Artifact correction method and device, magnetic resonance image processing method and device and computer equipment

Abstract

The invention relates to an artifact correction method and device, a magnetic resonance image processing method and device and computer equipment. The method comprises the following steps: acquiring a to-be-corrected image containing artifacts; performing down-sampling processing on the to-be-corrected image according to different down-sampling templates to obtain target down-sampling images corresponding to the down-sampling templates; inputting the target downsampling images into a pre-trained image reconstruction model to obtain reconstructed images corresponding to the target downsampling images; and weighting the reconstructed image to obtain an artifact correction image of the to-be-corrected image. By adopting the method, the artifacts in the magnetic resonance image can be reliably removed.

Description

Technical Field

[0001] The present application relates to the technical field of medical image processing, and in particular to an artifact correction, magnetic resonance image processing method, device and computer equipment. Background Art

[0002] As an important means of medical imaging, magnetic resonance imaging is widely used in clinical diagnosis. During the imaging process, the subject may move involuntarily and be affected by breathing, heartbeat, and the pulsation of blood vessels and spinal fluid. It is difficult to keep the scanned part still, resulting in motion artifacts in the scanned magnetic resonance images, which in turn affects the accuracy of clinical diagnosis.

[0003] In the prior art, a magnetic resonance image with motion artifacts and a corresponding magnetic resonance image without motion artifacts can be collected under pre-set constraints as training data sets to train a deep neural network, and motion artifacts in magnetic resonance images can be corrected based on the trained deep neural network. However, this method is only applicable to scenarios where motion artifacts occur that are the same or similar to the pre-set constraints. In the face of complex and changeable actual clinical applications, it is difficult to reliably remove motion artifacts.

[0004] Therefore, current artifact correction techniques for magnetic resonance images are unreliable. Summary of the invention

[0005] Based on this, it is necessary to provide an artifact correction, magnetic resonance image processing method, device, computer equipment, computer-readable storage medium and computer program product that can reliably remove artifacts in order to solve the above technical problems.

[0006] In a first aspect, the present application provides an artifact correction method. The method comprises:

[0007] Acquire an image to be corrected that contains artifacts;

[0008] Downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0009] Inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0010] The reconstructed image is weighted to obtain an artifact-corrected image of the image to be corrected.

[0011] In one embodiment, the downsampling template includes a first area and a second area, the first area corresponds to the low-frequency information of the image to be corrected in the phase encoding direction, the second area corresponds to the high-frequency information of the image to be corrected in the phase encoding direction, the sampling channel of the first area is full-pass, and the sampling channel of the second area is Gaussian distributed.

[0012] In one embodiment, downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template includes:

[0013] Determining target K-space data corresponding to the image to be corrected;

[0014] According to each of the downsampling templates, downsampling processing is performed on the phase encoding direction of the target K-space data respectively to obtain downsampling data corresponding to the target K-space data;

[0015] Performing inverse Fourier transform on each of the downsampled data to obtain the target downsampled image.

[0016] In one embodiment, determining the target K-space data corresponding to the image to be corrected includes:

[0017] Performing Fourier transform on the image to be corrected to obtain original K-space data corresponding to the image to be corrected;

[0018] Performing cropping processing on the original K-space data to obtain cropped K-space data;

[0019] Channel compression processing is performed on the cropped K-space data to obtain the target K-space data.

[0020] In a second aspect, the present application provides a method for processing magnetic resonance images. The method comprises:

[0021] Acquire an image to be corrected; the image to be corrected is a magnetic resonance image;

[0022] Performing downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, to obtain downsampling data sets corresponding to each sampling trajectory;

[0023] Inputting the target downsampled images corresponding to each of the downsampled data sets into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0024] The reconstructed image is weightedly processed to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

[0025] In one of the embodiments, the different sampling trajectories form mutual patterns in the K space along the phase encoding direction.

[0026] In one of the embodiments, when the target K-space dataset corresponding to the image to be corrected is a three-dimensional K-space dataset, after acquiring the image to be corrected, the method further includes:

[0027] A one-dimensional inverse Fourier transform is performed on the original K-space data set corresponding to the image to be corrected in the plane phase encoding direction.

[0028] In a third aspect, the present application also provides an artifact correction device. The device comprises:

[0029] A first acquisition module, used for acquiring an image to be corrected containing artifacts;

[0030] A first downsampling module is used to perform downsampling processing on the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0031] A first reconstruction module, used for inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0032] The first correction module is used to perform weighted processing on the reconstructed image to obtain an artifact-corrected image of the image to be corrected.

[0033] In a fourth aspect, the present application also provides a magnetic resonance image processing device. The device comprises:

[0034] A second acquisition module is used to acquire an image to be corrected; the image to be corrected is a magnetic resonance image;

[0035] A second downsampling module is used to perform downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, so as to obtain downsampling data sets corresponding to each sampling trajectory;

[0036] A second reconstruction module, used for inputting the target downsampled images corresponding to each of the downsampled data sets into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0037] The second correction module is used to perform weighted processing on the reconstructed image to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

[0038] In a fifth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0039] Acquire an image to be corrected that contains artifacts;

[0040] Downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0041] Inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0042] The reconstructed image is weighted to obtain an artifact-corrected image of the image to be corrected.

[0043] In a sixth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0044] Acquire an image to be corrected that contains artifacts;

[0045] Downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0046] Inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0047] The reconstructed image is weighted to obtain an artifact-corrected image of the image to be corrected.

[0048] In a seventh aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0049] Acquire an image to be corrected that contains artifacts;

[0050] Downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0051] Inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0052] The reconstructed image is weighted to obtain an artifact-corrected image of the image to be corrected.

[0053] The above-mentioned artifact correction, magnetic resonance image processing method, device, computer equipment, storage medium and computer program product obtain an image to be corrected containing artifacts, perform downsampling processing on the image to be corrected according to different downsampling templates, obtain a target downsampling image corresponding to each downsampling template, input each target downsampling image into a pre-trained image reconstruction model, obtain a reconstructed image corresponding to each target downsampling image, perform weighted processing on the reconstructed image, and obtain an artifact-corrected image of the image to be corrected; the artifacts in the magnetic resonance image can be modeled as a number of lines missing in the K-space phase encoding direction, and the K-space data of the magnetic resonance image can be downsampled using multiple different downsampling templates to remove the influence of the missing lines on the target downsampling image corresponding to the K-space data, and the target downsampling image can be reconstructed and weighted, so as to reliably remove the artifacts in the magnetic resonance image. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a schematic flow chart of an artifact correction method in one embodiment;

[0055] Figure 2 is a schematic diagram of a downsampling template in one embodiment;

[0056] Figure 3 is a schematic diagram of an artifact correction process in one embodiment;

[0057] Figure 4 A schematic diagram of a flow chart of an image reconstruction model training method in one embodiment;

[0058] Figure 5 A schematic diagram of a cyclic generation model in one embodiment;

[0059] Figure 6 A schematic diagram of a discriminator of a cyclic generation model in one embodiment;

[0060] Figure 7 A schematic diagram of a generator of a cyclic generation model in one embodiment;

[0061] Figure 8 is a schematic flow chart of a magnetic resonance image processing method in one embodiment;

[0062] Fig. 9 A comparison diagram of artifact correction of abdominal magnetic resonance images in one embodiment;

[0063] Fig.10 A comparison diagram of artifact correction of brain magnetic resonance images in one embodiment;

[0064] Fig.11 is a structural block diagram of an artifact correction device in one embodiment;

[0065] Fig.12 is a structural block diagram of a magnetic resonance image processing device in one embodiment;

[0066] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0068] In one embodiment, Figure 1 As shown, a method for artifact correction is provided. This embodiment takes the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0069] Step S110, obtaining an image to be corrected containing artifacts.

[0070] The image to be corrected may be a medical image, including but not limited to a magnetic resonance image.

[0071] The artifacts may be artifacts in medical images, including but not limited to motion artifacts.

[0072] In a specific implementation, a medical imaging device can be used to scan the detection object to obtain a medical image, and the obtained medical image can be identified. If artifacts are identified, the medical image is input into the terminal so that the terminal obtains the image to be corrected containing the artifacts.

[0073] In practical applications, the magnetic resonance images scanned by the magnetic resonance imaging device can be manually identified or identified through a neural network. If motion artifacts are identified in the magnetic resonance images, they are input into the terminal as images to be corrected containing artifacts.

[0074] Step S120 , downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template.

[0075] The downsampling template may be a template for downsampling a medical image in K space.

[0076] The target downsampled image may be a downsampled image of the image to be corrected.

[0077] In a specific implementation, at least one downsampling template may be pre-set, and the terminal uses each downsampling template to downsample the image to be corrected to obtain at least one target downsampled image, wherein each target downsampled image corresponds to a downsampling template.

[0078] In practical applications, N different downsampling templates can be pre-set, and each downsampling template can downsample the image to be corrected in the phase encoding direction of the K space. During the downsampling process, the image to be corrected can be firstly Fourier transformed to obtain the K space data corresponding to the image to be corrected, and then the K space data can be downsampled using the N downsampling templates to obtain N groups of downsampled data corresponding to the K space data, wherein each group of downsampled data corresponds to a downsampling template, and then the N groups of downsampled data are inverse Fourier transformed to obtain N target downsampled images corresponding to the image to be corrected.

[0079] Specifically, the image to be corrected can be subjected to Fourier transform to obtain a K-space data set corresponding to the image to be corrected, and the K-space data set can be subjected to downsampling processing of different sampling trajectories to obtain multiple downsampled data sets of K space, and each downsampled data set can be subjected to inverse Fourier transform to obtain a corresponding target downsampled image. Different sampling trajectories form mutual patterns in the K-space along the phase encoding direction, that is, the patterns between multiple different sampling trajectories are complementary.

[0080] Step S130 , inputting each target down-sampled image into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each target down-sampled image.

[0081] The image reconstruction model may be a model for reconstructing a downsampled image into a medical image.

[0082] The reconstructed image may be a medical image obtained by reconstructing the target downsampled image.

[0083] In a specific implementation, an image reconstruction model may be pre-trained, and the terminal inputs at least one target downsampled image into the pre-trained image reconstruction model to obtain at least one reconstructed image, wherein each reconstructed image corresponds to a target downsampled image.

[0084] In practical applications, N target downsampled images can be respectively input into a pre-trained image reconstruction model to obtain N reconstructed images, where each reconstructed image corresponds to a target downsampled image.

[0085] Step S140: performing weighted processing on the reconstructed image to obtain an artifact-corrected image of the image to be corrected.

[0086] In a specific implementation, the terminal may perform weighted processing on at least one reconstructed image to obtain an artifact-corrected image for correcting artifacts in the image to be corrected.

[0087] In practical applications, an artifact-corrected image without motion artifacts can be obtained by weighting N reconstructed images, wherein the image quality of the artifact-corrected image is higher than that of the image to be corrected. For example, the N pixel values ​​corresponding to the same voxel in N reconstructed images are accumulated and averaged to obtain an artifact-corrected image, wherein the weight of each reconstructed image is 1 / N.

[0088] The above-mentioned artifact correction method obtains an image to be corrected containing artifacts, performs downsampling processing on the image to be corrected according to different downsampling templates, obtains a target downsampling image corresponding to each downsampling template, inputs each target downsampling image into a pre-trained image reconstruction model, obtains a reconstructed image corresponding to each target downsampling image, and performs weighted processing on the reconstructed image to obtain an artifact corrected image of the image to be corrected; the artifacts in the magnetic resonance image can be modeled as a number of lines missing in the K-space phase encoding direction, and the K-space data of the magnetic resonance image is downsampled using multiple different downsampling templates to remove the influence of the missing lines on the target downsampling image corresponding to the K-space data, and the target downsampling image is reconstructed and weighted, so as to reliably remove the artifacts in the magnetic resonance image.

[0089] In one embodiment, the downsampling template includes a first area and a second area, the first area corresponds to the low-frequency information of the image to be corrected in the phase encoding direction, the second area corresponds to the high-frequency information of the image to be corrected in the phase encoding direction, the sampling channel of the first area is full-pass, and the sampling channel of the second area is Gaussian distributed.

[0090] The first region may be a low-frequency region in the K-space phase encoding direction, and the second region may be a high-frequency region in the K-space phase encoding direction.

[0091] In a specific implementation, each downsampling template may include a first region or a second region, the first region corresponds to the low-frequency information of the image to be corrected in the K-space phase encoding direction, and the second region corresponds to the high-frequency information of the image to be corrected in the K-space phase encoding direction. The sampling channel of the first region may be set to full pass to obtain all the low-frequency information of the image to be corrected, and the sampling channel of the second region may be set to Gaussian distribution to allow only part of the high-frequency information of the image to be corrected to pass.

[0092] Figure 2 A schematic diagram of a downsampling template is provided. Figure 2, N downsampling templates can be set, each downsampling template performs full sampling in the RO (Read Out) direction and performs downsampling in the PE (Phase Encoding) direction. The pixel value in the downsampling template is set to 1 to indicate that the K-space data of the pixel position is retained, and is set to 0 to indicate that the K-space data of the pixel position is not retained. As shown in the figure, several lines in the center of the PE direction can be divided into a first area, and the corresponding pixel values ​​are set to 1 to obtain all low-frequency information of the image to be corrected. The two sides of the first area are the second area. The number and position of the lines set to 1 in the second area are in accordance with the Gaussian distribution, so that part of the high-frequency information of the image to be corrected passes randomly. Furthermore, the number and position of the lines set to 1 in the second area of ​​different downsampling templates can be different.

[0093] In this embodiment, by making the downsampling template include a first area and a second area, the first area corresponds to the low-frequency information of the image to be corrected in the phase encoding direction, the second area corresponds to the high-frequency information of the image to be corrected in the phase encoding direction, the sampling channel of the first area is full-pass, and the sampling channel of the second area is Gaussian distributed, and a plurality of different downsampling templates that conform to the Gaussian distribution and are complementary can be used to downsample the K-space phase encoding direction of the image to be corrected, respectively, to ensure that each line outside the central limited range of the phase encoding direction is under-sampled, thereby ensuring that the lines in the phase encoding direction affected by motion artifacts can be removed, thereby improving the reliability of artifact correction.

[0094] In one embodiment, the above step S120 may specifically include: determining the target K space data corresponding to the image to be corrected; performing downsampling processing on the phase encoding direction of the target K space data according to each downsampling template to obtain downsampling data corresponding to the target K space data; and performing inverse Fourier transform on each downsampling data to obtain a target downsampling image.

[0095] The target K-space data may be the K-space data of the image to be corrected.

[0096] In a specific implementation, the terminal can first perform Fourier transform on the image to be corrected to obtain the target K space data corresponding to the image to be corrected, and then use at least one downsampling template to downsample the target K space data in the phase encoding direction to obtain at least one group of downsampled data corresponding to the target K space data, wherein each group of downsampled data corresponds to a downsampling template, and then perform inverse Fourier transform on each group of downsampled data to obtain at least one target downsampled image.

[0097] Figure 3 A schematic diagram of the artifact correction process is provided. Figure 3First, the image to be corrected is transformed into K-space data through two-dimensional Fourier transform, and then N downsampling templates are used to downsample the K-space data in the phase encoding direction to obtain N groups of downsampled data. After two-dimensional inverse Fourier transform, N target downsampled images are obtained. The generator is used as a pre-trained image reconstruction model to reconstruct the N target downsampled images to obtain N reconstructed images. According to the average value of the N reconstructed images, the artifact corrected image corresponding to the image to be corrected can be obtained.

[0098] In this embodiment, the target K-space data corresponding to the image to be corrected is determined; according to each downsampling template, the phase encoding direction of the target K-space data is downsampled respectively to obtain the downsampled data corresponding to the target K-space data; each downsampled data is inverse Fourier transformed to obtain the target downsampled image, and a plurality of different downsampling templates that conform to the Gaussian distribution and are complementary can be used to downsample the K-space phase encoding direction of the image to be corrected respectively, to ensure that each line outside the central limited range of the phase encoding direction is undersampled, thereby ensuring that the lines in the phase encoding direction affected by motion artifacts can be removed, thereby improving the reliability of artifact correction.

[0099] Moreover, by averaging the reconstructed images to obtain the artifact-corrected images, the motion artifacts in the image to be corrected can be diluted, compensating for the quality loss of the reconstructed image caused by the under-sampling of a single down-sampling template.

[0100] In one embodiment, the step of determining the target K space data corresponding to the image to be corrected may specifically include: performing Fourier transform on the image to be corrected to obtain the original K space data corresponding to the image to be corrected; cropping the original K space data to obtain cropped K space data; and performing channel compression on the cropped K space data to obtain the target K space data.

[0101] The original K-space data may be unprocessed K-space data.

[0102] The cropping process may be a process of rounding down the matrix size of the original K-space data.

[0103] In a specific implementation, the terminal may first perform a Fourier transform on the image to be corrected to obtain the original K space data corresponding to the image to be corrected, and then crop the original K space data by rounding down the matrix size of the original K space data. For example, the matrix size of the original K space data may be rounded down to a multiple of 16 to obtain cropped K space data, and then the cropped K space data may be channel compressed, for example, compressed to 12 channels to obtain target K space data. The size of the target K space data may be (floor(x / 16)*16, floor(y / 16)*16, 12), wherein floor represents rounding down, x represents the number of pixels in the PE direction of the original K space data, and y represents the number of pixels in the RO direction of the original K space data.

[0104] In this embodiment, by performing Fourier transform on the image to be corrected, the original K-space data corresponding to the image to be corrected is obtained; the original K-space data is cropped to obtain cropped K-space data; and the cropped K-space data is channel compressed to obtain target K-space data. This allows images to be corrected of different sizes to match the downsampling template, thereby increasing the universal applicability of the artifact correction method.

[0105] In one embodiment, Figure 4 As shown, a method for training an image reconstruction model is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0106] Step S210, obtaining a fully sampled sample image;

[0107] Step S220, inputting the fully sampled sample image into the cycle generation model to obtain a first cycle consistency loss value, a second cycle consistency loss value, and a discriminator loss value output by the cycle generation model;

[0108] Step S230, training the image reconstruction model to be trained in the cycle generation model according to the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value to obtain a pre-trained image reconstruction model; the pre-trained image reconstruction model is used for artifact correction.

[0109] The fully sampled sample image may be an artifact-free medical image sample.

[0110] Among them, the cycle generation model can be, but is not limited to, a model built based on cycleGAN (cycle generation adversarial network), ResNet (residual neural network) or DenseNet (dense convolutional network), and the cycle generation model can include an image reconstruction model.

[0111] The first cycle consistency loss value may be a loss value of a fully sampled image obtained by sequentially downsampling and image reconstruction. The second cycle consistency loss value may be a loss value of a downsampled image obtained by sequentially reconstructing and downsampling. The discriminator loss value may be a loss value between a fully sampled sample image and a reconstructed image.

[0112] In a specific implementation, the terminal can obtain a fully sampled sample image and input the fully sampled sample image into a cyclic generation model. On the one hand, the cyclic generation model can first use a downsampling template to downsample the fully sampled sample image to obtain a first downsampled image, and then input the first downsampled image into the image reconstruction model to be trained to obtain a first fully sampled image, and determine the first cycle consistency loss value between the fully sampled sample image and the first full sampled image; on the other hand, the cyclic generation model can also first input the second downsampled image corresponding to the fully sampled sample image into the image reconstruction model to be trained to obtain a second fully sampled image, and then use the downsampling template to downsample the second fully sampled image to obtain a third downsampled image, and determine the second cycle consistency loss value between the second downsampled image and the third downsampled image; in addition, the cyclic generation model can also determine the discriminator loss value between the fully sampled sample image and the second full sampled image. The image reconstruction model to be trained is adjusted according to the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value. If the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value do not meet the preset conditions, the step of obtaining the fully sampled sample image is returned until the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value meet the preset conditions to obtain the pre-trained image reconstruction model. The preset condition may be that the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value are all less than the preset loss value.

[0113] Figure 5 A schematic diagram of a cyclic generative model is provided, in which the generator can reconstruct the model for the image to be trained. The generator training process can be Figure 5 The A and B branches in the code are used to implement:

[0114] In branch A, the generator (G θ) is passed through the generator network to obtain a second fully sampled image generated by the generator network. The second fully sampled image is transformed to K space for undersampling by Fourier transform and then transformed back to the image domain to obtain a third downsampled image (undersampled image). The second downsampled image and the third downsampled image are respectively channel-merged and the second cycle consistency loss is calculated.

[0115] In branch B, the generator (G θ ) The fully sampled sample image on the left is transformed to K space for undersampling by Fourier transform, and then transformed back to the image domain to obtain the first downsampled image (undersampled image). The first downsampled image is input into the generator network to obtain the first fully sampled image generated by the generator network. The fully sampled sample image and the first fully sampled image are channel-merged and the first cycle consistency loss is calculated.

[0116] The discriminator training process can be Figure 5 The C branch in the implementation:

[0117] In the C branch, the second fully sampled image generated by the generator network in the A branch is merged through channels and then input into the discriminator together with the image after the full sampled sample image is merged through channels to calculate the discriminator loss.

[0118] It can be seen that in the training process of the above neural network, the original sample image (full sampling sample image) and the image after the sample image is downsampled (the second downsampled image) are used as constraints for the neural network training.

[0119] The above-mentioned image reconstruction model training method obtains a fully sampled sample image, inputs the fully sampled sample image into a cyclic generation model, obtains a first cycle consistency loss value, a second cycle consistency loss value and a discriminator loss value output by the cyclic generation model, and trains the image reconstruction model to be trained in the cyclic generation model according to the first cycle consistency loss value, the second cycle consistency loss value and the discriminator loss value to obtain a pre-trained image reconstruction model; the image reconstruction model can be trained based on the cyclic generation model to improve the accuracy of image reconstruction performed by the image reconstruction model, thereby increasing the accuracy of artifact correction.

[0120] In one embodiment, the above step S220 may specifically include: downsampling the full-sampled sample image according to a preset downsampling template to obtain a first downsampled image; inputting the first downsampled image into the image reconstruction model to be trained to obtain a first full-sampled image output by the image reconstruction model to be trained; and determining a first cycle consistency loss value based on the full-sampled sample image and the first full-sampled image.

[0121] The first down-sampled image may be an image obtained by down-sampling the artifact-free medical image, and the first fully sampled image may be an image obtained by reconstructing the first down-sampled image.

[0122] In the specific implementation, Figure 5 Taking the B branch as an example, for a fully sampled sample image, after Fourier transform, it can be downsampled through a downsampling template, and after inverse Fourier transform, a first downsampled image is obtained, and the first downsampled image is input into the image reconstruction model to be trained for image reconstruction to obtain a first fully sampled image, and the channels of the fully sampled sample image and the first fully sampled image are merged respectively to obtain the channel merged value of the full sampled sample image and the channel merged value of the first full sampled image. According to the difference between the two, the first cycle consistency loss value can be obtained.

[0123] In this embodiment, the full-sampled sample image is downsampled according to a preset downsampling template to obtain a first downsampled image; the first downsampled image is input into the image reconstruction model to be trained to obtain a first full-sampled image output by the image reconstruction model to be trained; a first cycle consistency loss value is determined based on the full-sampled sample image and the first full-sampled image, and the image reconstruction model can be trained based on the first cycle consistency loss value to improve the accuracy of image reconstruction performed by the image reconstruction model.

[0124] In one embodiment, the above step S220 may further specifically include: inputting a second downsampled image corresponding to the fully sampled sample image into the image reconstruction model to be trained to obtain a second fully sampled image output by the image reconstruction model to be trained; downsampling the second fully sampled image according to the downsampling template to obtain a third downsampled image; and determining a second cycle consistency loss value according to the second downsampled image and the third downsampled image.

[0125] The second downsampled image may be a downsampled image corresponding to the artifact-free medical image. The second fully sampled image may be an image obtained by reconstructing the second downsampled image. The third downsampled image may be an image obtained by downsampling the second fully sampled image.

[0126] In the specific implementation, Figure 5Taking branch A as an example, the fully sampled sample image can be downsampled to obtain a second downsampled image, and the second downsampled image can be input into the image reconstruction model to be trained for image reconstruction to obtain a second fully sampled image. After the second fully sampled image is Fourier transformed, it is downsampled through a downsampling template, and then inverse Fourier transformed to obtain a third downsampled image. The second downsampled image and the third downsampled image are respectively channel merged to obtain the second downsampled image channel merged value and the third downsampled image channel merged value. According to the difference between the two, the second cycle consistency loss value can be obtained.

[0127] In this embodiment, a second downsampled image corresponding to the fully sampled sample image is input into the image reconstruction model to be trained to obtain a second fully sampled image output by the image reconstruction model to be trained; the second fully sampled image is downsampled according to the downsampling template to obtain a third downsampled image; a second cycle consistency loss value is determined according to the second downsampled image and the third downsampled image, and the image reconstruction model can be trained according to the second cycle consistency loss value to improve the accuracy of image reconstruction performed by the image reconstruction model.

[0128] In one embodiment, the above step S220 may further specifically include: inputting the fully sampled sample image and the second fully sampled image into a preset discriminator to obtain a discriminator loss value output by the discriminator.

[0129] In a specific implementation, after obtaining the second fully sampled image, the cyclic generation model can perform channel merge on the second fully sampled image to obtain the second fully sampled image channel merged value, and input the second fully sampled image channel merged value and the full sampled sample image channel merged value into a pre-set discriminator, and obtain the discriminator loss value by calculating the mean square error between the second fully sampled image channel merged value and the full sampled sample image channel merged value.

[0130] In this embodiment, by inputting the fully sampled sample image and the second fully sampled image into a pre-set discriminator to obtain the discriminator loss value output by the discriminator, the image reconstruction model can be trained according to the discriminator loss value to further improve the accuracy of image reconstruction performed by the image reconstruction model.

[0131] In one embodiment, Figure 8 As shown, a magnetic resonance image processing method is provided. This embodiment takes the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0132] Step S310, obtaining an image to be corrected; the image to be corrected is a magnetic resonance image;

[0133] Step S320, performing downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, to obtain downsampling data sets corresponding to each sampling trajectory;

[0134] Step S330, inputting the target downsampled images corresponding to each downsampled data set into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each target downsampled image;

[0135] Step S340, weighted processing is performed on the reconstructed image to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

[0136] The target K-space data set may be all K-space data obtained by performing Fourier transform on the image to be corrected, and the downsampled data set may be all downsampled data obtained by performing downsample processing on the target K-space data set.

[0137] The target corrected image may be a magnetic resonance image after artifact correction.

[0138] In a specific implementation, the terminal can perform Fourier transform on the magnetic resonance image to be corrected to obtain a target K-space data set corresponding to the magnetic resonance image, pre-set multiple different sampling trajectories, and form mutual patterns between the multiple different sampling trajectories, that is, the patterns between the multiple different sampling trajectories are complementary, and down-sample the target K-space data set according to each sampling trajectory to obtain multiple down-sampled data sets of the target K-space data set, perform inverse Fourier transform on each down-sampled data set to obtain a target down-sampled image corresponding to each down-sampled data set, input the target down-sampled images into a pre-trained image reconstruction model, respectively, to obtain a reconstructed image corresponding to each target down-sampled image, and weight the multiple reconstructed images, for example, accumulate and average them, to obtain a target corrected image with higher image quality than the magnetic resonance image to be corrected.

[0139] In this embodiment, by acquiring an image to be corrected, downsampling processing is performed on a target K-space data set corresponding to the image to be corrected with different sampling trajectories to obtain a downsampling data set corresponding to each sampling trajectory, and the target downsampling image corresponding to each downsampling data set is input into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each target downsampling image, and weighted processing is performed on the reconstructed image to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected; the artifacts in the magnetic resonance image can be modeled as a number of lines missing in the K-space phase encoding direction, and the K-space data of the magnetic resonance image is downsampled using multiple different downsampling templates to remove the influence of the missing lines on the target downsampling image corresponding to the K-space data, and the target downsampling image is reconstructed and weighted, so that the artifacts in the magnetic resonance image can be reliably removed.

[0140] In one embodiment, different sampling trajectories form mutual patterns in the K-space along the phase encoding direction.

[0141] In a specific implementation, when downsampling the target K-space data set according to different sampling trajectories, different sampling trajectories may form mutual patterns along the K-space phase encoding direction, that is, the patterns of different sampling trajectories along the phase encoding direction are complementary.

[0142] like Figure 2 As shown, N different sampling trajectories form complementary stripes in the PE direction, so that the superposition of N different sampling trajectories has uniformity in the PE direction.

[0143] In this embodiment, by making different sampling trajectories form mutual patterns in the K space along the phase encoding direction, the influence of missing lines in the K space phase encoding direction on the magnetic resonance image can be removed, thereby removing artifacts in the magnetic resonance image.

[0144] In one embodiment, when the target K-space dataset corresponding to the image to be corrected is a three-dimensional K-space dataset, after the above step S310, it can also specifically include: performing a one-dimensional inverse Fourier transform on the original K-space dataset corresponding to the image to be corrected in the plane phase encoding direction.

[0145] In the specific implementation, the terminal performs Fourier transform on the magnetic resonance image to be corrected to obtain the original K-space data set, and performs cropping and channel compression on the original K-space data set to obtain the target K-space data set. If the magnetic resonance image to be corrected is three-dimensional data, the corresponding target K-space data set is a three-dimensional K-space data set. In this case, the original K-space data set obtained after Fourier transform can be subjected to a one-dimensional inverse Fourier transform in the SPE (S1ice Phase Encoding) direction, and then the above-mentioned cropping and channel compression processes are performed to obtain the target K-space data set.

[0146] In this embodiment, artifact correction of a three-dimensional scanning scene can be achieved by performing a one-dimensional inverse Fourier transform on the original K-space data set corresponding to the image to be corrected in the plane phase encoding direction.

[0147] In order to facilitate those skilled in the art to have a deeper understanding of the embodiments of the present application, a specific example will be described below.

[0148] The present application provides a retrospective motion artifact removal method based on a deep neural network, which can effectively remove and correct the motion artifacts in the image caused by the patient's occasional movement (such as insufficient breath holding time, occasional swinging of joints, head, etc. during scanning, etc.) in clinical practice, and takes into account both two-dimensional scanning scenes and three-dimensional scanning scenes. The method specifically includes the following contents:

[0149] 1. Cyclic generative model training

[0150] Construction of training data set: Taking the construction of head training data set as an example, according to the commonly used clinical head MRI scanning protocol, collect no less than 3000 layers of motion artifact-free images with different contrasts (T2, T1, PD) and different scanning orientations (sagittal, coronal, and transverse), turn off the K-space zero-filling interpolation operation, and ensure that the reconstructed image K-space is fully sampled.

[0151] Training data preprocessing: The matrix size of the acquired original K-space data is rounded down to a multiple of 16 and cropped. The cropped K-space data is channel-compressed to 12 channels, and then inverse Fourier transformed to the image domain. The real and imaginary parts are respectively taken in the channel dimension and connected to 24 channels. The images are stored in layers to obtain a fully sampled sample image. The size of the fully sampled sample image is (floor(x / 16)*16, floor(y / 16)*16, 24). It should be noted that for three-dimensional scanning scenes, the SPE direction can be subjected to a one-dimensional inverse Fourier transform and the above operation can be repeated.

[0152] Cycle generation model structure: The cycle generation model consists of Figure 6 The discriminator shown is Figure 7 The generator structure shown.

[0153] Cyclic generative model training process: Cyclic generative model training process is as follows Figure 5 As shown. Among them, the channel merging formula can be

[0154]

[0155] Among them, sqrt represents square root operation, sum represents sum operation, image represents image pixel, real represents the real part of image pixel, imag represents the imaginary part of image pixel, L cycle The cycle consistency loss is L1 loss, L patchGAN The discriminator loss is MSE (mean square error) loss.

[0156] The downsampling template has the same size as the input image, and uses the time point multiplication of the input image K space for downsampling. The template is set to 1 to retain the value of the corresponding position in the K space, and the template is set to 0 to set the value of the corresponding position in the K space to 0. Figure 5As shown, 36 lines in the center of the down-sampling template are set to 1, and the number and positions of the remaining lines set to 1 should conform to the Gaussian distribution.

[0157] 2. Application of Cycle Generation Model

[0158] Applications of cycle generation models include Figure 3 As shown in the figure, after the training of the cyclic generation model is completed, motion artifacts can be removed. A multi-channel image with motion artifacts is acquired from the magnetic resonance machine, and the matrix size of its original K-space data is rounded down to a multiple of 16 and cropped. The cropped K-space data is channel compressed to 12 channels to obtain the target K-space data. The target K-space data is downsampled in the PE direction using a downsampling template, and the target downsampled image is obtained after inverse Fourier transform. Assuming the number of downsampling templates is N, the number of target downsampled images is N, and N reconstructed images are obtained after entering the generator respectively. The corresponding pixel values ​​of the N reconstructed images are accumulated and averaged to obtain the final artifact-corrected image without motion artifacts.

[0159] It should be noted that the above-mentioned cycle generation model can be implemented by, but is not limited to, networks such as cycleGAN, ResNet or DenseNet. In addition, the number of lines retained in the center of the downsampling template, the number of network input channels, the loss function and the composition of the training data set are not limited to those described in the article.

[0160] Fig. 9 A comparison chart of artifact correction of abdominal MRI images is provided. Fig.10 A comparison chart of artifact correction in brain magnetic resonance images is provided. Fig. 9 and Fig.10 ,The above retrospective motion artifact removal method based on deep neural ,network has a good artifact correction effect.

[0161] The above-mentioned retrospective motion artifact removal method based on deep neural network models motion artifacts as a small number of destroyed phase encoding lines in Fourier space, which is more in line with the physical laws of magnetic resonance acquisition. In the training stage of the cyclic generation model, unsupervised training is adopted to model the motion artifact removal problem as the reconstruction problem of the undersampled Fourier space obtained after removing the phase encoding lines affected by motion. There is no need to collect motion artifact data for network training, avoiding the problem that the artificially simulated data pairs with motion artifacts and without motion artifacts that may appear in supervised learning do not match the actual clinical motion situation.

[0162] Moreover, this method uses multiple different downsampling templates that conform to a certain distribution law and are complementary to each other to downsample the Fourier space of the image with motion artifacts, which can ensure that each phase encoding line except the central limited range is undersampled, that is, it ensures that the phase encoding lines affected by motion artifacts will be removed. The images downsampled by different undersampling templates are reconstructed through the network, and the multiple images obtained are fused into one image, which can dilute the motion artifacts in the original image with a higher probability, and at the same time make up for the problem of single image quality loss caused by undersampling of a single template.

[0163] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0164] Based on the same inventive concept, the embodiment of the present application also provides an artifact correction and magnetic resonance image processing device for implementing the artifact correction and magnetic resonance image processing method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more artifact correction and magnetic resonance image processing device embodiments provided below can refer to the limitations of the artifact correction and magnetic resonance image processing method above, and will not be repeated here.

[0165] In one embodiment, Fig.11 As shown, an artifact correction device 400 is provided, comprising: a first acquisition module 410, a first downsampling module 420, a first reconstruction module 430 and a first correction module 440, wherein:

[0166] A first acquisition module 410 is used to acquire an image to be corrected containing artifacts;

[0167] A first downsampling module 420 is used to perform downsampling processing on the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template;

[0168] A first reconstruction module 430, configured to input each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0169] The first correction module 440 is used to perform weighted processing on the reconstructed image to obtain an artifact-corrected image of the image to be corrected.

[0170] In one embodiment, the downsampling template includes a first area and a second area, the first area corresponds to the low-frequency information of the image to be corrected in the phase encoding direction, the second area corresponds to the high-frequency information of the image to be corrected in the phase encoding direction, the sampling channel of the first area is full-pass, and the sampling channel of the second area is Gaussian distributed.

[0171] In one embodiment, the first downsampling module 420 is further used to determine the target K space data corresponding to the image to be corrected; according to each downsampling template, downsampling processing is performed on the phase encoding direction of the target K space data to obtain downsampling data corresponding to the target K space data; and inverse Fourier transform is performed on each downsampling data to obtain the target downsampling image.

[0172] In one embodiment, the first downsampling module 420 is further used to perform Fourier transform on the image to be corrected to obtain original K space data corresponding to the image to be corrected; perform cropping processing on the original K space data to obtain cropped K space data; and perform channel compression processing on the cropped K space data to obtain the target K space data.

[0173] In one embodiment, Fig.12 As shown, a magnetic resonance image processing device 500 is provided, comprising: a second acquisition module 510, a second downsampling module 520, a second reconstruction module 530 and a second correction module 540, wherein:

[0174] A second acquisition module 510 is used to acquire an image to be corrected; the image to be corrected is a magnetic resonance image;

[0175] A second downsampling module 520 is used to perform downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, so as to obtain downsampling data sets corresponding to each sampling trajectory;

[0176] A second reconstruction module 530, configured to input the target downsampled images corresponding to the downsampled data sets into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images;

[0177] The second correction module 540 is used to perform weighted processing on the reconstructed image to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

[0178] In one embodiment, different sampling trajectories form mutual patterns in the K-space along the phase encoding direction.

[0179] In one embodiment, when the target K-space dataset corresponding to the image to be corrected is a three-dimensional K-space dataset, the magnetic resonance image processing apparatus 500 further includes:

[0180] The one-dimensional inverse Fourier transform module is used to perform a one-dimensional inverse Fourier transform on the original K-space data set corresponding to the image to be corrected in the plane phase encoding direction.

[0181] Each module in the above-mentioned artifact correction and magnetic resonance image processing device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0182] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.13 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for artifact correction and magnetic resonance image processing is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0183] Those skilled in the art will understand that Fig.13The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0184] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0185] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0186] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0187] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0188] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0189] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0190] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for artifact correction, characterized in that: The method comprises: Acquire an image to be corrected that contains artifacts; Downsampling the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template; Inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images; The reconstructed image is weighted to obtain an artifact-corrected image of the image to be corrected.

2. The method according to claim 1, characterized in that The downsampling template includes a first area and a second area, the first area corresponds to the low-frequency information of the image to be corrected in the phase encoding direction, the second area corresponds to the high-frequency information of the image to be corrected in the phase encoding direction, the sampling channel of the first area is full-pass, and the sampling channel of the second area is Gaussian distributed.

3. The method according to claim 2, characterized in that The downsampling process is performed on the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template, including: Determining target K-space data corresponding to the image to be corrected; According to each of the downsampling templates, downsampling processing is performed on the phase encoding direction of the target K-space data respectively to obtain downsampling data corresponding to the target K-space data; Performing inverse Fourier transform on each of the downsampled data to obtain the target downsampled image.

4. The method according to claim 3, characterized in that The determining of the target K-space data corresponding to the image to be corrected includes: Performing Fourier transform on the image to be corrected to obtain original K-space data corresponding to the image to be corrected; Performing cropping processing on the original K-space data to obtain cropped K-space data; Channel compression processing is performed on the cropped K-space data to obtain the target K-space data.

5. A magnetic resonance image processing method, characterized in that: The method comprises: Acquire an image to be corrected; the image to be corrected is a magnetic resonance image; Performing downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, to obtain downsampling data sets corresponding to each sampling trajectory; Inputting the target downsampled images corresponding to each of the downsampled data sets into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images; The reconstructed image is weightedly processed to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

6. The method according to claim 5, characterized in that The different sampling trajectories form mutual patterns along the phase encoding direction in K space.

7. The method according to claim 5, characterized in that In the case where the target K-space dataset corresponding to the image to be corrected is a three-dimensional K-space dataset, after acquiring the image to be corrected, the method further includes: A one-dimensional inverse Fourier transform is performed on the original K-space data set corresponding to the image to be corrected in the plane phase encoding direction.

8. An artifact correction device, characterized in that: The device comprises: A first acquisition module, used for acquiring an image to be corrected containing artifacts; A first downsampling module is used to perform downsampling processing on the image to be corrected according to different downsampling templates to obtain a target downsampling image corresponding to each downsampling template; A first reconstruction module, used for inputting each of the target downsampled images into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images; The first correction module is used to perform weighted processing on the reconstructed image to obtain an artifact-corrected image of the image to be corrected.

9. A magnetic resonance image processing device, characterized in that: The device comprises: A second acquisition module is used to acquire an image to be corrected; the image to be corrected is a magnetic resonance image; A second downsampling module is used to perform downsampling processing of different sampling trajectories on the target K-space data set corresponding to the image to be corrected, so as to obtain downsampling data sets corresponding to each sampling trajectory; A second reconstruction module, used for inputting the target downsampled images corresponding to each of the downsampled data sets into a pre-trained image reconstruction model to obtain a reconstructed image corresponding to each of the target downsampled images; The second correction module is used to perform weighted processing on the reconstructed image to obtain a target corrected image; the image quality of the target corrected image is higher than the image quality of the image to be corrected.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images