Artifact elimination and image correction control system in MRI (Magnetic Resonance Imaging)

Abstract

The invention relates to the field of artificial intelligence, in particular to an artifact elimination and image correction control system in MRI (Magnetic Resonance Imaging), which comprises a data acquisition and preprocessing module, a hardware correction module, an artifact detection and classification module and an artifact correction and removal module. Through feature extraction and standardization processing, the accuracy of artifact detection and classification is improved; in the artifact correction and removal module, complex artifacts in an MRI image are removed by using an adversarial optimization process, the definition and contrast of the image are remarkably improved, in addition, a unit based on time sequence processing is also integrated, spatial-temporal features are extracted from a 2D image sequence, key information is provided for 3D reconstruction, the quality and diagnosis accuracy of the MRI image are remarkably improved, and the imaging quality of the MRI image is improved. And important technical support is provided for the development of clinical medical imaging.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an artifact elimination and image correction control system in MRI imaging. Background Art

[0002] With the advancement of science and technology, magnetic resonance imaging (MRI) has become one of the important tools of modern medical imaging. In the existing MRI imaging process, the presence of artifacts seriously affects the image quality and diagnostic accuracy. Therefore, effectively detecting, classifying and correcting these artifacts has become an important problem to be solved in MRI imaging technology. The existing artifact detection and classification methods have the problems of low detection accuracy, poor classification effect and low processing efficiency. The traditional classifier is inefficient when processing high-dimensional data, which leads to reduced accuracy of artifact detection, and is unstable during the processing process, resulting in poor artifact classification effect, affecting the subsequent correction and removal process; general artifact correction and removal methods have the problems of limited artifact elimination effect, low image quality and difficulty in three-dimensional reconstruction. Traditional methods are usually based on predefined rules and parameters, and it is difficult to deal with complex and diverse artifact types and forms. Conventional artifact correction methods may cause image blur, especially when processing artifact edge areas, details are easily lost, affecting the diagnostic value of the image. Due to the lack of spatiotemporal information integration, traditional 2D artifact correction methods are difficult to fully utilize the spatiotemporal information in 3D data and cannot effectively reconstruct high-quality 3D images. Summary of the invention

[0003] In view of the above situation, the present invention provides an artifact elimination and image correction control system in MRI imaging (Magnetic Resonance Imaging). The existing artifact detection and classification methods have the problems of low detection accuracy, poor classification effect and low processing efficiency. The present invention constructs a DNH classifier (Distributed Nearest Hashing, distributed nearest hashing), using DNH classifier for artifact detection, through feature extraction and standardization processing, the accuracy of artifact detection and classification is improved, the artifacts are classified, different correction and removal methods are adopted for different types of artifacts, the effect of artifact processing is improved, and distributed nearest hashing technology is used for feature processing and classification, which improves the efficiency of artifact detection and classification, and realizes efficient detection and accurate classification of artifacts in MRI images; in view of the problems that general artifact correction and removal methods have limited artifact elimination effect, low image quality, and difficulty in three-dimensional reconstruction, the present invention removes complex artifacts in MRI images by utilizing adversarial optimization process in the artifact correction and removal module, significantly improves the clarity and contrast of the image, obtains high-level feature information from the image after preliminary correction, captures subtle structural and texture features, and provides rich feature support for subsequent artifact removal. In addition, a unit based on time series processing is integrated to extract spatiotemporal features from 2D image sequences, provide key information for 3D reconstruction, greatly improve the overall quality and diagnostic accuracy of MRI images, and overcome the problem that the performance of traditional monitoring systems is reduced under artifact interference.

[0004] The present invention provides an artifact elimination and image correction control system in MRI imaging, including a data acquisition and preprocessing module, a hardware correction module, an artifact detection and classification module, and an artifact correction and removal module, which specifically includes the following contents:

[0005] The data acquisition and preprocessing module uses an MRI scanner to acquire MRI image data and performs preliminary processing, including noise filtering, offset field correction, and image standardization, and transmits the processed MRI image data to the artifact detection and classification module and the artifact correction and removal module;

[0006] The hardware correction module calibrates the gradient coil, the radio frequency coil and the system frequency, and performs environmental noise filtering and temperature correction at the same time;

[0007] The artifact detection and classification module uses a DNH classifier to perform artifact detection, obtain artifacts and classify them, and the classification includes motion artifacts, chemical shift artifacts and radio frequency interference;

[0008] The artifact correction and removal module uses the GAN model and the VGG-16 model combined with 3D reconstruction to perform artifact removal and image reconstruction.

[0009] Furthermore, in the artifact detection and classification module, the following contents are specifically included:

[0010] Obtain training data: Divide the processed MRI image data into image segments of 0.5 each, and each image segment is called a data point;

[0011] Feature extraction: add labels to each data point, including normal and artifact labels, and extract features from each data point. The extracted features include variance, peak-to-peak amplitude, and average rectified value;

[0012] Artifact detection and classification: The extracted features are standardized and all data points are detected using distributed nearest hashing, which includes the following steps:

[0013] Step Q1: Build a classifier, build a DNH classifier, use the hash map and primary key clustering order for processing, and define the DNH classifier parameters, including the following:

[0014] Number of hash functions: The hash function maps data points into a hash table, and the set range is [10,20];

[0015] Number of hash tables: stores hash function results, set in the range of [5, 10];

[0016] Wide column storage database: stores and manages hash map data structures;

[0017] Step Q2: Initialize the hash map and use the hash function and hash table to initialize the data structure of the hash map;

[0018] Step Q3: Feature input, input the standardized features and labels of each data point into the DNH classifier, and use the hash map and primary key clustering order to perform incremental training on the data points;

[0019] Step Q4: DNH classifier training, inserting data points into the hash map, establishing hash indexes, and mapping the standardized features of the data points into the hash table through hash functions;

[0020] Step Q5: Update the hash map, set the maximum number of iterations and stop updating when it is reached, and obtain the trained DNH classifier;

[0021] Step Q6: Acquire the data to be detected, and collect the unlabeled MRI image data to be detected from the data acquisition and preprocessing module;

[0022] Step Q7: data processing, preprocessing, feature extraction and labeling of unlabeled MRI image data to be detected;

[0023] Step Q8: Apply the DNH classifier and use the trained DNH classifier to perform artifact detection.

[0024] Furthermore, in the artifact correction and removal module, the GAN model and the VGG-16 model are combined with 3D reconstruction, which specifically includes the following steps:

[0025] Step S1: data preparation, using the MRI image data processed in the data acquisition and preprocessing module;

[0026] Step S2: feature extraction, extracting features from the processed MRI image data and standardizing them, the features include noise level, contrast-to-noise ratio and texture features;

[0027] Step S3: preliminary correction;

[0028] Step S4: Depth artifact removal.

[0029] Further, step S3 specifically includes the following steps:

[0030] Step S31: metal region detection, using threshold segmentation technology to identify metal regions in the processed MRI image data, marking the metal regions and generating a metal mask;

[0031] Step S32: artifact region detection, using a region growing algorithm to detect the artifact region caused by metal according to the metal mask, and marking the artifact region;

[0032] Step S33: interpolation correction, using high-order interpolation to process the metal area to obtain MRI image data after interpolation correction;

[0033] Step S34: image reconstruction, using an iterative reconstruction algorithm to reconstruct the MRI image data after interpolation correction;

[0034] Step S35: Model correction: correcting the artifact area using a correction method based on a physical model.

[0035] Furthermore, in step S35, the correction method based on the physical model includes the following contents:

[0036] Magnetic field inhomogeneity correction: Use magnetic field mapping to perform corrections and compensate for signal offsets caused by magnetic field inhomogeneity;

[0037] Gradient nonlinear correction: Based on the gradient nonlinear model, geometric distortion is corrected during image reconstruction;

[0038] Radio frequency interference correction: Detect radio frequency interference signals through spectrum analysis and use radio frequency filters to filter out interference;

[0039] Step S36: smoothing the artifact area, using a Gaussian filter to smooth the corrected artifact area and adjust the contrast.

[0040] Further, step S4 specifically includes the following steps:

[0041] Step S41: pre-training VGG-16 model;

[0042] Step S42: extracting high-level features from the preliminarily rectified image using a pre-trained VGG-16 model;

[0043] Step S43: 3D reconstruction, inputting high-level features into the LSTM unit based on the attention mechanism to obtain time series features;

[0044] Step S44: Establish a GAN model, which includes a discriminator and a generator, and specifically includes the following contents:

[0045] Discriminator: distinguishes between the generated artifact-free 3D images and the artifact-containing 3D images, and obtains feedback to be sent back to the generator for optimization;

[0046] Generator: Generate artifact-free 3D images based on time series features;

[0047] Step S45: Depth artifact removal: input the preliminarily corrected image into the GAN model, use the discriminator in the GAN model for judgment, and feed back to optimize the generator to obtain an artifact-free MRI image.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) The existing artifact detection and classification methods have the problems of low detection accuracy, poor classification effect and low processing efficiency. The present invention constructs a DNH classifier in the artifact detection and classification module, uses the DNH classifier for artifact detection, improves the accuracy of artifact detection and classification through feature extraction and standardization processing, classifies artifacts, adopts different correction and removal methods for different types of artifacts, improves the effect of artifact processing, uses distributed nearest hashing technology for feature processing and classification, improves the efficiency of artifact detection and classification, and realizes efficient detection and accurate classification of artifacts in MRI images;

[0050] (2) In view of the problems of limited artifact elimination effect, low image quality and difficulty in three-dimensional reconstruction in general artifact correction and removal methods, the present invention removes complex artifacts in MRI images by utilizing an adversarial optimization process in the artifact correction and removal module, significantly improving the clarity and contrast of the image, obtaining high-level feature information from the initially corrected image, capturing subtle structural and texture features, and providing rich feature support for subsequent artifact removal. In addition, a unit based on time series processing is integrated to extract spatiotemporal features from 2D image sequences, providing key information for 3D reconstruction, greatly improving the overall quality and diagnostic accuracy of MRI images, and overcoming the problem of reduced performance of traditional monitoring systems under artifact interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a module of an artifact elimination and image correction control system in MRI imaging provided by the present invention.

[0052] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0055] Example 1, see Figure 1 The present invention provides an artifact elimination and image correction control system in MRI imaging, including a data acquisition and preprocessing module, a hardware correction module, an artifact detection and classification module, and an artifact correction and removal module, which specifically includes the following contents:

[0056] The data acquisition and preprocessing module uses an MRI scanner to acquire MRI image data and performs preliminary processing, including noise filtering, offset field correction, and image standardization, and transmits the processed MRI image data to the artifact detection and classification module and the artifact correction and removal module;

[0057] The hardware correction module calibrates the gradient coil, the radio frequency coil and the system frequency, and performs environmental noise filtering and temperature correction at the same time;

[0058] The artifact detection and classification module uses a DNH classifier to perform artifact detection, obtain artifacts and classify them, and the classification includes motion artifacts, chemical shift artifacts and radio frequency interference;

[0059] The artifact correction and removal module uses the GAN model and the VGG-16 model combined with 3D reconstruction to perform artifact removal and image reconstruction.

[0060] Embodiment 2: This embodiment is based on the above embodiment and specifically includes the following contents in the artifact detection and classification module:

[0061] Obtain training data: Divide the processed MRI image data into image segments of 0.5 each, and each image segment is called a data point;

[0062] Feature extraction: add labels to each data point, including normal and artifacts, and perform feature extraction on each data point. The extracted features include variance, peak-to-peak amplitude, and average rectified value. The variance is the variance of the pixel values ​​of each image segment, the peak-to-peak amplitude is the difference between the maximum pixel value and the minimum pixel value of each image segment, and the average rectified value is the average absolute value of the pixel values ​​of each image segment.

[0063] Artifact Detection and Classification: The extracted features are normalized and all data points are detected using distributed nearest hashing.

[0064] Embodiment 3: This embodiment is based on the above embodiment and uses distributed recent hashing to detect all data points. In this embodiment, the following steps are specifically included:

[0065] Step Q1: Build a classifier, build a DNH classifier, use the hash map and primary key clustering order for processing, and define the DNH classifier parameters, including the following:

[0066] Number of hash functions: The hash function maps data points into the hash table, set to 15;

[0067] Number of hash tables: to store hash function results, set to 8;

[0068] Wide column storage database: Apache Cassandra is selected to store and manage hash map data structures;

[0069] Step Q2: Initialize the hash map and use the hash function and hash table to initialize the data structure of the hash map;

[0070] Step Q3: Feature input, input the standardized features and labels of each data point into the DNH classifier, and use the hash map and primary key clustering order to perform incremental training on the data points;

[0071] Incremental training of data points includes the following:

[0072] Step 1: Data point insertion: input the standardized features of the new data point into the DNH classifier, calculate the hash value of the data point using the hash function, and insert it into the corresponding hash table slot;

[0073] Step 2: Update the hash map, update the hash index in the hash map according to the new data point, insert the new data point into the hash table and maintain the structure of the hash map to ensure that it can correctly handle the new data point;

[0074] Step 3: Primary key clustering order, clustering data points using the primary key clustering order. In the DNH classifier, each data point is not simply inserted into the hash table, but also needs to be properly adjusted in the primary key clustering order according to its feature value;

[0075] Step 4: Adjust the clustering. Recalculate the distance between the data point and the cluster center according to the characteristics of the new data point, and adjust the location of the cluster center and the allocation of other data points as needed. This step requires the use of the hash map structure to quickly calculate the distance and make adjustments.

[0076] Step 5: Iterative update. During the entire incremental training process, the maximum number of iterations needs to be set to ensure the stability of the hash map and clustering structure. When the maximum number of iterations is reached, the update process is stopped to obtain the trained DNH classifier.

[0077] Step Q4: DNH classifier training, inserting data points into the hash map, establishing hash indexes, and mapping the standardized features of the data points into the hash table through hash functions;

[0078] Step Q5: Update the hash map, set the maximum number of iterations to 1000 and stop updating when it is reached, and obtain the trained DNH classifier;

[0079] Step Q6: Acquire the data to be detected, and collect the unlabeled MRI image data to be detected from the data acquisition and preprocessing module;

[0080] Step Q7: data processing, preprocessing, feature extraction and labeling of unlabeled MRI image data to be detected;

[0081] Step Q8: Apply the DNH classifier and use the trained DNH classifier to perform artifact detection;

[0082] In this embodiment, we have a set of MRI image data points, each of which contains multiple features, including pixel intensity, contrast, and texture features, and a label of normal or artifact;

[0083] Feature standardization: Each feature is standardized using zero mean unit variance standardization;

[0084] Hash function initialization: select 10 hash functions and 5 hash tables;

[0085] Data point insertion: Each new data point is mapped to a slot in the hash table through a hash function;

[0086] Primary key clustering order adjustment: adjust the clustering structure in the hash map according to the feature values ​​of the new data points to ensure that the new data points are correctly classified;

[0087] Iterative update: Stop updating after a maximum of 10 iterations.

[0088] Embodiment 4: This embodiment is based on the above embodiment. In the artifact correction and removal module, the GAN model and the VGG-16 model are combined with 3D reconstruction, and specifically includes the following steps:

[0089] Step S1: data preparation, using the MRI image data processed in the data acquisition and preprocessing module;

[0090] Step S2: feature extraction, extracting features from the processed MRI image data and standardizing them, the features include noise level, contrast-to-noise ratio and texture features;

[0091] Step S3: Preliminary correction, denoising and normalizing the MRI image data, using a Gaussian filter to remove noise in the image and normalizing the pixel values ​​to the range of [0,1];

[0092] Step S3 specifically includes the following steps:

[0093] Step S31: metal area detection, using threshold segmentation technology to identify the metal area in the processed MRI image data, the pixel value range of the MRI image is 0 to 4095, the threshold is set to 2000, the metal area is marked and a metal mask is generated;

[0094] Step S32: artifact region detection, using a region growing algorithm to detect the artifact region caused by metal according to the metal mask, and marking the artifact region;

[0095] Step S33: interpolation correction, using a cubic spline interpolation method to process the metal area to obtain interpolation-corrected MRI image data;

[0096] Step S34: image reconstruction, using an iterative reconstruction algorithm to reconstruct the MRI image data after interpolation correction;

[0097] Step S35: Model correction, using a correction method based on a physical model to correct the artifact area; in step S35, the correction method based on a physical model includes the following contents:

[0098] Magnetic field inhomogeneity correction: Use magnetic field mapping to perform corrections and compensate for signal offsets caused by magnetic field inhomogeneity;

[0099] Gradient nonlinear correction: Based on the gradient nonlinear model, geometric distortion is corrected during image reconstruction;

[0100] Radio frequency interference correction: Detect radio frequency interference signals through spectrum analysis and use radio frequency filters to filter out interference;

[0101] Step S36: smoothing the artifact area, using a Gaussian filter to smooth the corrected artifact area and adjust the contrast;

[0102] Step S4: Depth artifact removal.

[0103] Embodiment 5, based on the above embodiment, step S4 specifically includes the following steps:

[0104] Step S41: Pre-train the VGG-16 model using the public ImageNet dataset;

[0105] Step S42: extracting high-level features from the preliminarily rectified image using a pre-trained VGG-16 model;

[0106] Step S43: 3D reconstruction, inputting high-level features into the LSTM unit based on the attention mechanism to obtain time series features;

[0107] Step S44: Establish a GAN model, which includes a discriminator and a generator, and specifically includes the following contents:

[0108] Discriminator: used to distinguish between the generated artifact-free 3D images and images with artifacts, and provide feedback to the generator to optimize its performance. The discriminator network structure adopts a multi-layer convolutional neural network;

[0109] Generator: Generates artifact-free 3D images based on the time series features output by the LSTM unit. The generator network structure uses a convolutional neural network combined with a deconvolution layer to achieve image generation;

[0110] Step S45: Depth artifact removal: input the preliminarily corrected image into the GAN model, use the discriminator in the GAN model for judgment, and feed back to optimize the generator to obtain an artifact-free MRI image.

[0111] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0112] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0113] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An artifact elimination and image correction control system in MRI imaging, characterized in that: It includes data acquisition and preprocessing module, hardware correction module, artifact detection and classification module, and artifact correction and removal module, specifically including the following contents: The data acquisition and preprocessing module uses an MRI scanner to acquire MRI image data and performs preliminary processing, and transmits the processed MRI image data to the artifact detection and classification module and the artifact correction and removal module; The hardware correction module calibrates the gradient coil, the radio frequency coil and the system frequency, and performs environmental noise filtering and temperature correction at the same time; The artifact detection and classification module uses a DNH classifier to perform artifact detection, obtain artifacts and classify them; The artifact correction and removal module uses the GAN model and the VGG-16 model combined with 3D reconstruction to perform artifact removal and image reconstruction.

2. The artifact elimination and image correction control system in MRI imaging according to claim 1, characterized in that: In the artifact detection and classification module, the following contents are included: Obtain training data: Divide the processed MRI image data into image segments of 0.5 each, and each image segment is called a data point; Feature extraction: add labels to each data point, including normal and artifact labels, and perform feature extraction on each data point; Artifact detection and classification: The extracted features are standardized and all data points are detected using distributed nearest hashing, which includes the following steps: Step Q1: Build a classifier, build a DNH classifier, use the hash map and primary key clustering order for processing, and define the DNH classifier parameters, including the following: Number of hash functions: The hash function maps data points into a hash table, and the set range is [10, 20]; Number of hash tables: stores hash function results, set in the range of [5, 10]; Wide column storage database: stores and manages hash map data structures; Step Q2: Initialize the hash map and use the hash function and hash table to initialize the data structure of the hash map; Step Q3: Feature input, input the standardized features and labels of each data point into the DNH classifier, and use the hash map and primary key clustering order to perform incremental training on the data points; Step Q4: DNH classifier training, inserting data points into the hash map, establishing hash indexes, and mapping the standardized features of the data points into the hash table through the hash function; Step Q5: Update the hash map, set the maximum number of iterations and stop updating when it is reached, and obtain the trained DNH classifier; Step Q6: Acquire the data to be detected, and collect the unlabeled MRI image data to be detected from the data acquisition and preprocessing module; Step Q7: data processing, preprocessing, feature extraction and labeling of unlabeled MRI image data to be detected; Step Q8: Apply the DNH classifier and use the trained DNH classifier to perform artifact detection.

3. An artifact elimination and image correction control system in MRI imaging according to claim 2, characterized in that: In the artifact correction and removal module, the GAN model and the VGG-16 model are combined with 3D reconstruction, which specifically includes the following steps: Step S1: data preparation, using the MRI image data processed in the data acquisition and preprocessing module; Step S2: feature extraction, extracting features from the processed MRI image data and standardizing them; Step S3: preliminary correction; Step S4: Depth artifact removal.

4. The artifact elimination and image correction control system in MRI imaging according to claim 3, characterized in that: Further, step S3 specifically includes the following steps: Step S31: metal region detection, using threshold segmentation technology to identify metal regions in the processed MRI image data, marking the metal regions and generating a metal mask; Step S32: artifact region detection, using a region growing algorithm to detect the artifact region caused by metal according to the metal mask, and marking the artifact region; Step S33: interpolation correction, using high-order interpolation to process the metal area to obtain MRI image data after interpolation correction; Step S34: image reconstruction, using an iterative reconstruction algorithm to reconstruct the MRI image data after interpolation correction; Step S35: Model correction, using a correction method based on a physical model to correct the artifact area; Furthermore, in step S35, the correction method based on the physical model includes the following contents: Magnetic field inhomogeneity correction: Use magnetic field mapping to perform corrections and compensate for signal offsets caused by magnetic field inhomogeneity; Gradient nonlinear correction: Based on the gradient nonlinear model, geometric distortion is corrected during image reconstruction; Radio frequency interference correction: Detect radio frequency interference signals through spectrum analysis and use radio frequency filters to filter out interference; Step S36: smoothing the artifact area, using a Gaussian filter to smooth the corrected artifact area and adjust the contrast.

5. The artifact elimination and image correction control system in MRI imaging according to claim 4, characterized in that: Step S4 specifically includes the following steps: Step S41: pre-training VGG-16 model; Step S42: extracting high-level features from the preliminarily rectified image using a pre-trained VGG-16 model; Step S43: 3D reconstruction, inputting high-level features into the LSTM unit based on the attention mechanism to obtain time series features; Step S44: Establish a GAN model, which includes a discriminator and a generator, and specifically includes the following contents: Discriminator: distinguishes between the generated artifact-free 3D images and the artifact-containing 3D images, and obtains feedback to be sent back to the generator for optimization; Generator: Generate artifact-free 3D images based on time series features; Step S45: Depth artifact removal: input the preliminarily corrected image into the GAN model, use the discriminator in the GAN model for judgment, and feed back to optimize the generator to obtain an artifact-free MRI image.

Images