System and method for reducing artifacts in medical images using analog images

Abstract

The present disclosure provides methods and systems for reducing the amount of artifacts in image data. In one example, a method for an image processing system includes generating a plurality of analog images (402); generating a training image pair set based on the plurality of simulated magnetic resonance images (404); training an artifact removal neural network for the set using the training image (410); and generating an output of the artifact removal neural network based on the input acquired image (606), wherein generating the plurality of analog images includes generating an image from the RGB image (502), simulating motion in the analog image (504), simulating contrast in the analog image (506), and simulating phase contrast dynamics in the analog image (508).

Description

Technical Field

[0001] Embodiments of the subject matter disclosed herein relate to medical imaging, and more particularly to systems and methods for removing artifacts from medical images. Background Art

[0002] Medical images such as magnetic resonance (MR) images may be acquired at various phases. Multi-phase data provides a larger amount of data, whereby images are acquired at multiple time points, providing various imaging phases or echoes, but in turn requires longer scan times than single-phase imaging. In order to reduce scan time, each phase may be undersampled in k-space because there is redundant data between the acquired phases. Undersampling of data may result in increased artifacts in the output MR image, which may reduce the quality of the image and hinder diagnosis. Various approaches have been taken to reduce or remove these undersampling artifacts, including mutual reconstruction methods that exploit temporal information, however such methods generally require high computational requirements. In some examples, a convolutional neural network (CNN) may be trained to reduce artifacts in MR images, which allows for reduced computational requirements during reconstruction. The CNN may be trained based on an image pair comprising a first undersampled input image and a second fully sampled target (reference) image. The CNN learns to map the undersampled image to the fully sampled image, and when trained, the CNN may output an artifact-reduced version of the MR image for the MR image input into the CNN.

[0003] However, MR images often include motion, e.g., periodic motion, such as respiration. Obtaining motion-free images and / or fully sampled contrast-enhanced images with high temporal resolution to use as targets for training CNNs is challenging, and thus, the lack of sufficient benchmark data makes such CNNs infeasible for application. Summary of the invention

[0004] In one example, a method for an image processing system includes: generating a plurality of simulated images, wherein the plurality of simulated images include simulated motion, contrast, and phase contrast dynamics; generating a set of training image pairs based on a plurality of simulated magnetic resonance images; training an artifact removal neural network using the set of training image pairs; and generating an output of the artifact removal neural network based on an input acquired image, wherein generating the plurality of simulated images includes generating images from RGB images, simulating motion in the simulated images, simulating contrast in the simulated images, and simulating phase contrast dynamics in the simulated images.

[0005] It should be understood that the above brief description is provided to introduce in a simplified form selected concepts that are further described in the detailed description. It is not meant to identify key features or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims that follow the detailed description. Furthermore, the claimed subject matter is not limited to implementations that solve any disadvantages noted above or in any part of this disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Various aspects of the present disclosure may be better understood by reading the following detailed description and referring to the accompanying drawings, in which:

[0007] Figure 1 shows a block diagram of an MRI system according to one or more embodiments of the present disclosure;

[0008] Figure 2 A block diagram illustrating an exemplary embodiment of an image processing system according to one or more embodiments of the present disclosure;

[0009] Figure 3 A block diagram illustrating an exemplary embodiment of an artifact removal neural network training system for training a neural network according to one or more embodiments of the present disclosure;

[0010] Figure 4 A flowchart illustrating an exemplary method for training an artifact removal neural network according to one or more embodiments of the present disclosure is shown;

[0011] Figure 5 A flow chart illustrating an exemplary method for simulating an image to be used for training a neural network according to one or more embodiments of the present disclosure is shown;

[0012] Figure 6 A flow chart illustrating an exemplary method for removing artifacts from undersampled MR images using a trained artifact removal neural network according to one or more embodiments of the present disclosure is shown;

[0013] Fig. 7A shows exemplary simulated images of different motion stages according to one or more embodiments of the present disclosure;

[0014] Figure 7B showing a second exemplary simulated image of a different phase according to one or more embodiments of the present disclosure;

[0015] Figure 8 shows an exemplary input MR image, a target MR image, and an output MR image according to one or more embodiments of the present disclosure;

[0016] Fig. 9shows an architectural diagram of a first neural network according to one or more embodiments of the present disclosure; and

[0017] Fig.10 An architectural diagram of a second neural network according to one or more embodiments of the present disclosure is shown.

[0018] The accompanying drawings illustrate specific aspects of the described systems and methods. Together with the following description, the accompanying drawings show and explain the structures, methods and principles described herein. In the accompanying drawings, the sizes of components may be enlarged or otherwise modified for clarity. Well-known structures, materials or operations are not shown or described in detail to avoid obscuring various aspects of the described components, systems and methods. DETAILED DESCRIPTION

[0019] Provided herein are methods and systems for reducing artifacts in medical image data such as undersampled magnetic resonance (MR) images. Multiphase and / or multi-echo imaging describes an imaging technique in which imaging data is collected at various phases or echoes, such as as a contrast agent moves through a patient, which typically requires a longer scan time than single-phase and / or single-echo imaging. MRI imaging uses nuclear magnetic resonance, gradient fields, and hydrogen atoms in the subject to produce cross-sectional images with high spatial resolution. MRI scan time may be proportional or approximately proportional to the number of time-consuming phase encoding steps in k-space. In order to reduce scan time, multiphase and / or multi-echo data are typically undersampled in k-space. However, undersampling typically results in undersampling directional artifacts in MR images.

[0020] Various methods have been developed to reduce undersampling directional artifacts. For example, advanced joint or mutual reconstruction methods utilize time information in multiple phases and / or echoes to reduce artifacts in the output reconstructed image. However, such methods are plagued by high processing and / or computational requirements. Recently, deep learning-based methods have been developed to address the high processing requirements of methods such as joint reconstruction, however, the lack of appropriate benchmark data sets for deep learning methods to be trained is a challenge. For example, in vivo imaging often includes motion artifacts, which are due to periodic motion and / or in some cases non-periodic motion. Therefore, appropriately matched input images and target images for neural network training are not easy to obtain.

[0021] In order to increase the availability of training data for deep learning-based neural networks, this paper proposes systems and methods that rely on high-quality, high-resolution simulated training data. Although other methods have included training of deep learning-based networks, the generation of simulated images as training data allows 1-1 matching of input images and target images. Simulated images may include simulated motion, simulated contrast levels, and simulated phase contrast dynamics that may exist in certain MRI applications. Due to known cycles, periodic motions such as breathing are reproducible, and therefore motion phases can be simulated in simulated MR images. Various contrast levels at different phases in multi-phase and / or multi-echo imaging can also be reproduced in a similar manner to allow simulated MR images to include such data.

[0022] A training data set including training image pairs can be generated based on simulated images. For example, the simulated images can be based on high-quality reference images (e.g., natural images) from RGB channels. These high-quality simulated images can be corresponding targets in the training image pairs. Corresponding inputs can be generated by undersampling the simulated images in k-space, so that the input images have lower resolution due to undersampling directional artifacts. By training an artifact removal neural network based on these training image pairs of simulated images, artifact-reduced images based on the undersampled input images can be output and displayed. The use of deep learning and neural networks described herein can reduce scan times while reducing artifacts and allowing high temporal resolution for more accurate diagnosis and higher diagnostic confidence.

[0023] In an embodiment, the medical image may be obtained by, for example, Figure 1 The imaging system can be acquired by an MRI system such as the one shown. Figure 2 An image processing system such as the image processing system 202 reduces or removes artifacts from an image. The image processing system may include an artifact removal neural network model that takes an undersampled image as input and outputs an artifact-reduced version of the image. Figure 4 One or more steps of the method are used to train an artifact removal neural network model, such as combining Figure 3 The denoising neural network model can be trained based on synthetic MR data, which can be obtained by following Figure 5 The simulated data may include simulated motion phase and contrast phase, such as Fig. 7A and Figure 7B shown. Figure 8 An example of an undersampled input image, the corresponding high-quality target image, and the output image of the neural network is shown in FIG. Fig. 9 and Fig.10 , an architectural diagram of the first neural network and the second neural network as described herein is shown.

[0024] Figure 1 An exemplary imaging system that can be used to acquire medical imaging data is illustrated. Figure 1 A magnetic resonance imaging (MRI) system is illustrated, but it should be understood that other medical imaging systems may be used without departing from the scope of the present disclosure. Figure 1 A magnetic resonance imaging (MRI) apparatus 10 is shown, which includes a static magnetic field magnet unit 12, a gradient coil unit 13, an RF coil unit 14, an RF body or volume coil unit 15, a transmit / receive (T / R) switch 20, an RF driver unit 22, a gradient coil driver unit 23, a data acquisition unit 24, a controller unit 25, a patient couch or bed 26, a data processing unit 31, an operation console unit 32, and a display unit 33. In some embodiments, the RF coil unit 14 is a surface coil, which is a local coil that is usually placed near an anatomical structure of interest of a subject 16. Here, the RF body coil unit 15 is a transmission coil that transmits RF signals, and the local surface RF coil unit 14 receives MR signals. Therefore, the transmission body coil (e.g., the RF body coil unit 15) and the surface receiving coil (e.g., the RF coil unit 14) are independent but electromagnetically coupled components. The MRI apparatus 10 transmits electromagnetic pulse signals to a subject 16 placed in an imaging space 18 forming a static magnetic field to perform scanning to obtain magnetic resonance signals from the subject 16. One or more images of the subject 16 can be reconstructed based on the magnetic resonance signals thus obtained by the scanning.

[0025] The static magnetic field magnet unit 12 includes, for example, an annular superconducting magnet installed in an annular vacuum container. The magnet defines a cylindrical space around the subject 16 and generates a constant main static magnetic field B0.

[0026] The MRI apparatus 10 further includes a gradient coil unit 13, which forms a gradient magnetic field in an imaging space 18 so as to provide three-dimensional position information for a magnetic resonance signal received by an RF coil array. The gradient coil unit 13 includes three gradient coil systems, each of which generates a gradient magnetic field along one of three spatial axes perpendicular to each other, and generates a gradient field in each of a frequency encoding direction, a phase encoding direction, and a slice selection direction according to an imaging condition. More specifically, the gradient coil unit 13 applies a gradient field in a slice selection direction (or a scanning direction) of the subject 16 to select a slice; and the RF body coil unit 15 or the local RF coil array can transmit an RF pulse to the selected slice of the subject 16. The gradient coil unit 13 also applies a gradient field in a phase encoding direction of the subject 16 to phase encode a magnetic resonance signal from a slice excited by an RF pulse. Then the gradient coil unit 13 applies a gradient field in a frequency encoding direction of the subject 16 to frequency encode a magnetic resonance signal from a slice excited by an RF pulse.

[0027] The RF coil unit 14 is provided, for example, to surround a region to be imaged of the subject 16. In some examples, the RF coil unit 14 may be referred to as a surface coil or a receiving coil. In a static magnetic field space or an imaging space 18 in which a static magnetic field B0 is formed by the static magnetic field magnet unit 12, the RF coil unit 15 transmits RF pulses as electromagnetic waves to the subject 16 based on a control signal from the controller unit 25, thereby generating a high-frequency magnetic field B1. This excites the proton spins in the slice to be imaged of the subject 16. The RF coil unit 14 receives the electromagnetic waves generated when the proton spins thus excited in the slice to be imaged of the subject 16 return to alignment with the initial magnetization vector as magnetic resonance signals. In some embodiments, the RF coil unit 14 may transmit RF pulses and receive MR signals. In other embodiments, the RF coil unit 14 may be used only to receive MR signals, and not to transmit RF pulses.

[0028] The RF body coil unit 15 is disposed, for example, to surround the imaging space 18, and generates RF magnetic field pulses orthogonal to the main magnetic field B0 generated by the static magnetic field magnet unit 12 in the imaging space 18 to excite the nucleus. Compared with the RF coil unit 14 that can be disconnected from the MRI device 10 and replaced with another RF coil unit, the RF body coil unit 15 is fixedly attached and connected to the MRI device 10. In addition, although the local coil (such as the RF coil unit 14) can transmit or receive signals only to or from a local area of ​​the subject 16, the RF body coil unit 15 generally has a larger coverage area. For example, the RF body coil unit 15 can be used to transmit or receive signals to the whole body of the subject 16. The use of a receive-only local coil and a transmission body coil provides uniform RF excitation and good image uniformity at the expense of higher RF power deposited in the subject. For a transmit-receive local coil, the local coil provides RF excitation to the region of interest and receives MR signals, thereby reducing the RF power deposited in the subject. It should be understood that the specific use of the RF coil unit 14 and / or the RF body coil unit 15 depends on the imaging application.

[0029] The T / R switch 20 may selectively electrically connect the RF body coil unit 15 to the data acquisition unit 24 when operating in the receive mode, and may selectively electrically connect the RF body coil unit to the RF driver unit 22 when operating in the transmit mode. Similarly, the T / R switch 20 may selectively electrically connect the RF coil unit 14 to the data acquisition unit 24 when the RF coil unit 14 operates in the receive mode, and may selectively electrically connect the RF coil unit to the RF driver unit 22 when the RF coil unit operates in the transmit mode. When both the RF coil unit 14 and the RF body coil unit 15 are used for a single scan, for example, if the RF coil unit 14 is configured to receive MR signals and the RF body coil unit 15 is configured to transmit RF signals, the T / R switch 20 may direct the control signal from the RF driver unit 22 to the RF body coil unit 15 while directing the received MR signal from the RF coil unit 14 to the data acquisition unit 24. The coil of the RF body coil unit 15 may be configured to operate in a transmit-only mode or a transmit-receive mode. The coil of the local RF coil unit 14 may be configured to operate in a transmit-receive mode or a receive-only mode.

[0030] The RF driver unit 22 includes a gate modulator (not shown), an RF power amplifier (not shown), and an RF oscillator (not shown), which are used to drive an RF coil (e.g., the RF coil unit 15) and form a high-frequency magnetic field in the imaging space 18. The RF driver unit 22 modulates an RF signal received from the RF oscillator into a predetermined timing signal having a predetermined envelope based on a control signal from the controller unit 25 and using the gate modulator. The RF signal modulated by the gate modulator is amplified by the RF power amplifier and then output to the RF coil unit 15.

[0031] The gradient coil driver unit 23 drives the gradient coil unit 13 based on a control signal from the controller unit 25 and thereby generates a gradient magnetic field in the imaging space 18. The gradient coil driver unit 23 includes three driver circuits (not shown) corresponding to three gradient coil systems included in the gradient coil unit 13.

[0032] The data acquisition unit 24 includes a preamplifier (not shown), a phase detector (not shown), and an analog / digital converter (not shown) for acquiring a magnetic resonance signal received by the RF coil unit 14. In the data acquisition unit 24, the phase detector uses the output of the RF oscillator from the RF driver unit 22 as a reference signal, performs phase detection on the magnetic resonance signal received from the RF coil unit 14 and amplified by the preamplifier, and outputs the phase-detected analog magnetic resonance signal to the analog / digital converter to be converted into a digital signal. The digital signal thus obtained is output to the data processing unit 31.

[0033] The MRI apparatus 10 includes a table 26 for placing the subject 16 thereon. By moving the table 26 based on a control signal from the controller unit 25, the subject 16 can be moved inside and outside the imaging space 18.

[0034] The controller unit 25 includes a computer and a recording medium on which a program to be executed by the computer is recorded. The program, when executed by the computer, causes the various parts of the device to perform operations corresponding to a predetermined scan. The recording medium may include, for example, a ROM, a floppy disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, or a non-volatile memory card. The controller unit 25 is connected to the operation console unit 32 and processes an operation signal input to the operation console unit 32, and also controls the examination bed 26, the RF driver unit 22, the gradient coil driver unit 23, and the data acquisition unit 24 by outputting control signals to them. The controller unit 25 also controls the data processing unit 31 and the display unit 33 based on the operation signal received from the operation console unit 32 to obtain a desired image.

[0035] The operation console unit 32 includes a user input device such as a touch screen, a keyboard, and a mouse. The operation console unit 32 is used by the operator to input such data as an imaging protocol, and to set the area where the imaging sequence will be executed. The data on the imaging protocol and the imaging sequence execution area are output to the controller unit 25.

[0036] The data processing unit 31 includes a computer and a recording medium on which a program to be executed by the computer to perform predetermined data processing is recorded. The data processing unit 31 is connected to the controller unit 25, and performs data processing based on a control signal received from the controller unit 25. The data processing unit 31 is also connected to the data acquisition unit 24, and generates spectrum data by applying various image processing operations to the magnetic resonance signals output from the data acquisition unit 24.

[0037] The display unit 33 includes a display device, and displays an image on a display screen of the display device based on a control signal received from the controller unit 25. The display unit 33 displays, for example, images regarding input items for which an operator inputs operation data from the operation console unit 32. The display unit 33 also displays a two-dimensional (2D) slice image or a three-dimensional (3D) image of the subject 16 generated by the data processing unit 31.

[0038] Although an MRI system is described by way of example, it should be understood that the present techniques may also be useful when applied to images acquired using other imaging modalities, such as CT, tomosynthesis, PET, C-arm angiography, etc. The present discussion of the MRI imaging modality is provided merely as an example of one suitable imaging modality.

[0039] Reference now Figure 2 , an image processing system 202 of a medical imaging system 200 according to an embodiment is shown. In some embodiments, at least a portion of the image processing system 202 is disposed at a device (e.g., an edge device, a server, etc.) that is communicatively coupled to the medical imaging system 200 via a wired connection and / or a wireless connection. In some embodiments, at least a portion of the image processing system 202 is disposed at a separate device (e.g., a workstation) that can receive images from the medical imaging system 200 or from a storage device storing images / data generated by the medical imaging system 200. It should be understood that although an MRI system, MR images, and simulated MR images are described herein, other types of imaging and images are possible without departing from the scope of the present disclosure.

[0040] The image processing system 202 includes a processor 204 configured to execute machine-readable instructions stored in a non-transitory memory 206. The processor 204 may be a single-core or multi-core processor, and the program executed thereon may be configured for parallel processing or distributed processing. In some embodiments, the processor 204 may optionally include separate components spread across two or more devices, which may be remotely located and / or configured for collaborative processing. In some embodiments, one or more aspects of the processor 204 may be virtualized and performed by a remotely accessible networked computing device configured in a cloud computing configuration.

[0041] The non-transitory memory 206 may store a neural network module 208, a network training module 210, an inference module 212, and medical image data 214. The neural network module 208 may include a deep learning network and instructions for implementing the deep learning network to reduce or optionally remove noise from a medical image of the medical image data 214, as described in more detail below. The neural network module 208 may include one or more trained and / or untrained neural networks, and may also include various data or metadata related to the one or more neural networks stored therein.

[0042] The training module 210 may include instructions for training one or more of the neural networks implementing the deep learning model stored in the neural network module 208. In particular, the training module 210 may include instructions that, when executed by the processor 204, cause the image processing system 202 to perform one or more of the steps of the method 400 for training one or more neural networks in a training phase, which will be described below with reference to Figure 3 and Figure 4 The training module 210 may include a simulated MR image generator 211 that may be used to generate simulated images and training data to train one or more neural networks, as described below with reference to Figure 3 and Figure 5 In some embodiments, the training module 210 includes instructions for implementing one or more gradient descent algorithms, applying one or more loss functions and / or training routines for adjusting parameters of one or more neural networks of the neural network module 208. The non-transitory memory 206 also stores an inference module 212, which includes instructions for reducing the amount of artifacts in new image data using a trained deep learning model.

[0043] The non-transitory memory 206 also stores medical image data 214. The medical image data 214 may include, for example, medical images acquired via an MRI scanner, a CT scanner, a scanner for spectral imaging, or via different imaging modalities. For example, the medical image data 214 may store images of the same anatomical features of the same patient acquired via an MRI scanner. The medical image data 214 may also include histological images of anatomical structures generated from slices of anatomical samples.

[0044] In some embodiments, non-transitory memory 206 may include components disposed at two or more devices that may be remotely located and / or configured for collaborative processing. In some embodiments, one or more aspects of non-transitory memory 206 may include a remotely accessible networked storage device configured in a cloud computing configuration.

[0045] The image processing system 202 may be operably / communicatively coupled to a user input device 232 and a display device 234. The user input device 232 may include one or more of a touch screen, a keyboard, a mouse, a trackpad, a motion sensing camera, or other devices configured to enable a user to interact with and manipulate data within the image processing system 202. The display device 234 may include one or more display devices utilizing virtually any type of technology. In some embodiments, the display device 234 may include a computer monitor and may display medical images. The display device 234 may be combined with the processor 204, the non-volatile memory 206, and / or the user input device 232 in a shared housing, or may be a peripheral display device and may include a monitor, a touch screen, a projector, or other display device known in the art that enables a user to view medical images generated by the medical imaging system and / or interact with various data stored in the non-volatile memory 206. In some examples, the display device 234 may be Figure 1 The display unit 33 and the user input device 232 may be Figure 1 at least a portion of the operating console unit 32.

[0046] Image processing system 202 may be operatively / communicatively coupled to MRI scanner 236. MRI scanner 236 may be, for example, Figure 1The image processing system 202 may include any MRI imaging device such as the MRI device 10 of the present invention, which is configured to image a subject such as a patient, an inanimate object, one or more manufactured parts, and / or a foreign object such as a dental implant, a stent, and / or a contrast agent present in the body. The image processing system 202 may receive MRI imaging data from the MRI scanner 236, process the received MRI imaging data via the processor 204 based on instructions stored in one or more modules of the non-transitory memory 206, and / or store the received MRI imaging data in the medical image data 214.

[0047] It should be understood that Figure 1 The image processing system 202 shown is for illustration and not for limitation. Another suitable image processing system may include more, fewer, or different components.

[0048] refer to Figure 3 , shows an example of an artifact removal neural network training system 300, which can be used to train a neural network, such as an artifact removal neural network 302. Figure 3 In one or more operations described in more detail in method 300, an artifact removal neural network 302 may be trained to detect and reduce or optionally remove noise from a two-dimensional (2D) CT image. Artifact removal neural network training system 300 may be implemented by, for example, Figure 1 The image processing system 202 of the present invention is implemented to train the artifact removal neural network 302 to detect and reduce or optionally remove artifacts in the MR images.

[0049] In some embodiments, the artifact removal neural network 302 can be a deep neural network with multiple hidden layers. In one embodiment, the artifact removal neural network 302 is a convolutional neural network (CNN).

[0050] The artifact removal neural network 302 may be stored in a neural network module 301 of the image processing system. The neural network module 301 may be Figure 2 The artifact removal neural network training system 300 also includes a training module 304 that includes a training data set that includes a plurality of data training pairs, such as image pairs divided into training image pairs 306 and test image pairs 308. The training module 304 may be Figure 2 2 is a non-limiting example of a training module 210 of the image processing system 202.

[0051] The plurality of training image pairs 306 and test image pairs 308 may be selected to ensure that sufficient training data is available to prevent overfitting, whereby the artifact removal neural network 302 learns to map specific features of the training set samples that are not present in the test set.

[0052] Each image pair of the training image pair 306 and the test image pair 308 includes an input image and a target image. The input image and the target image can be simulated MR images generated by the simulated image generator 360 from RGB (red-green-blue) images, such as Figure 5 Further described. In various embodiments, the input image is a simulated MR image that has been undersampled in k-space. During training, the artifact removal neural network 302 can learn to distinguish undersampled directional artifacts from anatomical features of the target image.

[0053] The artifact removal neural network training system 300 may include a simulated image generator 360 that can be used to generate simulated images of different motion phases, phases, and / or contrasts. The simulated image generator 360 can generate a high-quality simulated MR image 312 and an undersampled simulated MR image 316. The undersampled simulated MR image 316 can be a corresponding undersampled version of the simulated MR image 312 that is undersampled in k-space 350.

[0054] The artifact removal neural network training system 300 may include a training data generator 310 that may be used to generate training image pairs 306 and test image pairs 308 for the training module 304. Images from high-quality simulated MR images 312 may be paired with images from undersampled simulated MR images 316 by the training data generator 310. Figure 5 An exemplary method for generating training data is described in more detail.

[0055] Once each image pair is generated, the image pair may be assigned to either the training image pair 306 or the testing image pair 308. In an embodiment, the image pairs may be randomly assigned to the training image pair 306 or the testing image pair 308 in a pre-established ratio. For example, the image pairs may be randomly assigned to the training image pair 306 or the testing image pair 308 such that 90% of the generated image pairs are assigned to the training image pair 306 and 10% of the generated image pairs are assigned to the testing image pair 308. Alternatively, the image pairs may be randomly assigned to the training image pair 306 or the testing image pair 308 such that 85% of the generated image pairs are assigned to the training image pair 306 and 15% of the generated image pairs are assigned to the testing image pair 308. It should be understood that the examples provided herein are for illustrative purposes and that the image pairs may be assigned to the training image pair 306 dataset or the testing image pair 308 dataset via different procedures and / or in different ratios without departing from the scope of the present disclosure.

[0056] The artifact removal neural network training system 300 may include a validator 320 that validates the performance of the artifact removal neural network 302 against the test image pairs 308. The validator 320 may take as input the partially trained artifact removal neural network 302 and the dataset of test image pairs 308, and may output an assessment of the performance of the partially trained artifact removal neural network 302 based on the dataset of test image pairs 308.

[0057] Once the artifact removal neural network 302 has been validated, the trained artifact removal neural network 322 (e.g., the validated artifact removal neural network 302) can be used to generate a set of artifact-reduced images 334 from a set of acquired (e.g., real) MR imaging data 332 (e.g., real MR images). The MR imaging data 332 may be undersampled data having lower quality due to artifacts. For example, the MR imaging data 332 may be acquired by an MR imaging device 330, which may be Figure 2 The trained artifact removal neural network 322 may be stored in an inference module 321 of an image processing system (e.g., Figure 2 within the inference module 212).

[0058] Now turn to Figure 4 , a flowchart illustrating a method 400 for training an artifact removal neural network is shown. According to an exemplary embodiment, the artifact removal neural network may be Figure 3 A non-limiting example of an artifact removal neural network 302 of an artifact removal neural network training system 300. The method 400 may be performed by, for example, Figure 2 In an embodiment, some operations of method 400 may be stored in a non-transitory memory of the image processing system (e.g., stored in a memory such as a memory such as a memory such as a memory such as a computer). Figure 2 The image processing system 202 includes a training module 210 and a training module 211 of the image processing system 202, and a processor of the image processing system (e.g., Figure 2 The artifact removal neural network may be trained based on training data including one or more sets of image pairs. As described below, each image pair in the one or more sets of image pairs may include a high-quality simulated MR image and a corresponding undersampled simulated MR image. In some embodiments, the one or more sets of image pairs may be stored in a medical image dataset (such as a medical image dataset of the image processing system) Figure 2 In addition, it should be understood that although the method 400 is described herein with respect to MRI imaging, MR imaging, and simulated MR imaging, other imaging modalities are possible without departing from the scope of the present disclosure.

[0059] Method 400 begins at 402, where method 400 includes generating an image that simulates motion, contrast, and phase dynamics, such as Figure 5 Further described. In short, in some examples, simulated motion can include simulating various motion phases of periodic or instantaneous motion types. Simulating contrast can include generating images at various signal intensities. Simulating phase contrast dynamics can include simulating phase changes in images. The simulated images can be based on natural images, which can provide a more robust image dataset for training.

[0060] At 404, method 400 includes generating a dataset of training image pairs based on the simulated images. Each training pair may include a target high-quality simulated image (e.g., a target simulated MR image) and a corresponding version of the simulated image that is undersampled (e.g., an input simulated MR image), where high quality herein means little to no artifacts (e.g., a fully sampled image). Thus, generating the dataset of training image pairs may include generating a high-quality simulated MR image as a target, as shown at 406, and generating an undersampled MR image as an input (e.g., a corresponding undersampled version of the high-quality simulated MR image), as shown at 408.

[0061] At 410, method 400 includes training an artifact removal neural network based on the training pairs. More specifically, training the artifact removal neural network based on the image pairs includes training the artifact removal neural network to learn to map an undersampled image (e.g., an undersampled simulated MR image) to a high quality image. In some embodiments, the artifact removal neural network may include a generative neural network. In some embodiments, the artifact removal neural network may include a generative neural network having a U-net architecture. In some embodiments, the artifact removal neural network may include one or more convolutional layers, which in turn include one or more convolutional filters (e.g., a convolutional neural network architecture).

[0062] For example, as shown at 412, training an artifact removal neural network may include training an artifact removal neural network for multi-phase or multi-echo data. The multi-phase or multi-echo trained network may be applied to a joint reconstruction method that may utilize information in a multi-phase or multi-echo MR image to remove artifacts caused by undersampling. In such examples, the input and target output data may be formatted as xy phases (or echoes). A simulated multi-phase or multi-echo image may be an input matched to a target 1-1. In some examples, the output may be a selected phase or echo or quantitative map. Fig. 9 The architecture of such a neural network is further described.

[0063] In other examples, as shown at 414, training the artifact removal neural network may include training the artifact removal neural network for single phase data. Such a trained network may be an iterative network. Similar to a network trained with multiple phases or multiple echoes, training the single phase data network may include an input matched to the target simulated image 1-1. Training the network for single phase data may be based on the input matched to the target simulated image 1-1. Fig.10 The architecture described is similar to the architecture.

[0064] The convolutional filters of either architecture may include a plurality of weights, wherein the values ​​of these weights are learned during a training procedure. The convolutional filters may correspond to one or more visual features / patterns, thereby enabling the artifact removal neural network to recognize and extract features from the medical image. In other embodiments, the artifact removal neural network may not be a convolutional neural network, but a different type of neural network.

[0065] Training an artifact removal neural network based on image pairs may include iteratively inputting the input image of each training image pair into an input layer of the artifact removal neural network. In some embodiments, each pixel intensity value of the input image may be input into a different neuron of the input layer of the artifact removal neural network. The artifact removal neural network may map the input image to a corresponding target image by propagating the input image from the input layer, through one or more hidden layers, until it reaches the output layer of the artifact removal neural network. In some embodiments, the output of the artifact removal neural network may be an image with a reduced amount of under-sampling directional artifacts and thus improved image quality compared to an under-sampled MR image input using a conventional reconstruction method. In addition, deploying an artifact removal neural network trained to reduce artifacts may reduce processing and / or computational requirements compared to, for example, advanced joint reconstruction methods using temporal information. As a result, scan time may also be reduced.

[0066] The artifact removal neural network may be configured to iteratively adjust one or more of a plurality of weights of the artifact removal neural network based on an evaluation of the difference between an input image and a target image included in each image pair of the training image pairs so as to minimize a loss function. In one embodiment, the loss function is a mean absolute error (MAE) loss function, in which the difference between the input image and the target image is compared and summed on a pixel-by-pixel basis. In another embodiment, the loss function may be a structural similarity index (SSIM) loss function. In other embodiments, the loss function may be a minimum loss function or a Wasserstein loss function. It should be understood that the examples provided herein are for illustrative purposes and that other types of loss functions may be used without departing from the scope of the present disclosure.

[0067] The weights and biases of the artifact removal neural network may be adjusted based on the difference between the output image of the relevant image pair and the target (e.g., reference) image. The difference (or loss) as determined by the loss function may be back-propagated through the neural learning network to update the weights (and biases) of the convolutional layers. In some embodiments, back-propagation of the loss may occur according to a gradient descent algorithm, wherein the gradient (first-order derivative or an approximation of the first-order derivative) of the loss function is determined for each weight and bias of the deep neural network. Each weight (and bias) of the artifact removal neural network is then updated by adding the negative of the product of the gradient determined (or approximated) for the weight (or bias) to a predetermined step size. The weights and biases may be repeatedly updated until the weights and biases of the artifact removal neural network converge, or the rate of change of the weights and / or biases of the deep neural network is below a threshold for each iteration of weight adjustment.

[0068] To avoid overfitting, the training of the artifact removal neural network can be periodically interrupted to verify the performance of the artifact removal neural network relative to the test image pair. In an embodiment, the training of the artifact removal neural network can be terminated when the performance of the artifact removal neural network relative to the test image pair converges (e.g., when the error rate on the test set converges to a threshold of a minimum value or converges to within a threshold of a minimum value). In this manner, the artifact removal neural network can be trained to generate a reconstruction of the input image, wherein the reconstruction of the input image includes fewer artifacts than the input image.

[0069] In some embodiments, the evaluation of the performance of the artifact removal neural network may include a combination of a minimum error rate and a quality assessment, or a different function of the minimum error rate achieved on each of the test image pairs and / or one or more quality assessments, or another factor for evaluating the performance of the artifact removal neural network. It should be understood that the examples provided herein are for illustrative purposes, and other loss functions, error rates, quality assessments, and / or performance assessments may be included without departing from the scope of the present disclosure.

[0070] Reference now Figure 5 , shows a method for generating training data to train an artifact removal neural network (such as Figure 3 Flowchart of method 500 for training an artifact removal neural network 302 of an artifact removal neural network training system 300. The training data may include various types of reference images, where the reference images have a higher resolution and / or contrast-to-noise ratio (CNR) than the undersampled MR images. Portions of method 500 may be performed by, for example, Figure 2 In an embodiment, some operations of method 500 may be stored in a non-transitory memory of the image processing system (e.g., stored in a memory such as a memory such as a memory such as a memory such as a computer). Figure 2The image processing system 202 includes a training module 210 and a training module 211 of the image processing system 202, and a processor of the image processing system (e.g., Figure 2 The image processing system 202 may be executed by the processor 204 of the image processing system 202.

[0071] At 502, method 500 includes generating a simulated (e.g., synthetic) image from an RGB image. The RGB image may be a reference image, a natural image, etc., which is not an image of a subject acquired specifically using a medical imaging system. Generating a simulated complex-valued image from the RGB image may include creating a randomized phase using the RGB channels and simulating linear and / or nonlinear phases. White noise or colored noise may be added to the simulated complex-valued image.

[0072] At 504, method 500 includes simulating motion in the simulated image. In some examples, the simulated motion may include generating a randomized motion field to distort the image and create images of different motion phases. As previously discussed, medical imaging data such as MR imaging data often undergo periodic motion due to processes such as breathing, bowel movements, etc. This periodic motion makes 1-1 matching of undersampled MR data and fully sampled MR data difficult. The simulated image may include simulated motion that simulates the motion phase seen in such imaging data. Therefore, the training data may include motion data as seen in the MR imaging data. The motion field may compromise rigid and / or non-rigid motion. The motion field may be generated by simulating motion (such as random rotation, random shift, periodic respiratory motion, etc.). The motion field may also be derived from MR or other modality images (such as CT, ultrasound, video, etc.). Multiple randomized motion fields may also be generated and applied periodically to simulate periodic motion. Images of different motion phases are then generated by distorting the reference image using the simulated motion field. In some examples, raw k-space data may be generated from the distorted image by simulating MR acquisition (such as radial, propeller, spiral acquisition, etc.). Simulated MR images of different motion phases can be generated from the simulated raw k-space data with motion.

[0073] Temporary turn Fig. 7A , shows examples of simulated MR images of various motion phases. A first image 702 may be warped to a first motion phase of a randomized motion field. A second image 704 may be warped to a second motion phase of a randomized motion field. A third image 706 may be warped to a third motion phase of a randomized motion field. In some examples, each of the motion phases may replicate a known periodic motion phase. In this way, the simulated images may include data of various motion phases to allow an artifact removal neural network to be trained to recognize such motion phases, and thus the trained artifact removal neural network may be able to remove artifacts from undersampled MR images that include motion.

[0074] Return to Figure 5At 506, method 500 includes simulating contrast in the simulated image. In some examples, simulating contrast may include simulating signal evaluation and modulating signal intensity in each RGB channel. The signal evaluation may be a contrast enhancement curve. In some examples, MR imaging data may be acquired for a contrast enhancement protocol, whereby a contrast agent is injected into a subject and taken up by various tissues in different ways, which defines different signal intensities in the MR image. In multi-phase and / or multi-echo MR imaging, imaging data may be acquired at various time phases and / or echoes, which produces different levels of contrast (e.g., different contrast phases) for each phase / echo. Due to the different contrast during acquisition, each phase or echo may provide visualization of a certain type of tissue (e.g., fat versus water). Simulating such different contrasts may provide simulated images of different contrast levels (e.g., with different signal intensities), as may be seen in multi-phase, multi-echo, and / or other contrast-enhanced MR imaging data. Similar to simulating motion, simulation of different contrasts may enable training of an artifact removal neural network using input images of different contrast levels and therefore at different phases or echoes. In some examples, the contrast phase may be simulated in the same set of images for which motion is simulated. In other examples, the contrast phase may be simulated for a new set of images that is different from those for which motion is simulated.

[0075] Temporary turn Figure 7B , examples of simulated images of different phases or echoes (e.g., having different contrast phases) are shown. A first image 710 may be simulated to replicate or estimate a first contrast phase or echo. A second image 712 may be simulated to replicate or estimate a second contrast phase or echo. A third image 714 may be simulated to replicate or estimate a third contrast phase or echo. In some examples, each of the phases or echoes may replicate a known phase or echo used in MR imaging. In this way, the simulated images may include data for various phases and / or echoes to allow an artifact removal neural network to be trained to recognize such different contrast levels, and thus the trained artifact removal neural network may be able to remove artifacts from undersampled MR images that include different contrast levels from different phases and / or echoes.

[0076] Back to Figure 5At 508, method 500 includes simulating phase contrast dynamics in the simulated image. In some examples, simulating phase contrast dynamics may include simulating phase changes with and / or without signal changes in the simulated image. In some MR or other modality applications, certain protocols utilize phase and / or echo changes, such as susceptibility weighted imaging (e.g., susceptibility weighted angiography), fast spoiled gradient echo imaging, 4D blood flow imaging, etc. Simulating such contrast dynamics in the simulated image may allow an artifact removal neural network to be trained based on input images that include such data. In some examples, phase contrast dynamics may be simulated in the same set of images in which motion and / or contrast phases are simulated. In other examples, phase contrast dynamics may be simulated for new sets of images that are different from those sets of images in which motion and / or contrast phases are simulated. In this manner, artifacts of various types of imaging data may be removed by an artifact removal neural network that is trained based on training images generated from various simulated images.

[0077] At 510, method 500 includes generating a training image based on the simulated image. Figure 4 The training images may include an input image and a target image. The simulated image including motion data, contrast data and phase contrast dynamic data may be a high-quality simulated MR image. The high-quality simulated MR image may be a target image for the training image. The input image may be a corresponding undersampled version of the simulated MR image, whereby the image is undersampled in k-space. The training image sets may each include a 1-1 matched input image and a target image.

[0078] In some examples, a first training image set may be based on images that simulate motion, but do not include images that simulate contrast at various phases. The first training image set may be used to train a first neural network, which may be used, for example, for single-phase MR imaging. In other examples, a second training image set may be based on images that simulate contrast at various phases, but do not include images that simulate motion. The second training image set may be used to train a second neural network, which may be used, for example, for MR imaging of low-motion or no-motion areas of the body (e.g., brain MRI, joint MRI, etc.). In yet another example, a third training image set may be based on images that simulate both motion and contrast at various phases. The third training image set may be used to train a third neural network, which may be deployed for MR imaging data that includes periodic motion and is multi-phase (e.g., multi-phase abdominal MRI). It should be understood that these are non-limiting examples, and other training image sets may be generated using a combination of motion, contrast phase, and phase-contrast dynamics. In this way, simulated images may be generated based on the intended application of the neural network to be trained using the images.

[0079] Due to motion and / or contrast phase changes in the acquired MR data, training data based on real MR images is not feasible. The method described herein provides for the generation of training data that simulates motion and contrast phase changes to allow the use of relevant training images to train a neural network. In this way, images that include under-sampling directional artifacts can be de-artifacted via a trained neural network, thereby reducing the computational power required for the reconstruction method.

[0080] Reference now Figure 6 , shows the use of deploying Figure 3 Flow chart of a method 600 of using an artifact removal neural network such as the artifact removal neural network 302 to reduce artifacts in real MR images. The method 600 may be performed by, for example, Figure 2 Some operations of method 600 may be stored in a non-volatile memory of the image processing system (e.g., stored in a memory device). Figure 2 ) in the inference module 212 of the image processing system 202 and by a processor of the image processing system (e.g., Figure 2 In various embodiments, the image processing system 202 may be executed by the processor 204 of the image processing system 202. Figure 5 The synthetic training data generated by the method 500 is as described above with reference to Figure 4 The denoising neural network is trained as described in method 400. Specifically, the training data may include one or more sets of image pairs, the one or more sets of image pairs including simulated MR images, the simulated MR images including undersampled input images and fully sampled or otherwise high-quality target images.

[0081] Method 600 begins at 602, where method 600 includes receiving MR imaging data acquired from a subject. As an example, the MR imaging data may include an MR image of the subject. For example, a Figure 3 MR images may be acquired by an MRI device such as an MR imaging device 330 or an MRI scanner 236 of the image processing system 202. The acquired MR images may have the same region of interest and / or may include the same set of anatomical structures as the set of RGB images (e.g., reference images, natural images, etc.) and simulated images on which the training of the artifact removal neural network is based. In some embodiments, the subject of the acquired MR image may be similar to the subject of the set of RGB images. For example, the subject may be a child, wherein the acquired MR image may be input into a first artifact removal neural network trained based on the child's RGB images; the subject may be a female, wherein the acquired MR image may be input into a second artifact removal neural network trained based on a female reference image; the subject may be a male, wherein the acquired MR image may be input into a third artifact removal neural network trained based on a male reference image; and so on.

[0082] At 604, the acquired MR imaging data is input into the trained artifact removal neural network. In various embodiments, inputting the acquired MR imaging data into the trained artifact removal neural network includes inputting the image data of each pixel of the acquired MR image into a corresponding node of the input layer of the denoising neural network. The value of the image data can be multiplied by the weight at the corresponding node and propagated to the output layer of the denoising neural network through various hidden layers (e.g., convolutional layers). The output layer can include a node corresponding to each pixel of the output MR image, wherein the output MR image is based on the image data output by each node. The output image can have fewer artifacts than the input image, wherein the amount of artifacts of the input image is reduced or removed by the trained artifact removal neural network.

[0083] At 606, method 600 includes, at a display screen (eg, Figure 2 The artifact-reduced image output by the trained denoising neural network is displayed on a display device 234 of the image processing system. During an examination of the subject, the artifact-reduced image can be displayed in real time on the display screen so that an operator of the image processing system (e.g., a caregiver) can view the artifact-reduced image during the examination. The artifact-reduced image can also be output to a storage device or a picture archiving and communication system (PACS) for subsequent retrieval and / or remote viewing. For example, the artifact-reduced image can be used to diagnose a condition of the subject. By reducing the amount of artifacts in the acquired MR images, the caregiver can see the anatomical features of the subject more clearly, thereby making it easier to diagnose the condition.

[0084] Figure 8 An exemplary input MR image, a target MR image, and an output MR image generated using a trained artifact removal neural network (such as the artifact removal neural network described herein) are shown. The input image 800 can be an undersampled MR image acquired of a patient. Specifically, the input image 800 can be an abdominal MR image. The input image 800 can have a first artifact amount (e.g., undersampled directional artifacts). The target image 804 can be a fully sampled (e.g., high quality) version of the undersampled MR image with a second artifact amount. The second artifact amount can be less than the first artifact amount of the input image 800.

[0085] The output image 802 may be output by the trained artifact removal neural network based on the input image 800 (e.g., where the input image 800 is input into the trained artifact removal neural network to generate the output image 802). Figure 3 and FIG. 7A to FIG. 7B as described and as Figure 5The artifact removal neural network is trained based on the simulated MR images generated as described in the method 500. In other words, the artifact removal neural network is trained based on an image pair including a target MR image with a small amount of artifacts (e.g., the simulated MR image 312) and its corresponding undersampled version (e.g., the undersampled simulated MR image 316). The output image 802 may have a third artifact amount. The third artifact amount may be more closely similar to the second artifact amount of the target image 804 than the first artifact amount. In this way, the output image 802 may have a reduced artifact amount compared to the input image 800.

[0086] refer to Fig. 9 , shows an architecture diagram of the first neural network 900. When trained for multi-phase or multi-echo MR imaging, the first neural network 900 may be an artifact removal neural network as described herein. The first neural network 900 may be used to extract the artifact removal information from the image obtained by, for example, Figure 1 The first neural network 900 can be used in a system such as an MRI device 10 to reduce artifacts in an undersampled MR image acquired by an MR imaging system. Figure 3 The artifact removal neural network training system 300 and other artifact removal neural network training systems are trained.

[0087] The first neural network 900 may have an input layer, a plurality of convolutional layers, and an output layer. In the input layer, an input image 902 may be input into the first neural network 900 and mapped to a feature set. The input image 902 may be undersampled multi-phase or multi-echo data. The first neural network 900 may include a series of mappings from the input image 902 to a plurality of iterative images (e.g., trainable blocks). For example, the input image 902 may be mapped to a first iterative image set 904. The first iterative image set 904 may be mapped to a second iterative image set 906 and a third iterative image set 908. The second iterative image set 906 may also be mapped to a third iterative image set 908, as shown by a solid arrow 920. The third iterative image set 908 may be mapped to an output image 910. In some examples, the images may also have residual connections, as indicated by dashed lines in the figure. For example, a first residual connection 912 may exist between the input image 902 and the output image 910, a second residual connection 914 may exist between each of the iterative images of the first iterative image set 904, a third residual connection 916 may exist between each of the iterative images of the second iterative image set 906, and a fourth residual connection 918 may exist between each of the iterative images of the third iterative image set 908. Each of the iterative images may be connected to a previous image and a next image, wherein each iterative image receives input from a previous iterative image and transforms / maps the received input to an output to produce a next iterative image. In some examples, the convolution kernel size may increase or decrease between iterative images, and in other examples, the convolution kernel size may remain unchanged between iterative images.

[0088] The output image 910 may be an estimated image based on the input image 902 and multiple iterated images based on the input layer and the output layer of the first neural network 900. The output image 910 may approximate a reference image based on the training of the first neural network 900, where the reference image is a target image of a pair of training images. Thus, the first neural network 900 illustrates a mapping of the transformations that occur when the input image 902 is propagated through the layers of the network.

[0089] Training the first neural network 900 based on the training image pairs may include iteratively inputting the input image as described herein into an input layer. In some embodiments, each pixel intensity value of the input image may be input into a different neuron of the input layer of the first neural network 900. In some examples, the output of the first neural network 900 includes a 2D matrix of values, where each value corresponds to a different intensity of a pixel of the input image, and where the different intensity of each pixel of the output image generates a reconstruction of the input image, where an amount of artifacts in one or more regions of the output image is lower than an amount of artifacts in one or more regions of the input image.

[0090] like Figure 4 As described above, the weights (and biases) of the convolutional layers in the first neural network 900 are learned during training. During training, the difference between the output image 910 and the reference image (e.g., the benchmark data included in the corresponding training pair) can be back-propagated through the layers of the first neural network 900 to update the weights (and biases) of the convolutional layers according to the loss function. As described above, the first neural network 900 can be trained based on multiple data training pairs.

[0091] Steering Fig.10 , shows an architecture diagram of the second neural network 1000. When trained for single-phase MR imaging, the second neural network 1000 may be an artifact removal neural network as described herein. The second neural network 1000 may be used to extract the artifacts from the image obtained by, for example, Figure 1 The second neural network 1000 can be used in a system such as an MRI device 10 to reduce artifacts in an undersampled MR image acquired by an MR imaging system such as an MRI device 10. Figure 3 The artifact removal neural network training system 300 and other artifact removal neural network training systems are trained.

[0092] The second neural network 1000 may have an input layer, a plurality of convolutional layers, and an output layer. In the input layer, an input image 1004 may be input into the second neural network 1000 and mapped to a feature set. The input image 1004 may be undersampled data 1002. The second neural network 1000 may include a series of mappings from the input image 1004 to a plurality of iteration images 1006. Each iteration image in the iteration images may be connected to a previous image and a next image, wherein each iteration image receives an input from a previous iteration image and transforms / maps the received input to an output to generate a next iteration image. In addition, a residual connection 1012 may exist between non-adjacent iteration images 1006. The nth iteration image may be mapped to an output image 1008. The output image 1008 may be an estimated image based on the input image 1004 and a plurality of iteration images 1006 based on the input layer and the output layer of the second neural network 1000. The output image 1008 may approximate a reference image 1010, where the reference image 1010 is a target image of a pair of training images, based on the training of the second neural network 1000. Thus, the second neural network 1000 illustrates a mapping of the transformations that occur when the input image 1004 is propagated through the layers of the network.

[0093] Similar to first neural network 900, training second neural network 1000 based on training image pairs may include iteratively inputting an input image as described herein into an input layer. In some embodiments, each pixel intensity value of the input image may be input into a different neuron of the input layer of second neural network 1000. In some examples, the output of second neural network 1000 includes a 2D matrix of values, where each value corresponds to a different intensity of a pixel of the input image, and where the different intensity of each pixel of the output image generates a reconstruction of the input image, where an amount of artifacts in one or more regions of the output image is lower than an amount of artifacts in one or more regions of the input image.

[0094] like Figure 4 As described above, the weights (and biases) of the convolutional layers in the second neural network 1000 are learned during training. During training, the difference between the output image 1008 and the reference image (e.g., the benchmark data included in the corresponding training pair) can be back-propagated through the layers of the second neural network 1000 to update the weights (and biases) of the convolutional layers according to the loss function. As described above, the second neural network 1000 can be trained based on multiple data training pairs.

[0095] Thus, described herein are systems and methods for improving the performance of an artifact reduction or denoising neural network by training the neural network based on simulated (e.g., synthetic) MR images rather than real MR images. By using simulated MR images rather than real MR images, the neural network can more easily learn to distinguish artifacts from anatomical features in the image. Unlike other methods based on generating simulated MR images using machine learning models, the simulated MR images described herein can be created based on RGB channels and factors present in real MR images, such as motion, contrast of different phases, and phase dynamics can be simulated in the simulated MR images. The simulated MR images can be generated from RGB channels with a higher resolution or CNR than the real MR images, and can be undersampled in k-space to generate their corresponding undersampled versions. In this way, a higher quality output MR image can be generated based on undersampled multi-phase or multi-echo input MR images because the artifact removal neural network is trained to detect and remove artifacts from such undersampled images. In addition, compared to methods for mitigating artifacts in undersampled MR imaging data (such as advanced joint or mutual reconstruction methods), the trained artifact removal neural network can reduce the computational and processing power required to remove artifacts in MR images. Since the reconstruction method does not require artifact removal, processing requirements are reduced.

[0096] The present disclosure also provides support for a method for an image processing system, the method comprising: generating a plurality of simulated images; generating a set of training image pairs based on the plurality of simulated images; training an artifact removal neural network using the set of training image pairs; and generating an output of the artifact removal neural network based on an input acquired image, wherein generating a plurality of simulated images comprises generating an image from an RGB image, simulating motion in the simulated image, simulating contrast in the simulated image, and simulating phase contrast dynamics in the simulated image. In a first example of the method, each training image pair in the set of training image pairs comprises an input image and a target image. In a second example of the method (optionally including the first example), the input image is an undersampled version of the simulated image, and the target image is a high-quality simulated image. In a third example of the method (optionally including one or both of the first and second examples), the input image and the target image of a given training image pair are 1-1 matched. In a fourth example of the method (optionally including one or more or each of the first to third examples), the plurality of simulated images are simulated magnetic resonance (MR) images. In a fifth example of the method (optionally including one or more or each of the first to fourth examples), the input acquisition image is an MR image acquired by an MRI scanner. In a sixth example of the method (optionally including one or more or each of the first to fifth examples), the MR image acquired by the MRI scanner is undersampled and includes undersampling directional artifacts. In a seventh example of the method (optionally including one or more or each of the first to sixth examples), the method further includes: testing the trained artifact removal neural network using a test image pair generated from a simulated image. In an eighth example of the method (optionally including one or more or each of the first to seventh examples), the artifact removal neural network is a multi-phase network or a multi-echo network. In a ninth example of the method (optionally including one or more or each of the first to eighth examples), the artifact removal neural network is a single-phase network or a single-echo network.

[0097] The present disclosure also provides support for an image processing system, the image processing system comprising: a processor capable of being communicatively coupled to a non-transitory memory storing a neural network, the memory comprising instructions that, when executed, cause the processor to: receive a plurality of simulated MR images, each simulated MR image being generated by simulating motion, contrast, and phase contrast dynamics; for each simulated MR image, generate an undersampled version of the simulated MR image; create a corresponding plurality of image pairs, each image pair comprising a simulated MR image as a target reference image and a corresponding undersampled version of the simulated MR image as an input image; train the neural network using the image pairs; deploy the trained neural network to generate an artifact-reduced image based on an MR image acquired from a scanned subject; and display the artifact-reduced image on a display device of the image processing system. In a first example of the system, the MR image acquired from the scanned subject is an undersampled multi-phase / multi-echo image. In a second example of the system (optionally including the first example), each of the image pairs is 1-1 matched. In a third example of the system (optionally including one or both of the first and second examples), generating the undersampled version of the simulated MR image includes undersampling the simulated MR image in k-space.

[0098] The present disclosure also provides support for a method for creating a simulated magnetic resonance (MR) image, the simulated magnetic resonance (MR) image being used to train a model to reduce the amount of artifacts in an acquired MR image, the method comprising: obtaining a set of reference images; simulating a motion phase in one or more of the reference images; simulating a contrast phase in one or more of the reference images; simulating phase contrast dynamics in one or more of the reference images; generating a simulated MR image of the reference image based on the simulated motion phase, contrast phase, and phase contrast dynamics. In a first example of the method, the reference image has a higher resolution than the acquired MR image. In a second example of the method (optionally including the first example), the simulated MR image is undersampled in k-space to generate a lower resolution version of the simulated MR image. In a third example of the method (optionally including one or both of the first and second examples), the method further comprises: generating a training image pair, wherein the simulated MR image is a target, and the lower resolution version of the simulated MR image is an input, and wherein the input and the target are in xy phase format. In a fourth example of the method (optionally including one or more or each of the first to third examples), the model is a neural network model, and the method further includes training the neural network model based on training data including training image pairs. In a fifth example of the method (optionally including one or more or each of the first to fourth examples), the method further includes: deploying the neural network model to generate an artifact-reduced version of the input MR image.

[0099] When introducing the elements of various embodiments of the present disclosure, the articles "one", "a kind of" and "the" are intended to mean that there are one or more such elements. The terms "first", "second", etc. do not represent any order, amount or importance, but are used to distinguish one element from another element. The terms "include", "comprises", and "have" are intended to be inclusive, and mean that additional elements may also exist in addition to the listed elements. As used herein, the terms "connected to", "coupled to", etc., an object (e.g., material, element, structure, member, etc.) may be connected to or coupled to another object, regardless of whether the one object is directly connected or coupled to another object, or whether there are one or more intervening objects between the one object and another object. In addition, it should be understood that reference to "an embodiment" or "embodiment" of the present disclosure is not intended to be interpreted as excluding the existence of additional embodiments that are also combined with the cited features.

[0100] In addition to any previously indicated modifications, those skilled in the art may devise numerous other variations and alternative arrangements without departing from the spirit and scope of the present specification, and the appended claims are intended to cover such modifications and arrangements. Therefore, although the information has been described in detail and in detail as described above in conjunction with what are currently considered to be the most practical and preferred aspects, it will be apparent to those of ordinary skill in the art that many modifications, including but not limited to form, function, mode of operation, and use, may be made without departing from the principles and concepts set forth herein. Likewise, as used herein, the examples and embodiments are intended to be illustrative in all respects only and should not be construed as limiting in any way.

Claims

1. A method for an image processing system, comprising: generating a plurality of simulated images (402); generating a set of training image pairs based on the plurality of simulated images (404); Training an artifact removal neural network using the set of training image pairs (410); as well as Generating an output of the artifact removal neural network based on the input acquired images (606), wherein generating the plurality of simulated images includes generating images from RGB images (502), simulating motion in the simulated images (504), simulating contrast in the simulated images (506), and simulating phase contrast dynamics in the simulated images (508).

2. The method of claim 1, wherein each training image pair in the set of training image pairs comprises an input image (408) and a target image (406).

3. The method of claim 2, wherein the input image is an undersampled version of the simulated image (408) and the target image is a high quality simulated image (406).

4. The method of claim 3, wherein the input image and the target image of a given training image pair are 1-1 matched. The method of claim 1 , wherein the plurality of simulated images are simulated magnetic resonance (MR) images.

6. The method according to claim 1, wherein the input acquired image is an MR image acquired by an MRI scanner (10).

7. The method of claim 6, wherein the MR images acquired by the MRI scanner (10) are undersampled and include undersampling orientation artifacts.

8. The method of claim 1, further comprising testing a trained artifact removal neural network using a test image pair (308) generated from the simulated image.

9. The method of claim 1, wherein the artifact removal neural network is a multi-phase network or a multi-echo network.

10. The method of claim 1, wherein the artifact removal neural network is a single phase network (1000) or a single echo network (900).

11. An image processing system, comprising: A processor (204) communicatively coupled to a non-transitory memory (206) storing a neural network, the memory comprising instructions that, when executed, cause the processor to: receiving a plurality of simulated MR images (406), each simulated MR image being generated by simulating motion, contrast, and phase contrast dynamics; For each simulated MR image, generating an undersampled version of the simulated MR image (408); creating a corresponding plurality of image pairs (404), each image pair comprising a simulated MR image as a target reference image and a corresponding undersampled version of the simulated MR image as an input image; training the neural network using the image pairs (410); deploying the trained neural network to generate artifact-reduced images from MR images acquired from a scanned subject (600); and The artifact-reduced image is displayed on a display device of the image processing system (606).

12. The image processing system of claim 11, wherein the MR image acquired from the scanned subject is an undersampled multi-phase / multi-echo image (602).

13. The image processing system of claim 11, wherein each of the image pairs is 1-1 matched.

14. The image processing system of claim 11, wherein generating the undersampled version of the simulated MR image comprises undersampling the simulated MR image in k-space (408).

15. The image processing system of claim 11, wherein the simulated MR image has a higher resolution than the undersampled version of the simulated MR image.

Images