Model training method, image artifact correction method, device, equipment and medium

Abstract

The invention provides a model training method and device, an image artifact correction method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of image processing, and the model training method comprises the steps: carrying out the registration of each first scanning image in a first time sequence image group according to a second time sequence image group, for each first scanning image, performing artifact correction on the first scanning image according to the target motion correction information corresponding to the first scanning image to obtain a third time sequence image group, and performing artifact correction on the first scanning image according to the target motion correction information corresponding to the first scanning image to obtain a second time sequence image group; and determining an image difference between the third time sequence image group and the second time sequence image group according to a target loss function formed by the normalized cross-correlation loss function and the gradient function, and training an image processing model according to the image difference. According to the method, the non-rigid motion artifacts in the CT scanning image can be efficiently and accurately corrected, the image quality is improved, and the accuracy of a clinical diagnosis result is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to a model training method, an image artifact correction method, a device, an electronic device, and a computer-readable storage medium in the field of image processing technology. Background Art

[0002] Computed Tomography (CT), as an important medical imaging technology, has been widely used in clinical diagnosis, especially playing a key role in the imaging of dynamic organs such as the heart, lungs, and abdomen. However, the quality of CT scan images is often significantly affected by motion artifacts. Motion artifacts are mainly divided into rigid motion (such as translation and rotation in head CT) and non-rigid motion (such as organ deformation caused by heart beating and breathing motion). Due to its complexity, non-rigid motion artifacts often manifest as blurred organ contours and distorted structures, seriously affecting the recognition and diagnosis of lesions. Therefore, how to solve the influence of non-rigid motion artifacts on the quality of CT scan images has become an urgent problem to be solved. Summary of the Invention

[0003] Embodiments of this application provide a model training method, an image artifact correction method, a device, an electronic device, and a computer-readable storage medium. The embodiments of this application can efficiently and accurately correct non-rigid motion artifacts in CT scan images, while retaining the details and edge information of the images, achieving an improvement in the quality of CT scan images, which is beneficial to improving the accuracy of diagnostic results and enhancing the clinical diagnostic effect.

[0004] In a first aspect, a model training method is provided. The model training method includes: inputting a first temporal image group and a second temporal image group of a first scanning object into an image processing model; wherein, the first temporal image group includes a plurality of first scanning images with motion artifacts, the scanning angles corresponding to the plurality of first scanning images are continuous and non-repeating, the second temporal image group includes a plurality of second scanning images without motion artifacts, the scanning angles corresponding to the plurality of second scanning images are continuous and non-repeating, the first scanning images in the first temporal image group and the second scanning images in the second temporal image group are in one-to-one correspondence and have the same scanning angle; registering each first scanning image in the first temporal image group according to the second temporal image group to obtain target motion correction information corresponding to each of the plurality of first scanning images; for each first scanning image, performing artifact correction on the first scanning image according to the target motion correction information corresponding to the first scanning image to obtain a third temporal image group; determining the image difference between the third temporal image group and the second temporal image group according to the target loss function, and training the image processing model according to the image difference; wherein, the target loss function includes a first loss function and a second loss function, the first loss function is used to measure the global similarity between the second temporal image group and the third temporal image group, and the second loss function is used to measure the difference in edge information and detail information between the scanning images in the third temporal image group and the second scanning images.

[0005] Based on the above technical solution, in the embodiment of the present application, by registering each first scanning image in the first temporal image group according to the second temporal image group to obtain target motion correction information corresponding to each of the plurality of first scanning images, for each first scanning image, performing artifact correction on the first scanning image according to the target motion correction information corresponding to the first scanning image to obtain a third temporal image group, determining the image difference between the third temporal image group and the second temporal image group according to the target loss function composed of the normalized cross-correlation loss function and the gradient function, and training the image processing model according to the image difference. On the one hand, it realizes the end-to-end optimization of motion correction information for the temporal image group in the deep learning network, can significantly reduce the computational complexity of the artifact motion estimation process in the image, and simplifies the artifact motion estimation process. On the other hand, by designing the target loss function composed of the normalized cross-correlation loss function and the gradient function for the supervised training of the model, the obtained third temporal image group is closer to the second temporal image group, realizing the simultaneous optimization of the global similarity and local detail information of the image, which is beneficial to improving the effect of motion artifact correction in the image.

[0006] In a possible implementation, the registering each first scanned image in the first time-series image group according to the second time-series image group to obtain target motion correction information corresponding to each of the multiple first scanned images includes: for each first scanned image in the first time-series image group, extracting multi-scale features of the first scanned image; performing feature difference on the multi-scale features of the first scanned image to obtain artifact features regarding motion artifacts corresponding to the first scanned image; mapping the artifact features to a motion field regarding the motion artifacts to obtain initial motion correction information corresponding to the first scanned image; and correcting the initial motion correction information based on the artifact features to obtain the target motion correction information corresponding to each of the multiple first scanned images.

[0007] In a possible implementation, the correcting the initial motion correction information based on the artifact features to obtain the target motion correction information corresponding to each of the multiple first scanned images includes: determining a weight value corresponding to the initial motion correction information based on the artifact features; and multiplying the initial motion correction information by the weight value to obtain the target motion correction information corresponding to each of the multiple first scanned images.

[0008] In a possible implementation, the determining a weight value corresponding to the initial motion correction information based on the artifact features includes at least one of the following steps: obtaining the weight value by performing global non-linear compression and correlation mapping on the artifact features; and obtaining the weight value by performing local time-series sliding convolution and pattern extraction on the artifact features.

[0009] In a possible implementation, the performing artifact correction on the first scanned image according to the target motion correction information corresponding to the first scanned image to obtain a third time-series image group includes: extracting spatial transformation parameters from the first scanned image and the target motion correction information corresponding to the first scanned image; generating a sampling grid based on the spatial transformation parameters; and resampling the first scanned image based on the sampling grid to obtain the third time-series image group.

[0010] In a second aspect, an image artifact correction method is provided. The image artifact correction method includes: performing multi-angle reconstruction on the original scan image of a second scanning object to obtain a fourth time-series image group; wherein, the fourth time-series image group includes a plurality of fourth scan images, and the scanning angles corresponding to the plurality of fourth scan images are continuous and non-repeating; inputting the fourth time-series image group into an image processing model, and performing artifact correction on each fourth scan image in the fourth time-series image by the image processing model to output a fifth time-series image group; wherein, the image processing model is trained based on the above-mentioned model training method; and fusing the plurality of fifth scan images in the fifth time-series image group to obtain the artifact-corrected scan image of the second scanning object.

[0011] Based on the above technical solution, in the embodiment of the present application, by performing multi-angle reconstruction on the original scan image of a second scanning object to obtain a fourth time-series image group, inputting the fourth time-series image group into an image processing model trained based on the above-mentioned model training method, performing artifact correction on each fourth scan image in the fourth time-series image by the image processing model to output a fifth time-series image group, and fusing the plurality of fifth scan images in the fifth time-series image group to obtain the artifact-corrected scan image of the second scanning object, the combination of multi-angle image reconstruction and a deep learning network is realized for motion artifact processing of CT scan images, which can effectively reduce motion artifacts in CT scan images, improve image quality, and is beneficial to improving the accuracy of diagnosis results and enhancing the clinical diagnosis effect.

[0012] In a possible implementation manner, the performing multi-angle reconstruction on the original scan image of a second scanning object to obtain a fourth time-series image group includes: intercepting an image within a first angle range from the original scan image to obtain a first image to be processed; wherein, the angle range of the original scan image includes the first angle range; splitting the first image to be processed into a plurality of image blocks according to a plurality of different second angle ranges to obtain a plurality of second images to be processed; wherein, the union of the plurality of second angle ranges is the first angle range; and reconstructing the plurality of second images to be processed to obtain the fourth time-series image group.

[0013] In a possible implementation manner, the fusing the plurality of fifth scan images in the fifth time-series image group to obtain the artifact-corrected scan image of the second scanning object includes: performing pixel-level averaging on the plurality of fifth scan images to obtain the artifact-corrected scan image.

[0014] In a third aspect, a model training device is provided. The model training device includes:

[0015] An input module for inputting a first temporal image group and a second temporal image group of a first scanning object into an image processing model; wherein, the first temporal image group includes a plurality of first scanning images with motion artifacts, the scanning angles corresponding to the plurality of first scanning images are continuous and non-repeating, the second temporal image group includes a plurality of second scanning images without motion artifacts, the scanning angles corresponding to the plurality of second scanning images are continuous and non-repeating, and the first scanning images in the first temporal image group and the second scanning images in the second temporal image group are in one-to-one correspondence and have the same scanning angle;

[0016] A first processing module for registering each first scanning image in the first temporal image group according to the second temporal image group to obtain target motion correction information corresponding to each of the plurality of first scanning images;

[0017] A second processing module for, for each first scanning image, performing artifact correction on the first scanning image according to the target motion correction information corresponding to the first scanning image to obtain a third temporal image group;

[0018] A training module for determining the image difference between the third temporal image group and the second temporal image group according to the target loss function, and training the image processing model according to the image difference; wherein, the target loss function includes a first loss function and a second loss function, the first loss function is used to measure the global similarity between the second temporal image group and the third temporal image group, and the second loss function is used to measure the difference in edge information and detail information between the scanning images in the third temporal image group and the second scanning images.

[0019] In a fourth aspect, an image artifact correction device is provided, and the image artifact correction device includes:

[0020] An image reconstruction module for performing multi-angle reconstruction on the original scanning image of a second scanning object to obtain a fourth temporal image group; wherein, the fourth temporal image group includes a plurality of fourth scanning images, and the scanning angles corresponding to the plurality of fourth scanning images are continuous and non-repeating;

[0021] An image correction module for inputting the fourth temporal image group into an image processing model, and performing artifact correction on each fourth scanning image in the fourth temporal image by the image processing model to output a fifth temporal image group; wherein, the image processing model is trained based on the above model training method;

[0022] An image fusion module for fusing the plurality of fifth scanning images in the fifth temporal image group to obtain an artifact correction scanning image of the second scanning object.

[0023] Fifth aspect, there is provided an electronic device, including a memory and a processor. The memory is used for storing executable program codes, and the processor is used for calling and running the executable program codes from the memory, such that the electronic device executes the image artifact correction method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0024] Sixth aspect, there is provided a computer program product, including: computer program codes, when the computer program codes are run on a computer, such that the computer executes the above-mentioned model training method or image artifact correction method.

[0025] Seventh aspect, there is provided a computer-readable storage medium, storing computer program codes, when the computer program codes are run on a computer, such that the computer executes the above-mentioned model training method or image artifact correction method. Description of the Drawings

[0026] Figure 1 Fig. shows a schematic flowchart of a model training method provided by an embodiment of the present application;

[0027] Figure 2 Fig. shows a schematic structural diagram of an image processing model provided by an embodiment of the present application;

[0028] Figure 3 Fig. shows a schematic flowchart of an image artifact correction method provided by an embodiment of the present application

[0029] Figure 4 Fig. shows a schematic structural diagram of a model training device provided by an embodiment of the present application;

[0030] Figure 5 Fig. shows a schematic structural diagram of an image artifact correction device provided by an embodiment of the present application;

[0031] Figure 6 Fig. shows a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0032] Hereinafter, the technical solutions in the present application will be clearly and elaborately described in conjunction with the drawings. Among them, in the description of the embodiments of the present application, unless otherwise specified, " / " means "or". For example, A / B may mean A or B. The "and / or" in the text is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of the present application, "a plurality of" means two or more than two.

[0033] Hereinafter, the terms "first" and "second" are for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0034] As an important medical imaging technology, computed tomography (CT) has been widely used in clinical diagnosis, especially playing a key role in the imaging of dynamic organs such as the heart, lungs, and abdomen. However, the quality of CT scan images is often significantly affected by motion artifacts. Motion artifacts are mainly divided into rigid motion (such as translation and rotation in head CT) and non-rigid motion (such as organ deformation caused by heart beating and breathing motion). Due to its complexity, non-rigid motion artifacts often manifest as blurred organ contours and distorted structures, seriously affecting the recognition and diagnosis of lesions (such as atherosclerosis, lung nodules, tumors, etc.).

[0035] Currently, to address the problem of motion artifacts, some technical means have been proposed, such as electrocardiogram (ECG) gating technology, respiratory gating technology, and algorithm-based motion correction methods. However, these methods have obvious limitations in practical applications: ECG gating has high requirements for the heart rate stability of patients and cannot completely eliminate the influence of respiratory motion; respiratory gating depends on patient cooperation, is complex to operate and time-consuming; and existing algorithms often have large computational amounts, low efficiency, and limited correction effects when dealing with non-rigid motion.

[0036] Based on the above problems, the embodiments of the present application provide a model training method, an image artifact correction method, a device, an electronic device, and a computer-readable storage medium. The embodiments of the present application can efficiently and accurately correct non-rigid motion artifacts (hereinafter referred to as motion artifacts) in CT scan images, while retaining the details and edge information of the images, achieving an improvement in the quality of CT scan images, which is beneficial to improving the accuracy of diagnostic results and enhancing the clinical diagnostic effect.

[0037] The following is an embodiment of a model training method provided in the specification of the present application.

[0038] Figure 1 The schematic flowchart of a model training method provided by the embodiments of the present application is shown. As Figure 1 shown, the model training method provided by the embodiments of the present application is applied to an electronic device (such as a computer). The above model training method includes the following steps:

[0039] S110: Input the first temporal image group and the second temporal image group of the first scanning object into the image processing model.

[0040] In an exemplary embodiment, the image processing model belongs to a deep learning network, which is used to remove (correct) non-rigid motion artifacts in CT scan images to improve the image quality of CT scan images. Before training the image processing model, it is necessary to prepare sample image data and reference image data. The process of obtaining the sample image data and the reference image data includes: performing a complete circular scan on a first scanning object (such as a patient) by CT to obtain the original CT scan image of the first scanning object, that is, the scanning angle of the original CT scan image of the first scanning object is 360°. Among them, the original CT scan image of the first scanning object has motion artifacts.

[0041] After obtaining the original CT scan image of the first scanning object, a local image within a first angular range is intercepted from the original CT scan image of the first scanning object. The first angular range is determined according to the difference between the starting point and the ending point of intercepting the local image. If the difference between the starting point and the ending point is 180 degrees, then the first angular range can be any one of [0°, 180°], [45°, 225°], [90°, 270°], etc. Among them, the difference between the starting point and the ending point can be set according to actual needs, and the embodiments of the present application do not make any limitations.

[0042] For example, if the first angular range is [0°, 180°], a local image corresponding to [0°, 180°] is intercepted from the original CT scan image of the first scanning object to obtain a sample image to be processed. Then, based on the scanning angle from 0° to 180°, the sample image to be processed is split into N parts, where N is a positive integer greater than or equal to 2. For example, N = 10, that is, the sample image to be processed is split into 10 first sub-sample images. The target angular ranges corresponding to the 1st to 10th first sub-sample images are respectively: [0°, 18°], [18°, 36°], [36°, 54°], [54°, 72°], [72°, 90°], [90°, 108°], [108°, 126°], [126°, 144°], [144°, 162°]. After obtaining N first sub-sample images, partial angular reconstruction is performed for each target angular range, that is, based on the target angular range of each first sub-sample image, each first sub-sample image is reconstructed to obtain N reconstructed sub-sample images, and a first time-series image group is generated through the N reconstructed sub-sample images. Among them, each reconstructed sub-sample image has motion artifacts. For the convenience of distinction, the reconstructed sub-sample image is referred to as the first scan image, that is, the first time-series image group includes multiple first scan images with motion artifacts, and the scanning angles corresponding to the multiple first scan images are continuous and non-repetitive. The first time-series image group is the sample image data. The first time-series image group obtained through multi-angle reconstruction can more finely reflect the dynamic changes of motion artifacts and refine the motion process of the artifacts, facilitating more accurate capture of the motion characteristics of the artifacts in the subsequent process, thereby providing more accurate input data for subsequent motion artifact correction.

[0043] After obtaining the sample image to be processed, other artifact removal processing methods are used to perform artifact removal processing on the sample image to be processed to obtain an artifact-free sample image. Then, in the same way as generating the first time-series image group based on the sample image to be processed, the artifact-free sample image is processed to obtain a second time-series image group. The second time-series image group includes multiple second scan images without motion artifacts, and the scanning angles corresponding to the multiple second scan images are continuous and non-repetitive. The second time-series image group is the reference image data. In addition, a historical CT scan image with good quality and few motion artifacts from the previous scan of the first scanning object can be obtained. The scanning angle range of the historical CT scan image is the first angular range. Then, in the same way as generating the first time-series image group based on the sample image to be processed, the historical CT scan image is processed to obtain a second time-series image group.

[0044] The images in the first temporal image group are aligned with those in the second temporal image group, that is, the first scanned images in the first temporal image group correspond one by one to the second scanned images in the second temporal image group and have the same scanning angles. For example, the 1st first scanned image in the first temporal image group corresponds to the 1st second scanned image in the second temporal image group, and the target angle range of the 1st first scanned image in the first temporal image group is [0°, 18°], and the target angle range of the 1st second scanned image in the second temporal image group is also [0°, 18°], and so on. Among them, the resolution of the first scanned images is the same as that of the second scanned images.

[0045] After obtaining the first temporal image group and the second temporal image group of the first scanned object, input the first temporal image group and the second temporal image group into the image processing model to start training the image processing model.

[0046] S120: Register each first scanned image in the first temporal image group according to the second temporal image group to obtain the target motion correction information corresponding to each of the multiple first scanned images.

[0047] After inputting the first temporal image group and the second temporal image group into the image processing model, use the second temporal image group as a reference to register each first scanned image Fi in the first temporal image group. After the registration is completed, obtain the initial motion correction information corresponding to the first scanned image Fi, and then correct (such as weight adjustment) the initial motion correction information corresponding to the first scanned image Fi to obtain the target motion correction information corresponding to the first scanned image Fi, so as to obtain the target motion correction information corresponding to each of the N first scanned images. Among them, the role of the target motion correction information is to guide the artifact correction operation, that is, how to adjust the first scanned image with motion artifacts to a state without motion artifacts.

[0048] S130: For each first scanned image, perform artifact correction on the first scanned image according to the target motion correction information corresponding to the first scanned image to obtain the third temporal image group.

[0049] After obtaining the target motion correction information corresponding to each of the N first scanned images, the first scanned image Fi is subjected to artifact correction according to the first scanned image Fi and the corresponding target motion correction information (which can also be understood as motion compensation for the artifacts in the first scanned image Fi), and the first scanned image Fi after artifact correction is obtained, so as to obtain a plurality of first scanned images after artifact correction. A third temporal image group (which can also be called a target temporal image group) is generated through the plurality of first scanned images after artifact correction. The scanned images in the third temporal image group are aligned with the scanned images in the first temporal image group. For the sake of easy distinction, the scanned images in the third temporal image group are called third scanned images, that is, the third scanned images are the first scanned images after artifact correction. Among them, performing artifact correction on the first scanned images in the first temporal image group is to make the image quality of the third scanned images in the obtained third temporal image group as close as possible to the image quality of the second scanned images in the second temporal image group. The resolution of the third scanned images is the same as that of the first scanned images.

[0050] S140: Determine the image difference between the third temporal image group and the second temporal image group according to the target loss function, and train the image processing model according to the image difference.

[0051] The target loss function includes a first loss function (also called a normalized cross-correlation loss function) and a second loss function (also called a gradient loss function). The first loss function is used to measure the global similarity between the second temporal image group and the third temporal image group, and the second loss function is used to measure the difference in edge information and detail information between the third scanned images in the third temporal image group and the second scanned images. By generating the target loss function through the above two loss functions, the global similarity and local detail information of the image can be optimized simultaneously, thereby improving the effect of motion artifact correction.

[0052] The first loss function L 1 (I c ,I r ) is shown by formula (1). The second loss function L 2 (I c ,I r ) is shown by formula (2). The formula of the target loss function L is shown by formula (3):

[0053]

[0054] L = L 1 (I c ,I r ) + L 2 (I c ,I r )(3)

[0055] Among them, Ic (i,j) and I r (i,j) respectively represent the pixel values of the third scanned image and the second scanned image at the pixel point (i,j), μ c and μ r respectively represent the pixel means of the third scanned image and the second scanned image. The minus sign "-" in formula (1) indicates that the loss function needs to be minimized, that is, to maximize the normalized cross-correlation value. and respectively represent the horizontal gradient and the vertical gradient of the image at (i,j) (calculated by the Sobel operator or other gradient operators). Only the x and y dimensions need to be calculated in the 2D network calculation. If it is a 3D network, the x, y, and z dimensions need to be calculated.

[0056] After obtaining the third time-series image group and the second time-series image group, input the third time-series image group and the second time-series image group into the target loss function to obtain a loss result, that is, the image difference between the third time-series image group and the second time-series image group. Then, based on this image difference, perform backpropagation on the image processing model to implement iterative training of the image processing model until the image processing model converges, and then stop the training of the image processing model. After stopping the training of the image processing model, save the model parameters of the image processing model, that is, the image processing model can be officially applied to the artifact removal processing of other CT scanned images.

[0057] In the embodiment of the present application, by registering each first scanned image in the first time-series image group according to the second time-series image group to obtain the target motion correction information corresponding to each first scanned image, for each first scanned image, performing artifact correction on the first scanned image according to the target motion correction information corresponding to the first scanned image to obtain the third time-series image group, determining the image difference between the third time-series image group and the second time-series image group according to the target loss function composed of the normalized cross-correlation loss function and the gradient function, and training the image processing model according to the image difference. On the one hand, it realizes end-to-end optimization of motion correction information for time-series image groups in a deep learning network, can significantly reduce the computational complexity of the artifact motion estimation process, and simplifies the artifact motion estimation process. On the other hand, by designing a target loss function composed of a normalized cross-correlation loss function and a gradient function for supervised training of the model, the obtained third time-series image group is closer to the second time-series image group, realizing simultaneous optimization of the global similarity and local detail information of the image, which is beneficial to improving the effect of motion artifact correction in the image.

[0058] The image processing model includes a registration module, a weight adjustment module, a fusion module, and a spatial transformation module. The weight adjustment module includes, but is not limited to, a Fully Convolutional Network (FCN), a 1D Convolutional Neural Network (1DCNN), etc. The registration module includes an encoder and a decoder. The output end of the encoder is respectively connected to the input end of the decoder and the input end of the weight adjustment module. The output ends of the decoder and the weight adjustment module are both connected to the input end of the fusion module. The output end of the fusion module is connected to the input end of the spatial transformation module.

[0059] Exemplarily, for example, the encoder includes m convolutional layers, where m is a positive integer greater than or equal to 1. The m convolutional layers are connected in series in sequence. The decoder includes a residual network and m deconvolutional layers, and the m deconvolutional layers are connected in series in sequence. Among them, the input end of the first convolutional layer is the input end of the image processing model (i.e., the input end of the encoder). The output end of the last convolutional layer (the m-th convolutional layer) (i.e., the output end of the encoder) is connected to the input end of the residual network and the input end of the weight adjustment module. The output end of the residual network is connected to the input end of the first deconvolutional layer. The output end of the last convolutional layer (the m-th deconvolutional layer) is connected to the input end of the fusion module. The output end of the first convolutional layer is connected to the input end of the first deconvolutional layer. The output end of the second convolutional layer is connected to the input end of the second deconvolutional layer, and so on. The output end of the m-th convolutional layer is connected to the input end of the m-th deconvolutional layer. The spatial transformation module includes a local network layer, a grid generator, and a resampler. The local network layer, the grid generator, and the resampler are connected in series in sequence. The output end of the fusion module is connected to the input end of the local network layer. The output end of the resampler is the output end of the image processing model. As Figure 2 shown, Figure 2 FIG. shows a schematic structural diagram of the image processing model provided by the embodiment of the present application, where m = 3.

[0060] It should be noted that the network layers constituting the encoder and decoder in the embodiment of the present application are not limited to the above description. The embodiment of the present application can also use other network layers to construct the encoder and decoder, and specific details are not limited.

[0061] In a possible implementation manner, the above-mentioned registering each first scanned image in the first time-series image group according to the second time-series image group to obtain the target motion correction information corresponding to each first scanned image includes the following steps:

[0062] For each first scanned image in the first time-series image group, extract the multi-scale features of the first scanned image;

[0063] Perform feature difference on the multi-scale features of the first scanned image to obtain the artifact features regarding motion artifacts corresponding to the first scanned image;

[0064] Map the artifact features to a motion field regarding motion artifacts to obtain the initial motion correction information corresponding to the first scanned image;

[0065] Based on the artifact features, correct the initial motion correction information to obtain the target motion correction information corresponding to each of the multiple first scanned images.

[0066] As Figure 1 and Figure 2 shown, input the first temporal image group and the second temporal image group into the encoder. The encoder extracts the multi-scale features of the first scanned image Fi in the first temporal image group, and then performs feature difference processing on the multi-scale features of the first scanned image Fi to obtain the artifact features regarding motion artifacts corresponding to the first scanned image Fi, that is, the artifact features of the motion artifacts in the first scanned image Fi. The encoder outputs the artifact features corresponding to the first scanned image Fi, and the artifact features corresponding to the first scanned image Fi are input into the decoder. Specifically: Input the first scanned image Fi and the second scanned image Ei into the first convolutional layer, and then after being processed by m convolutional layers, each of the m convolutional layers outputs a convolution result. The m-th convolutional layer outputs the artifact features corresponding to the first scanned image Fi (that is, the convolution result output by the m-th convolutional layer). Furthermore, the artifact features corresponding to the first scanned image Fi are input into the residual network. The convolution result of the first convolutional layer is input into the first deconvolutional layer, the convolution result of the second convolutional layer is input into the second deconvolutional layer, and so on. The convolution result of the m-th convolutional layer is input into the m-th deconvolutional layer. Among them, through the convolution operation of m convolutional layers, the multi-scale features of the input image can be extracted, that is, each convolutional layer will learn different-scale features, including from low-level features (such as edges, textures) to high-level features (such as object shapes, structures). The artifact features can accurately describe the characteristics and distributions of the motion artifacts in the first scanned image Fi. The multi-scale features include local details (such as edges, textures, etc.) and global structure information (such as shapes, positional relationships, etc.).

[0067] After the artifact features corresponding to the first scanned image Fi are input into the decoder, the decoder maps the artifact features corresponding to the first scanned image Fi into a motion field regarding motion artifacts, obtaining the initial motion correction information regarding motion artifacts corresponding to the first scanned image Fi. Specifically: after the artifact features corresponding to the first scanned image Fi are input into the residual network and the convolution results of the first convolutional layer to the m-th convolutional layer are respectively input into their corresponding transposed convolutional layers, the m-th transposed convolutional layer outputs the initial motion correction information regarding motion artifacts corresponding to the first scanned image Fi; among them, by inputting the output result of each convolutional layer into the transposed convolutional layer connected to each tape layer, multi-scale feature information can be retained, information loss can be reduced, the context understanding ability can be improved, gradient propagation can be improved, and the robustness of the model can be enhanced; the motion field of motion artifacts can be represented as a displacement vector field or a deformation matrix, which describes the motion state of the pixel points of the motion artifacts.

[0068] Considering the influence of factors such as the estimation error of the network and data noise, the obtained initial motion correction information may not be accurate enough. Therefore, after obtaining the initial motion correction information, the initial motion correction information corresponding to the first scanned image Fi is corrected based on the artifact features corresponding to the first scanned image Fi, obtaining the target motion correction information corresponding to the first scanned image Fi. In this way, the target motion correction information corresponding to each of the N first scanned images can be obtained. By correcting the initial motion correction information corresponding to each first scanned image, the accuracy of the motion correction information can be improved, making the quality of the scanned image obtained after artifact correction higher and the motion artifacts less.

[0069] In a possible implementation manner, the above-mentioned correction of the initial motion correction information based on the artifact features to obtain the target motion correction information corresponding to each of the multiple first scanned images includes the following steps:

[0070] Determine the weight value corresponding to the initial motion correction information based on the artifact features;

[0071] Multiply the initial motion correction information by the weight value to obtain the target motion correction information corresponding to each of the multiple first scanned images.

[0072] Such as Figure 1 and Figure 2As shown, after obtaining the artifact features corresponding to the first scanned image Fi, that is, the artifact features at the scanning angle where the first scanned image Fi is located, then based on the artifact features corresponding to the first scanned image Fi, input them into the weight adjustment module. The weight adjustment module assigns weight values to the initial motion correction information corresponding to the first scanned image Fi based on the artifact features corresponding to the first scanned image Fi. Specifically, the weight adjustment module will analyze the spatial distribution, intensity, and correlation with other features of the artifact features to determine which features are more important for the final motion correction information. For example, for areas with more significant motion artifacts, higher weight values may be assigned; while for areas with less significant motion artifacts, lower weight values may be assigned. The weight values are represented in the form of vectors, and the weight values are used to represent the degree of importance of the features.

[0073] After assigning weight values to the artifact features corresponding to the first scanned image Fi, input the weight values and the initial motion correction information corresponding to the first scanned image Fi into the fusion module. The fusion module performs a weighted fusion operation, that is, calculates the product of the initial motion correction information corresponding to the first scanned image Fi and the weight values to obtain the target motion correction information corresponding to the first scanned image Fi, which realizes the correction of the initial motion correction information corresponding to the first scanned image Fi, and thus can obtain the target motion correction information corresponding to each of the N first scanned images.

[0074] By using the weight adjustment module to assign weight values to the artifact features of each first scanned image, the weights of the reconstructed images at different angles are dynamically adjusted, which can effectively highlight important features and suppress irrelevant features; by multiplying the initial motion correction information by the weight values in the fusion module, accurate target motion correction information can be generated, which is beneficial to promoting the model to more accurately and stably complete the motion artifact correction task, thereby improving the image quality.

[0075] In a possible implementation manner, the above-mentioned determining the weight values corresponding to the initial motion correction information based on the artifact features includes at least one of the following steps:

[0076] Obtain the weight values by performing global non-linear compression and correlation mapping on the artifact features;

[0077] Obtain the weight values by performing local temporal sliding convolution and pattern extraction on the artifact features.

[0078] For the case where the weight adjustment module includes a fully connected network, perform global non-linear compression and correlation mapping on the artifact features through the fully connected network to obtain the weight values. Specifically, flatten the artifact features into a global vector, and use the fully connected network to model the non-linear relationship between the artifact intensity and the weights (such as the more severe the motion artifact, the lower the weight) to obtain the weight values.

[0079] For the case where the weight adjustment module includes a one-dimensional convolutional network, local temporal sliding convolution and pattern extraction are performed on the artifact features through the one-dimensional convolutional network to obtain weight values. Specifically, the one-dimensional convolutional network slides the convolution kernel along the time axis based on the artifact features to extract the local temporal patterns of the artifacts (such as periodic breathing or heartbeat artifacts), and outputs a weight curve that matches the motion period of the artifacts to obtain the weight values.

[0080] The weight adjustment module composed of a fully connected network and the weight adjustment module composed of a one-dimensional convolutional network assign weight values to the initial motion correction information corresponding to each first scanned image, improving the flexibility of weight value calculation.

[0081] In a possible implementation manner, the above-mentioned artifact correction of the first scanned image according to the target motion correction information corresponding to the first scanned image to obtain the third temporal image group includes the following steps:

[0082] Extract the spatial transformation parameters from the first scanned image and the target motion correction information corresponding to the first scanned image;

[0083] Generate an adoption grid based on the spatial transformation parameters;

[0084] Resample the first scanned image based on the adoption grid to obtain the third temporal image group.

[0085] As Figure 1 and Figure 2 shown, the local network layer is a small neural network used to learn and extract spatial transformation parameters from the first scanned image Fi and the target motion correction information corresponding to the first scanned image Fi. The spatial transformation parameters can describe how the pixel points in the first scanned image Fi are aligned to the artifact-free state through geometric transformation. For example, the local network layer includes a convolutional layer, a fully connected layer or a regression layer. The convolutional layer is used to extract the features of the first scanned image Fi and the target motion correction information corresponding to the first scanned image Fi, and generate spatial transformation parameters through the fully connected layer or the regression layer.

[0086] The grid generator generates a sampling grid according to the spatial transformation parameters, and the sampling grid defines how to map the pixel points in the first scanned image Fi to new spatial positions. The resampler resamples the first scanned image Fi based on the sampling grid (which can also be understood as spatial transformation), that is, recalculates the pixel values of the first scanned image Fi according to the sampling grid. For example, interpolation methods (such as bilinear interpolation or cubic interpolation) are used to calculate the pixel values of each point in the sampling grid, generating the first scanned image Fi after spatial transformation. In this way, N first scanned images after spatial transformation are obtained, and a third temporal image group is generated from the N first scanned images after spatial transformation. That is, the third scanned image in the third temporal image group has removed motion artifacts and is aligned with the second scanned image in the second temporal image group.

[0087] The above model training method adopted in the embodiments of the present application is used for the training of an image processing model, and has the following advantages:

[0088] 1. By optimizing the motion correction information end-to-end for the temporal image group in the deep learning network, the computational complexity of the motion artifact estimation process in the image can be significantly reduced, the motion artifact estimation process is simplified, and it is beneficial to shorten the time for removing motion artifacts from the image.

[0089] 2. By designing an objective loss function composed of a normalized cross-correlation loss function and a gradient function for the supervised training of the model, the obtained third temporal image group is closer to the second temporal image group, realizing the simultaneous optimization of the global similarity and local detail information of the image, which is beneficial to improving the effect of motion artifact correction in the image.

[0090] 3. A weight adjustment module is additionally added to the model. During the model training process, the weight adjustment module realizes the dynamic adjustment of the weights of the reconstructed images at different angles, can more accurately capture the difference information of the artifacts in the temporal motion, thereby further optimizing the motion correction information output by the model, can significantly improve the accuracy of the motion correction information, and is beneficial to promoting the model to more accurately and stably complete the motion artifact correction task, thereby improving the image quality.

[0091] The following is an embodiment of a model training method provided in the specification of the present application.

[0092] Figure 3 The schematic flowchart of an image artifact correction method provided in the embodiments of the present application is shown. As Figure 3 shown, the image artifact correction method provided in the embodiments of the present application is applied to an electronic device (such as a computer). The above image artifact correction method includes the following steps:

[0093] S210: Reconstruct the original scan image of the second scan object from multiple angles to obtain a fourth sequence of images.

[0094] S220: Input the fourth sequence of images into an image processing model, and let the image processing model correct the artifacts in each fourth scan image in the fourth sequence of images, and output a fifth sequence of images;

[0095] S230: Fuse multiple fifth scan images in the fifth sequence of images to obtain an artifact-corrected scan image of the second scan object.

[0096] In an exemplary embodiment, any patient can be selected as the second scan object. The second scan object and the above-mentioned first scan object can be the same patient or different patients. The original scan image of the second scan object refers to the original CT scan image of the second scan object, and the scan angle of the original scan image of the second scan object is 360°. After obtaining the original scan image of the second scan object, reconstruct the original scan image of the second scan object from multiple angles to obtain a fourth sequence of images. Among them, the fourth sequence of images includes multiple fourth scan images, the scan angles corresponding to the multiple fourth scan images are continuous and non-repetitive, and the fourth scan images include fourth scan images with motion artifacts.

[0097] The image processing model is trained based on the above model training method. After obtaining the fourth sequence of images, input the fourth sequence of images into the trained image processing model. The image processing model corrects the motion artifacts in each fourth scan image in the fourth sequence of images and outputs a fifth sequence of images. The fifth sequence of images includes multiple fifth scan images without motion artifacts, the scan angles corresponding to the multiple fifth scan images are continuous and non-repetitive, and the fourth scan images in the fourth sequence of images and the fifth scan images in the fifth sequence of images correspond one by one and have the same scan angle. The resolution of the fifth scan images is the same as that of the fourth scan images.

[0098] Since the fifth scan images in the fifth sequence of images are all independent images, after obtaining the fifth sequence of images, fuse the multiple fifth scan images in the fifth sequence of images to realize fusing the multiple fifth scan images in the fifth sequence of images into a complete image, that is, obtain an artifact-corrected scan image of the second scan object. The artifact-corrected scan image has no motion artifacts compared with the original scan image of the second scan object, and diagnosticians can diagnose the condition of the second scan object based on the artifact-corrected scan image of the second scan object.

[0099] In an embodiment of the present application, by performing multi-angle reconstruction on the original scan image of the second scan object to obtain a fourth time-series image group, inputting the fourth time-series image group into an image processing model trained based on the above model training method, performing artifact correction on each fourth scan image in the fourth time-series image by the image processing model, outputting a fifth time-series image group, and fusing multiple fifth scan images in the fifth time-series image group to obtain an artifact-corrected scan image of the second scan object, the technical solution realizes the combination of multi-angle image reconstruction and a deep learning network for motion artifact processing in CT scan images, can effectively reduce motion artifacts in CT scan images, improve image quality, and is beneficial to improving the accuracy of diagnostic results and enhancing the clinical diagnostic effect.

[0100] In a possible implementation manner, the above-mentioned multi-angle reconstruction of the original scan image of the second scan object to obtain a fourth time-series image group includes the following steps:

[0101] Intercept an image within a first angle range from the original scan image to obtain a first image to be processed;

[0102] Split the first image to be processed into multiple image blocks according to multiple different second angle ranges to obtain multiple second images to be processed; wherein, the union of the multiple second angle ranges is the first angle range;

[0103] Reconstruct the multiple second images to be processed to obtain a fourth time-series image group.

[0104] The multi-angle reconstruction of the original scan image of the second scan object is specifically a partial angle reconstruction at multiple angles of the original scan image of the second scan object. For example, the scan angle of the original scan image of the second scan object is 360°, that is, the angle range of the original scan image of the second scan object is [0°, 360°]. Intercept a local image within a first angle range from the original scan image of the second scan object to obtain a first image to be processed, and the angle range of the original scan image includes the first angle range.

[0105] The first angle range is determined according to the difference between the starting point and the ending point of intercepting the local image. If the difference between the starting point and the ending point is 180 degrees, then the first angle range can be any one of [0°, 180°], [45°, 225°], [90°, 270°], etc. For example, if the first angle range is [0°, 180°], then intercept the local image corresponding to [0°, 180°] from the original scan image of the second scan object to obtain a first image to be processed.

[0106] The union of multiple second angular ranges is the first angular range. For example, the multiple second angular ranges are respectively: [0°, 18°], [18°, 36°], [36°, 54°], [54°, 72°], [72°, 90°], [90°, 108°], [108°, 126°], [126°, 144°], [144°, 162°]. After obtaining the first image to be processed, the first image to be processed is split into multiple image blocks according to multiple different second angular ranges to obtain multiple second images to be processed. For example, 10 second images to be processed are obtained, and the second angular ranges corresponding to the 10 second images to be processed are respectively: [0°, 18°], [18°, 36°], [36°, 54°], [54°, 72°], [72°, 90°], [90°, 108°], [108°, 126°], [126°, 144°], [144°, 162°].

[0107] After performing image reconstruction on multiple second images to be processed, multiple second images to be processed after reconstruction are obtained. The second images to be processed after reconstruction are called fourth scanned images, that is, a fourth temporal image group is generated through multiple fourth scanned images.

[0108] By performing multi-angle partial angle reconstruction on the original scanned image of the second scanned object, the actual artifact movement can be refined, the movement artifact features can be extracted more precisely, which is beneficial to improving the accuracy of motion correction information. At the same time, the combination with the deep learning network not only reduces the complexity of the artifact movement estimation process, simplifies the artifact movement estimation process, but also can effectively reduce the motion artifacts in the CT scanned image, improve the image quality, and has broad application prospects and important clinical application value.

[0109] In a possible implementation manner, the above-mentioned fusion of multiple fifth scanned images in the fifth temporal image group to obtain the artifact correction scanned image of the second scanned object includes the following steps:

[0110] Perform pixel-level averaging on multiple fifth scanned images to obtain the artifact correction scanned image.

[0111] For example, the fifth temporal image group includes N fifth scanned images, and the pixel value of each fifth scanned image at position (a, b) is S n (a, b), n = 1, 2, 3,..., N. The pixel value S of the artifact correction scanned image at position (a, b) is obtained through the following formula (3). jz Obtained through the following formula (3).

[0112]

[0113] By adopting the method of pixel-level averaging, multiple fifth scanned images are fused into an artifact-corrected scanned image. Image fusion can be achieved without complex algorithms, which not only removes motion artifacts but also ensures the global consistency of the entire image.

[0114] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the method embodiment of the present application.

[0115] Figure 4 The structural schematic diagram of a model training apparatus provided by an embodiment of the present application is shown, as Figure 4 shown, the model training apparatus 400 includes:

[0116] An input module 410, configured to input a first time-series image group and a second time-series image group of a first scanned object into an image processing model; wherein, the first time-series image group includes multiple first scanned images with motion artifacts, the scanning angles corresponding to the multiple first scanned images are continuous and non-repetitive, the second time-series image group includes multiple second scanned images without motion artifacts, the scanning angles corresponding to the multiple second scanned images are continuous and non-repetitive, the first scanned images in the first time-series image group and the second scanned images in the second time-series image group correspond one-to-one and have the same scanning angle;

[0117] A first processing module 420, configured to register each first scanned image in the first time-series image group according to the second time-series image group, to obtain target motion correction information corresponding to each of the multiple first scanned images;

[0118] A second processing module 430, configured to, for each first scanned image, perform artifact correction on the first scanned image according to the target motion correction information corresponding to the first scanned image, to obtain a third time-series image group;

[0119] A training module 440, configured to determine the image difference between the third time-series image group and the second time-series image group according to the target loss function, and train the image processing model according to the image difference; wherein, the target loss function includes a first loss function and a second loss function, the first loss function is used to measure the global similarity between the second time-series image group and the third time-series image group, and the second loss function is used to measure the difference in edge information and detail information between the scanned images in the third time-series image group and the second scanned images.

[0120] In a possible implementation manner, the first processing module 420 includes:

[0121] A first processing unit, configured to extract multi-scale features of each first scanned image in the first time-series image group; perform feature difference on the multi-scale features of the first scanned image to obtain artifact features regarding motion artifacts corresponding to the first scanned image;

[0122] A second processing unit, configured to map the artifact features to a motion field regarding the motion artifacts to obtain initial motion correction information corresponding to the first scanned image;

[0123] A third processing unit, configured to correct the initial motion correction information based on the artifact features to obtain target motion correction information corresponding to each of the multiple first scanned images.

[0124] In a possible implementation manner, the third processing unit includes:

[0125] A first processing subunit, configured to determine a weight value corresponding to the initial motion correction information based on the artifact features;

[0126] A second processing subunit, configured to multiply the initial motion correction information by the weight value to obtain target motion correction information corresponding to each of the multiple first scanned images.

[0127] In a possible implementation manner, the first processing subunit is specifically configured to obtain the weight value by performing global non-linear compression and correlation mapping on the artifact features; and / or obtain the weight value by performing local time-series sliding convolution and pattern extraction on the artifact features.

[0128] In a possible implementation manner, the second processing module 430 is specifically configured to extract spatial transformation parameters from the first scanned image and the target motion correction information corresponding to the first scanned image; generate a sampling grid based on the spatial transformation parameters; resample the first scanned image based on the sampling grid to obtain the third time-series image group.

[0129] It should be noted that when the model training device provided in the above embodiments executes the model training method, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the model training device provided in the above embodiments and the model training method embodiments belong to the same concept. Therefore, for the details not disclosed in the device embodiments of the present application, please refer to the above model training method embodiments of the present application, which will not be elaborated here.

[0130] Figure 5The figure shows a schematic structural diagram of an image artifact correction device provided by an embodiment of the present application. As Figure 5 shown, the image artifact correction device 500 includes:

[0131] An image reconstruction module 510, configured to perform multi-angle reconstruction on the original scan image of a second scanned object to obtain a fourth time-series image group; wherein, the fourth time-series image group includes multiple fourth scan images, and the scan angles corresponding to the multiple fourth scan images are continuous and non-repetitive;

[0132] An image correction module 520, configured to input the fourth time-series image group into an image processing model, and the image processing model corrects the artifacts of each fourth scan image in the fourth time-series image and outputs a fifth time-series image group; wherein, the image processing model is trained based on the above model training method;

[0133] An image fusion module 530, configured to fuse multiple fifth scan images in the fifth time-series image group to obtain an artifact-corrected scan image of the second scanned object.

[0134] In a possible implementation manner, the image reconstruction module 510 is specifically configured to intercept an image within a first angle range from the original scan image to obtain a first image to be processed; wherein, the angle range of the original scan image includes the first angle range; split the first image to be processed into multiple image blocks according to multiple different second angle ranges to obtain multiple second images to be processed; wherein, the union of the multiple second angle ranges is the first angle range; perform reconstruction on the multiple second images to be processed to obtain the fourth time-series image group.

[0135] In a possible implementation manner, the image fusion module 530 is specifically configured to perform pixel-level averaging on the multiple fifth scan images to obtain the artifact-corrected scan image.

[0136] It should be noted that when the image artifact correction device provided in the above embodiment executes the image artifact correction method, only the above-mentioned division of each functional module is used for illustration. In actual application, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the image artifact correction device provided in the above embodiment and the embodiment of the image artifact correction method belong to the same concept. Therefore, for the details not disclosed in the device embodiment of the present application, please refer to the above embodiment of the image artifact correction method of the present application, and details will not be repeated here.

[0137] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0138] Figure 6 The figure shows a schematic structural diagram of an electronic device provided by an embodiment of the present application, as Figure 6 shown, the electronic device 600 includes: a memory 601 and a processor 602. Among them, an executable program code 6011 is stored in the memory 601, and the processor 602 is configured to call and execute the executable program code 6011 to execute a model training method or an image artifact correction method.

[0139] In this embodiment, the electronic device can be divided into functional modules according to the above method examples. For example, it can correspond to each functional module, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0140] In the case of dividing each functional module corresponding to each function, the electronic device may include: an input module, a first processing module, a second processing module, a third processing module, a training module, an image reconstruction module, an image correction module, an image fusion module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be cited in the function description of the corresponding functional module, and will not be repeated here.

[0141] The electronic device provided in this embodiment is used to execute the above model training method or image artifact correction method, so the same effect as the above implementation method can be achieved.

[0142] In the case of adopting an integrated unit, the electronic device may include a processing module and a storage module. Among them, the processing module can be used to control and manage the actions of the electronic device. The storage module can be used to support the electronic device to execute relevant program codes and data, etc.

[0143] Among them, the processing module can be a processor or a controller, which can implement or execute various exemplary logical blocks, modules, and circuits described in combination with the disclosure of the present application. The processor can also be a combination that realizes computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory.

[0144] This embodiment also provides a computer-readable storage medium, in which computer program code is stored. When the computer program code runs on a computer, the computer is enabled to execute the above relevant method steps to implement a model training method or an image artifact correction method in the above embodiment.

[0145] This embodiment also provides a computer program product. When the computer program product runs on a computer, it causes the computer to execute the above-related steps to implement a model training method or an image artifact correction method in the above embodiment.

[0146] In addition, the electronic device provided in the embodiment of the present application may specifically be a chip, a component or a module. The electronic device may include a processor and a memory connected thereto; wherein, the memory is used to store instructions. When the electronic device runs, the processor may call and execute the instructions to cause the chip to execute a model training method or an image artifact correction method in the above embodiment.

[0147] Among them, the electronic device, the computer-readable storage medium, the computer program product or the chip provided in this embodiment are all used to execute the corresponding model training method or image artifact correction method provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding model training method or image artifact correction method provided above, which will not be elaborated here.

[0148] Through the description of the above embodiments, those skilled in the art can understand that for the convenience and simplicity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0149] In the embodiments provided in the present application, it should be understood that the disclosed device and method can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0150] The above content is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A model training method, characterized in that: The model training method comprises: Inputting a first time-sequential image group and a second time-sequential image group of a first scan object into an image processing model; wherein the first time-sequential image group includes a plurality of first scan images with motion artifacts, and the scanning angles corresponding to the plurality of first scan images are continuous and non-repetitive, and the second time-sequential image group includes a plurality of second scan images without motion artifacts, and the scanning angles corresponding to the plurality of second scan images are continuous and non-repetitive, and the first scan images in the first time-sequential image group correspond to the second scan images in the second time-sequential image group in one-to-one correspondence and the scanning angles are the same; Registering each first scanned image in the first time-sequential image group according to the second time-sequential image group to obtain target motion correction information corresponding to each of the plurality of first scanned images; For each first scanned image, performing artifact correction on the first scanned image according to target motion correction information corresponding to the first scanned image to obtain a third time-series image group; The image difference between the third time-series image group and the second time-series image group is determined according to the target loss function, and the image processing model is trained according to the image difference; wherein the target loss function includes a first loss function and a second loss function, the first loss function is used to measure the global similarity between the second time-series image group and the third time-series image group, and the second loss function is used to measure the difference in edge information and detail information between the scanned image in the third time-series image group and the second scanned image.

2. The model training method according to claim 1, characterized in that: The registering each first scanned image in the first time-sequential image group according to the second time-sequential image group to obtain target motion correction information corresponding to each of the plurality of first scanned images comprises: For each first scanned image in the first time-series image group, extracting multi-scale features of the first scanned image; Performing feature differentiation on multi-scale features of the first scanned image to obtain artifact features related to motion artifacts corresponding to the first scanned image; Mapping the artifact feature into a motion field about the motion artifact to obtain initial motion correction information corresponding to the first scanned image; The initial motion correction information is corrected based on the artifact feature to obtain target motion correction information corresponding to each of the plurality of first scanned images.

3. The model training method according to claim 2, characterized in that: The correcting the initial motion correction information based on the artifact feature to obtain target motion correction information corresponding to each of the plurality of first scanned images includes: Determine a weight value corresponding to the initial motion correction information based on the artifact feature; The initial motion correction information is multiplied by the weight value to obtain target motion correction information corresponding to each of the plurality of first scanned images.

4. The model training method according to claim 3, characterized in that: Determining the weight value corresponding to the initial motion correction information based on the artifact feature comprises at least one of the following steps: Obtaining the weight value by performing global nonlinear compression and association mapping on the artifact features; The weight value is obtained by performing local temporal sliding convolution and pattern extraction on the artifact feature.

5. The model training method according to claim 3, characterized in that: The performing artifact correction on the first scanned image according to the target motion correction information corresponding to the first scanned image to obtain a third time-series image group comprises: extracting spatial transformation parameters from the first scanned image and target motion correction information corresponding to the first scanned image; Based on the spatial transformation parameters, generating a mesh; The first scanned image is resampled based on the adopted grid to obtain the third time-series image group.

6. A method for correcting image artifacts, characterized in that: The image artifact correction method comprises: Performing multi-angle reconstruction on the original scanned image of the second scanned object to obtain a fourth time-series image group; wherein the fourth time-series image group includes a plurality of fourth scanned images, and the scanning angles corresponding to the plurality of fourth scanned images are continuous and non-repetitive; Input the fourth time series image group into an image processing model, and the image processing model performs artifact correction on each fourth scan image in the fourth time series image, and outputs a fifth time series image group; wherein the image processing model is trained based on the model training method according to any one of claims 1 to 5; The plurality of fifth scanned images in the fifth time-series image group are fused to obtain an artifact-corrected scanned image of the second scanned object.

7. The image artifact correction method according to claim 6, characterized in that: The performing multi-angle reconstruction on the original scanned image of the second scanned object to obtain a fourth time-series image group comprises: An image within a first angle range is intercepted from the original scanned image to obtain a first image to be processed; wherein the angle range of the original scanned image includes the first angle range; Splitting the first image to be processed into a plurality of image blocks according to a plurality of different second angle ranges to obtain a plurality of second images to be processed; wherein the union of the plurality of second angle ranges is the first angle range; The plurality of second images to be processed are reconstructed to obtain the fourth time-series image group.

8. The image artifact correction method according to claim 6, characterized in that: The fusing of the plurality of fifth scanned images in the fifth time-series image group to obtain the artifact-corrected scanned image of the second scanned object comprises: The plurality of fifth scanned images are averaged at the pixel level to obtain the artifact-corrected scanned image.

9. A model training device, characterized in that: The model training device comprises: An input module, used for inputting a first time-sequential image group and a second time-sequential image group of a first scan object into an image processing model; wherein the first time-sequential image group includes a plurality of first scan images with motion artifacts, and the scanning angles corresponding to the plurality of first scan images are continuous and non-repeating, and the second time-sequential image group includes a plurality of second scan images without motion artifacts, and the scanning angles corresponding to the plurality of second scan images are continuous and non-repeating, and the first scan images in the first time-sequential image group correspond to the second scan images in the second time-sequential image group in one-to-one correspondence and the scanning angles are the same; A first processing module, configured to register each first scanned image in the first time-sequential image group according to the second time-sequential image group, to obtain target motion correction information corresponding to each of the plurality of first scanned images; A second processing module is used to perform artifact correction on each first scanned image according to target motion correction information corresponding to the first scanned image to obtain a third time-series image group; A training module, used to determine the image difference between the third time-series image group and the second time-series image group according to the target loss function, and train the image processing model according to the image difference; wherein the target loss function includes a first loss function and a second loss function, the first loss function is used to measure the global similarity between the second time-series image group and the third time-series image group, and the second loss function is used to measure the difference in edge information and detail information between the scanned image in the third time-series image group and the second scanned image.

10. An image artifact correction device, characterized in that: The image artifact correction device comprises: An image reconstruction module, configured to reconstruct the original scanned image of the second scanned object at multiple angles to obtain a fourth time-series image group; wherein, the fourth time-series image group includes a plurality of fourth scanned images, and the scanning angles corresponding to the plurality of fourth scanned images are continuous and non-repetitive; an image correction module, configured to input the fourth time-series image group into an image processing model, and the image processing model performs artifact correction on each fourth scan image in the fourth time-series image, and outputs a fifth time-series image group; wherein the image processing model is trained based on the model training method according to any one of claims 1 to 5; An image fusion module is used to fuse multiple fifth scan images in the fifth time-series image group to obtain an artifact-corrected scan image of the second scan object.

11. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable program codes; A processor, used to call and run the executable program code from the memory, so that the electronic device executes the model training method as described in any one of claims 1 to 5, or the image artifact correction method as described in any one of claims 6 to 8.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the model training method as described in any one of claims 1 to 5, or the image artifact correction method as described in any one of claims 6 to 8.

Images