining a first voxel value of the scanned image to be corrected and a second voxel value of the target scanned image.

[0105] The first voxel value is used to represent the voxel value of the human body region presented in the scanned image to be corrected. The second voxel value is the voxel value of the human body region presented in the target scanned image. The computer device can obtain the voxel value of the human body region in the scanned image to be corrected by analyzing data such as grayscale values and pixel values of the human body region in the scanned image to be corrected. Similarly, the voxel value of the human body region in the target scanned image can be obtained.

[0106] Step S206 : performing voxel value correction on the target scan image based on the difference between the first voxel value and the second voxel value.

[0107] The first voxel value is compared with the second voxel value to determine the difference between the second voxel value and the first voxel value. In the human body region in the target scanned image, this difference is added to the second voxel value to eliminate the problem of inconsistent voxel distribution, ensuring that the voxel distribution of the target scanned image is consistent with that of the scanned image to be corrected.

[0108] The artifact removal method for scanned images provided in this embodiment is based on the fact that strip artifacts are easier to remove than ring artifacts. Here, the ring artifacts in the scanned image to be corrected are converted into strip artifacts through coordinate system transformation, which facilitates better artifact removal in the scanned image and improves the artifact removal effect. A first image with strip artifacts is subtracted from a second image with strip artifacts removed to obtain a strip artifact image, which is then converted into a ring artifact image through coordinate transformation. The scanned image to be corrected and the ring artifact image are subtracted to obtain a target scanned image with the ring artifacts removed. This allows for maximum preservation of image detail information, thereby ensuring the quality of subsequent reconstructed images. By correcting the voxel values of the target scanned image, the voxel distribution of the corrected target scanned image is consistent with that of the original scanned image to be corrected, making the ring artifact removal effect more pronounced.

[0109] In some optional embodiments, the preset unsupervised network model in step S202 includes a generator and a discriminator, such as Figure 3 As shown, a preset unsupervised network model is trained in a polar coordinate system. The training method of the preset unsupervised network model includes:

[0110] Step S301: input slices of the first image into a generator to obtain a composite image.

[0111] The generator is used to convert the first image into a composite image so that the composite image is close to the second image. In some optional embodiments, the generator includes multiple downsampling units, multiple residual units, and multiple upsampling units. Accordingly, the above step S301 may include:

[0112] Step d1: extract features from slices of the first image to obtain a feature map of the first dimension.

[0113] In step d2, the feature map is input into multiple downsampling units to obtain a downsampled feature map of the second dimension.

[0114] In step d3, the downsampled feature map is input into multiple residual units, and a residual feature map is output.

[0115] In step d4, the residual feature map is input into multiple upsampling units to obtain a composite image, where the dimension of the composite image is equal to that of the first image.

[0116] The network structure of the generator G is composed of a U-type network Unet and a residual network Resnet. It mainly includes several downsampling units, several residual units based on Resnet, and several upsampling units. The number of downsampling units and upsampling units is the same. The first dimension is used to represent the size of the feature map, and the second dimension is used to represent the size of the downsampled feature map.

[0117] Here, we take 2 downsampling blocks, 9 Resnet-based residual blocks, and 2 upsampling blocks as an example to illustrate the generation process of the synthetic image:

[0118] First, the first image is divided into multiple slices of the same size, and the slices of the first image are sequentially passed through the convolution layer, the normalization layer, and the ReLU activation layer for feature extraction to obtain a feature map of size 256×256.

[0119] Next, the feature map of the first dimension is fed into two serially connected downsampling units, reducing the size of the feature map from 256×256 to 64×64. To mitigate the difficulties and degradation of deep neural network training, nine residual units are added to the network architecture of the generator G. Each residual unit consists of several (e.g., two) convolutional layers, several (e.g., two) normalization layers, and a residual layer. The downsampled feature map passes through the convolutional, normalization, and residual layers in sequence, outputting a residual feature map.

[0120] Then, the residual feature map is input into two serially connected upsampling units to obtain a feature map with the same size as the first image. The feature map output by the downsampling unit is the composite image.

[0121] By combining multiple downsampling units, multiple residual units, and multiple upsampling units to obtain a synthetic image, the problems of network model training difficulty and degradation are eliminated, ensuring the training effect of the unsupervised network model.

[0122] Step S302: input the composite image into the discriminator to obtain a recognition result for the composite image.

[0123] The discriminator receives the synthesized image output by the generator, compares the synthesized image with the real image without stripe artifacts, and outputs an array of all 0s or all 1s to distinguish between real samples and fake samples. Specifically, if the discriminator outputs an array of all 0s, it is classified as a fake sample; if it outputs an array of all 1s, it is classified as a real sample.

[0124] In some optional embodiments, the discriminator includes multiple downsampling units. Accordingly, the above step S302 may include:

[0125] Step e1: crop the composite image into multiple image blocks of preset dimensions.

[0126] In step e2, each image block is sequentially input into a plurality of down-sampling units, and the last down-sampling unit outputs the discrimination result corresponding to each image block.

[0127] Step e3: perform mean processing on each discrimination result to obtain the recognition result.

[0128] The preset dimension is used to represent the size of the image block. rCT (i.e., detecting whether the synthesized image is consistent with the real image without stripe artifacts) adopts the PatchGAN discriminator architecture, that is, using local image blocks (patches) of size 70x70 and assigning a result to each patch. This is equivalent to cropping the synthesized image into multiple overlapping patches of size 70x70, and inputting each patch into the discriminator to obtain the recognition result for each patch. The individual recognitions are averaged to obtain the final recognition result.

[0129] Discriminator D rCT It contains multiple downsampling units, each of which includes a convolutional layer, a normalization layer, and a ReLU layer. Taking five downsampling units as an example, each image block is sequentially input into the five concatenated downsampling units, reducing the image block size from 256×256 to 14×14. Each pixel in the final layer's output image is converted to the range [-1, 1] to correspond to the discrimination result for each patch. The individual discrimination results are averaged to obtain the average discrimination result, which is the final recognition result. This patch-level discriminator architecture has fewer parameters than the entire image-level discriminator, simplifying the discriminator network architecture while achieving image discrimination.

[0130] Step S303: Based on the synthesized image and the recognition result, the generator and the discriminator are iterated using the back propagation algorithm until the discriminator determines that the synthesized image output by the generator is an image without stripe artifacts.

[0131] The discriminator inputs its recognition results for the synthesized image into the generator, allowing the generator to combine the recognition results to determine the difference between the synthesized image and the image without striping artifacts. The generator can then optimize its generation of the synthesized image to make it as close as possible to the image without striping artifacts.

[0132] During training, the back-propagation algorithm is used to train the generator and discriminator separately. Both are continuously iteratively optimized, so that the synthetic image output by the generator continues to approach the image without stripe artifacts. At the same time, the recognition ability of the discriminator is also continuously enhanced, thereby improving the quality and credibility of the generated image.

[0133] It should be noted that the above-mentioned preset unsupervised network model can also include a generator F and a discriminator D uCT Among them, the generator F is used to generate a synthetic image with stripe artifacts from an image without stripe artifacts; the discriminator D uCT Used to identify whether the synthetic image with stripe artifacts is consistent with the first image with stripe artifacts. Generator F and Discriminator D uCTThe training method is the same as the above generator G and discriminator D rCT The same, no further description here.

[0134] By cropping the synthesized image into multiple image blocks and using multiple downsampling units to identify each image block, the recognition results corresponding to each image block are obtained, thereby reducing the architectural parameters of the discriminator and making the training of the discriminator more convenient.

[0135] Therefore, in the process of training the unsupervised network model, through the continuous iteration of the generator and the discriminator, the synthetic image output by the generator is constantly close to the real image, and at the same time the identification ability of the discriminator is constantly enhanced, thereby improving the quality and credibility of the synthetic image.

[0136] In some optional embodiments, the above method may further include:

[0137] Step f1, based on the synthetic image output by the preset unsupervised network model, determining the loss function of the preset unsupervised network model.

[0138] Step f2: Optimize the preset unsupervised network model based on the loss function.

[0139] The loss function of the pre-set unsupervised network model mainly consists of three parts: adversarial loss, PatchNCEloss, and consistency loss. The pre-set unsupervised network model uses adversarial loss to promote the visual similarity between the synthesized image sCT and the image rCT without striping artifacts, making the distribution patterns of the two as consistent as possible.

[0140] Specifically, the adversarial loss of the generator G is as follows:

[0141] L GAN (G, D rCT , uCT, rCT)

[0142] =E rct～rCT [log D rCT (rct)]+E uct～uCT [log(1-D rCT (G(uct)))]

[0143] Among them, L GAN (G, D rCT , uCT, rCT) represents the adversarial loss of the generator G; uCT is an image with stripe artifacts; rCT is an image without stripe artifacts; E rct～rCT Indicates the distribution state of rCT image; E uct～uCT represents the distribution state of the uCT image; G represents the generator G, which is used to generate an image without stripe artifacts from an image with stripe artifacts; Dr CTRepresents the discriminator for rCT images.

[0144] The adversarial loss of generator F is similar to that of generator G, and its adversarial loss is as follows:

[0145] L GAN (F, D CT , uCT, rCT)

[0146] =E uct～uCT [logD uCT (uct)]+E rct～rCT [log(1-D uCT (F(rct)))]

[0147] Among them, L GAN (F, D CT , uCT, rCT) represents the adversarial loss of the generator F; uCT is an image with stripe artifacts; rCT is an image without stripe artifacts; E rct～rCT Indicates the distribution state of rCT image; E uct～uCT represents the distribution state of the uCT image; F represents the generator F, which is used to generate an image with stripe artifacts from an image without stripe artifacts; D uCT Represents the discriminator for uCT images.

[0148] A noise contrast estimation framework is used to maximize the mutual information between the corresponding patches of the image uCT with stripe artifacts and the synthetic image sCT. Specifically, the patches in the image are regarded as objects to construct positive and negative samples. The idea of contrastive learning is used to associate and contrast the two example samples of the "query sample" and its "positive sample" with other example samples ("negative samples") in the dataset. The same position is selected in the generated sCT and uCT, and we denote the query sample, positive sample, and negative sample as and Here, an (N+1)-way classification problem is established, using cross entropy loss, which is expressed as follows:

[0149]

[0150] τ is used to control the distance between the query sample and other positive and negative samples, and its default value is 0.07. Represents query sample v and positive sample v + By minimizing l, the mutual information between the query sample and the positive sample is maximized, and the mutual information between the query sample and the negative sample is minimized.

[0151] In two different image domains, two different embedding layers are used to extract effective independent features of uCT and rCT. enc and H uCT , used to extract uCT image domain features. Similarly, EmbedingrCT is composed of F enc and H rCT The composition is used to extract rCT image domain features. In order to learn the independent specificity between the two image domains of uCT and rCT, the weights are not shared between the two Embeddings.

[0152] From G enc (uCT) selects L layer features and feeds them into a two-layer MLP projection head H uCT , encode the uCT image into features and stack Generated features. Where l∈{1,2,3...,L} represents the selected output features of the lth layer. In fact, each feature is a patch in the image. The spatial position of each selected layer is represented by s∈{1,2,3...,S l}, S l Indicates the number of spatial positions in the lth layer. Each time we select a query sample query, we find the corresponding feature as the positive feature The remaining features are considered as negative features C l Represents the number of channels in each layer. Similarly, the output rCT is encoded into Therefore, our goal is to match the corresponding patches of uCT and rCT. The loss function PatchNCE loss for converting uCT to rCT image can be expressed as:

[0153]

[0154]

[0155] In the process of converting rCT images into uCT,

[0156] In order to avoid unnecessary changes in the generator and ensure that the generated sCT maintains the same structure as the original uCT, we use consistency loss to improve training efficiency. The consistency loss Identity loss can be expressed as:

[0157] L identity (G, F) = E uct～uCT [||F(uct)-uct||1]+E rCT[||G(rct)-rct||1]

[0158] Finally, the total loss function can be expressed as:

[0159] L(G, F, D uCT , D rCT , H uCT , H rCT )

[0160] =λ GAN (L GAN (G, D rCT ,uCT,rCT)+L GAN (F, D uCT ,uCT,rCT))+λ NCE L PatchNCEuCT (G, H uCT , H rCT , uCT)+λ NCE L PatchNCErCT (F, H uCT , H rCT , rCT)+λ idt L identity (G, F)

[0161] Then, the preset unsupervised network model is optimized through the determined loss function, so that the images generated by the preset unsupervised network model are more visually similar to the real scanned images, so that the distribution patterns of the two are as consistent as possible, thereby maximizing the artifact removal effect of the scanned images.

[0162] As one or more specific application examples of the present invention, the above method is described here in conjunction with the elimination of ring artifacts in a specific scanned image. Figure 4 The flowchart of the artifact removal method shown in FIG. 1 mainly includes:

[0163] (1) Adding simulated ring artifacts: In the real CT data Io ri The simulated ring artifact is added to obtain the original CT image, that is, the scanned image to be corrected I0.

[0164] (2) Polar coordinate transformation: The scanned image to be corrected I0 with annular artifacts in the Cartesian coordinate system is transformed into the first image P0 with stripe artifacts in the polar coordinate system through polar coordinate transformation.

[0165] (3) Constructing a synthetic network: Applying a preset unsupervised network model in a polar coordinate system to eliminate the stripe artifacts in the first image P0 to generate the second image P f , and then P0 and P f The residual image P with image details and stripe artifacts is generated by subtraction res, remove the image details by mean filtering to obtain the stripe artifact image P ring .

[0166] (4) Polar coordinate inverse transformation: for the stripe artifact image P ring Perform inverse coordinate transformation to obtain the ring artifact image I in the Cartesian coordinate system ring and use I0 minus I ring Get the corrected target scan image I corr .

[0167] The above method was applied to simulated scanned images to be corrected and tested. The experimental results are as follows: Figure 5 and Figure 6 As shown, the display window width / center is set to 500 and 0HU respectively.

[0168] Figure 5 The following are histograms comparing the standard deviations of selected regions of interest (ROIs) in CT images before and after correction. The first row shows the axial image, and the second row shows the standard deviations of selected ROIs in the CT images before and after correction. (a) is the uncorrected CT image with ring artifacts; (b) is the corrected image; and (c) is the reference image without ring artifacts. The results show that the corrected CT image has reduced noise compared to the original scan image to be corrected.

[0169] Figure 6 The following are histograms comparing the standard deviations of selected ROIs in simulated CT data before and after correction. The first row shows the simulated CT images, and the second row shows histograms comparing the standard deviations of selected ROIs in the simulated CT images before and after correction. (a) is the uncorrected CT image with ring artifacts; (b) is the corrected image; and (c) is the reference image without ring artifacts. The results show that the noise in the simulated CT images is reduced after correction.

[0170] Furthermore, the above-mentioned artifact removal method of scanned images was tested and evaluated on 21 patients with a total of 2944 CT images. The experimental results are as follows: Figure 7-Figure 9 As shown, the display window width / center is set to 500 and 0HU respectively.

[0171] Figure 7Schematic diagram of artifact removal results on real patient CT images. The first row shows axial images, and the images in the second row are magnified images of the indicated boxes indicated by the dotted lines in the first row. (a) is an uncorrected CT image with ring artifacts; (b) is the corrected image; (c) is a reference image without ring artifacts; (d) is the difference between (a) and (b); (e) and (f) are magnified images of the box in (a); (g) and (h) are magnified images of the box in (b); (i) and (j) are magnified images of the box in (c); and (k) and (l) are magnified images of the box in (d). The results show that this artifact removal method for scanned images significantly suppresses ring artifacts in CT images.

[0172] Figure 8 Figure 1 shows the CT value changes at two lines in a CT image. The first row shows the axial image, and the second row shows the CT value changes at the vertical and horizontal lines in the first row. (a) is an uncorrected CT image with ring artifacts; (b) is the corrected image; and (c) is a reference image without ring artifacts.

[0173] Figure 9 Schematic diagram of a uniform axial CBCT slice of a patient. (a) is an uncorrected image with ring artifacts; (b) is the image corrected using the above method; (c) is the result of voxel correction based on (b); (d) and (e) show the CT value changes at the vertical and horizontal lines in the first row.

[0174] pass Figure 7-Figure 9 ,It can be clearly seen that compared with the CT image before correction, the ring artifacts of the ,corrected CT image have been removed, and the image detail data are ,retained. At the same time, the corrected CT image matches the real reference CT image well.

[0175] This embodiment also provides a device for removing artifacts from a scanned image. This device is used to implement the above-mentioned embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0176] This embodiment provides a device for removing artifacts from a scanned image. Figure 10 Shown, including:

[0177] The image acquisition module 401 is used to acquire a scanned image with ring artifacts to be corrected.

[0178] The image conversion module 402 is configured to convert the ring artifacts in the scanned image to be corrected into stripe artifacts, thereby obtaining a first image with stripe artifacts.

[0179] The artifact removal module 403 is configured to remove strip artifacts in the first image based on a preset unsupervised network model to generate a second image.

[0180] The correction module 404 is configured to remove artifacts based on the first image and the second image to obtain a target scan image.

[0181] In some optional embodiments, the image conversion module 402 includes:

[0182] The first position acquisition unit is used to acquire the first position of each pixel point in the scanned image to be corrected in a Cartesian coordinate system.

[0183] The second position determining unit is used to convert the first position corresponding to each pixel point into a polar coordinate system to obtain the second position of each pixel point in the polar coordinate system.

[0184] The second image generating unit is configured to obtain the first image with stripe artifacts based on the second position.

[0185] In some optional embodiments, the image conversion module 402 may further include:

[0186] The conversion detection unit is used to detect and obtain a plurality of unconverted pixel points generated during the coordinate conversion process.

[0187] The pixel generation unit is used to generate a plurality of unconverted pixel points by using a bilinear interpolation algorithm.

[0188] In some optional embodiments, the correction module 404 includes:

[0189] The first processing unit is configured to perform difference processing based on the first image and the second image to obtain a stripe artifact image.

[0190] The inverse transform unit is used to perform inverse transform processing on the strip artifact image to generate a ring artifact image.

[0191] The second processing unit is configured to perform a difference process based on the scanned image to be corrected and the ring artifact image to obtain a target scanned image.

[0192] In some optional embodiments, the first processing unit includes:

[0193] The residual image generating subunit is used to perform a difference between the first image and the second image to generate a residual image.

[0194] The filtering processing subunit is used to perform filtering processing on the residual image to obtain a striped artifact image.

[0195] In some optional embodiments, the inverse transform unit includes:

[0196] The third position acquisition subunit is used to acquire the third position of each pixel point in the strip artifact image in the polar coordinate system.

[0197] The fourth position determining subunit is configured to convert the third position corresponding to each pixel point into a Cartesian coordinate system to obtain a fourth position of each pixel point in the Cartesian coordinate system.

[0198] The ring artifact generating subunit is configured to generate a ring artifact image based on each pixel point and its corresponding fourth position.

[0199] In some optional embodiments, the apparatus for removing artifacts from a scanned image further includes:

[0200] The voxel acquisition unit is used to acquire a first voxel value of the scanned image to be corrected and a second voxel value of the target scanned image.

[0201] The voxel correction unit is configured to perform voxel value correction on the target scan image based on a difference between the first voxel value and the second voxel value.

[0202] In some optional embodiments, the apparatus for removing artifacts from a scanned image further includes:

[0203] The model training unit is used to train the preset unsupervised network model in the polar coordinate system.

[0204] In some optional embodiments, the preset unsupervised network model includes a generator and a discriminator, and the above-mentioned model training unit includes:

[0205] The image synthesis subunit is used to input the slices of the first image into the generator to obtain a synthesized image.

[0206] The image discrimination subunit is used to input the composite image into the discriminator to obtain a recognition result for the composite image.

[0207] The iterative subunit is used to iterate the generator and the discriminator using a back-propagation algorithm based on the synthetic image and the recognition result until the discriminator determines that the synthetic image output by the generator is an image without stripe artifacts.

[0208] In some optional embodiments, the generator includes multiple downsampling units, multiple residual units and multiple upsampling units, and the above-mentioned image synthesis subunit is specifically used to: perform feature extraction on slices of the first image to obtain a feature map of the first dimension; input the feature map into multiple downsampling units to obtain a downsampling feature map of the second dimension; input the downsampling feature map into multiple residual units to output a residual feature map; input the residual feature map into multiple upsampling units to obtain a synthesized image, and the dimension of the synthesized image is equal to the dimension of the first image.

[0209] In some optional embodiments, the discriminator includes multiple downsampling units, and the above-mentioned image discrimination subunit is specifically used to: crop the composite image into multiple image blocks of preset dimensions; input each image block into multiple downsampling units in turn, and output the discrimination results corresponding to each image block through the last downsampling unit; perform mean processing on each discrimination result to obtain a recognition result.

[0210] In some optional embodiments, the model training unit further includes:

[0211] The loss function determination subunit is used to determine the loss function of the preset unsupervised network model based on the synthetic image output by the preset unsupervised network model.

[0212] The model optimization subunit is used to optimize the preset unsupervised network model based on the loss function.

[0213] The further functional description of the above modules, units and sub-units is the same as that of the above corresponding embodiments and will not be repeated here.

[0214] The device for removing artifacts from scanned images in this embodiment is presented in the form of functional units, where the units refer to ASIC circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0215] The embodiment of the present invention also provides a computer device having the above Figure 10 The artifact removal device of the scanned image is shown.

[0216] See also Figure 11 , Figure 11 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 11As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 11 A processor 10 is taken as an example.

[0217] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0218] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0219] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0220] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0221] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0222] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0223] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for removing artifacts from a scanned image, characterized in that: The method comprises: Acquire a scan image to be corrected with a ring artifact; Converting the ring artifacts in the scanned image to be corrected into stripe artifacts to obtain a first image with the stripe artifacts; Using a preset unsupervised network model to eliminate the stripe artifacts in the first image to generate a second image; Performing artifact removal based on the first image and the second image to obtain a target scanned image, including: performing subtraction processing on the first image and the second image to obtain a striped artifact image; performing inverse transformation processing on the striped artifact image to generate a ring artifact image; performing subtraction processing on the scanned image to be corrected and the ring artifact image to obtain the target scanned image; The step of performing a difference process based on the first image and the second image to obtain a striped artifact image includes: performing a difference process on the first image and the second image to generate a residual image; and performing a filtering process on the residual image to obtain the striped artifact image. The strip artifact image is inversely transformed to generate a ring artifact image, including: obtaining the third position of each pixel point in the strip artifact image in the polar coordinate system; converting the third position corresponding to each pixel point to the Cartesian coordinate system to obtain the fourth position of each pixel point in the Cartesian coordinate system; and generating the ring artifact image based on the each pixel point and its corresponding fourth position.

2. The method according to claim 1, characterized in that The step of converting the ring artifacts in the scanned image to be corrected into stripe artifacts to obtain a first image with the stripe artifacts includes: Obtaining a first position of each pixel point in the scanned image to be corrected in a Cartesian coordinate system; Converting the first position corresponding to each pixel point into a polar coordinate system to obtain a second position of each pixel point in the polar coordinate system; Based on the second position, a first image with the stripe artifact is obtained.

3. The method according to claim 2, characterized in that The method further comprises: Detecting and obtaining multiple unconverted pixel points generated during the coordinate conversion process; The unconverted pixel points are generated by using a bilinear interpolation algorithm.

4. The method according to claim 1, wherein Also includes: Acquire a first voxel value of the scanned image to be corrected and a second voxel value of the target scanned image; Based on the difference between the first voxel value and the second voxel value, voxel value correction is performed on the target scan image.

5. The method according to claim 1, wherein The preset unsupervised network model includes a generator and a discriminator, and the training method of the preset unsupervised network model includes: Inputting a slice of the first image into a generator to obtain a composite image; Inputting the composite image into a discriminator to obtain a recognition result for the composite image; Based on the synthetic image and the recognition result, the generator and the discriminator are iterated using a back propagation algorithm until the discriminator determines that the synthetic image output by the generator is an image without stripe artifacts.

6. The method according to claim 5, characterized in that The generator includes a plurality of downsampling units, a plurality of residual units, and a plurality of upsampling units, and the slices of the first image are input into the generator to obtain a synthesized image, including: Performing feature extraction on the slice of the first image to obtain a feature map of a first dimension; Inputting the feature map into the multiple downsampling units to obtain a downsampled feature map of the second dimension; Inputting the downsampled feature map into the plurality of residual units, and outputting a residual feature map; The residual feature map is input into the multiple upsampling units to obtain a composite image, where the dimension of the composite image is equal to the dimension of the first image.

7. The method according to claim 5, characterized in that The discriminator includes a plurality of downsampling units, and the synthesized image is input into the discriminator to obtain a recognition result for the synthesized image, including: Cropping the composite image into a plurality of image blocks of preset dimensions; Inputting each of the image blocks into the multiple downsampling units in sequence, and outputting the discrimination results corresponding to each of the image blocks through the last downsampling unit; Perform mean processing on each of the discrimination results to obtain the recognition result.

8. The method according to claim 5, characterized in that Also includes: Determining a loss function of the preset unsupervised network model based on the synthetic image output by the preset unsupervised network model; The preset unsupervised network model is optimized based on the loss function.

9. A device for removing artifacts from a scanned image, characterized in that: The device comprises: An image acquisition module, used for acquiring a scanned image to be corrected with a ring artifact; An image conversion module, configured to convert the ring artifacts in the scanned image to be corrected into stripe artifacts, thereby obtaining a first image with the stripe artifacts; an artifact removal module, configured to remove the stripe artifacts in the first image based on a preset unsupervised network model to generate a second image; a correction module, configured to remove artifacts based on the first image and the second image to obtain a target scan image; The correction module includes: a first processing unit for performing a difference process based on the first image and the second image to obtain a stripe artifact image; an inverse transformation unit for performing an inverse transformation process on the stripe artifact image to generate a ring artifact image; and a second processing unit for performing a difference process based on the scanned image to be corrected and the ring artifact image to obtain the target scanned image. The first processing unit includes: a residual image generating subunit for performing a difference between the first image and the second image to generate a residual image; a filtering processing subunit for performing a filtering process on the residual image to obtain the stripe artifact image; Among them, the inverse transformation unit includes: a third position acquisition subunit, used to obtain the third position of each pixel point in the strip artifact image in the polar coordinate system; a fourth position determination subunit, used to convert the third position corresponding to each pixel point to the Cartesian coordinate system to obtain the fourth position of each pixel point in the Cartesian coordinate system; a ring artifact generation subunit, used to generate the ring artifact image based on the each pixel point and its corresponding fourth position.

10. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for removing artifacts from a scanned image according to any one of claims 1 to 8 by executing the computer instructions.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for removing artifacts from a scanned image according to any one of claims 1 to 8.

Images