System and method for artifact reduction in magnetic resonance images

Abstract

A computer-implemented method (350) of reducing (356) artifacts in a multi-channel magnetic resonance (MR) image includes receiving (352) a plurality of MR image sets acquired by a radio frequency (RF) coil assembly (38) having a plurality of channels. Each MR image set includes a plurality of slices of MR images acquired by one of the plurality of channels. The method (350) further includes estimating (354) a plurality of artifact sets in the plurality of MR image sets by inputting the plurality of MR image sets into the neural network model (304). Each artifact set corresponds to one of the plurality of channels. The method (350) further includes reducing (356) artifacts in the plurality of MR image sets based on the estimated artifacts, deriving (358) an artifact-reduced MR image by combining the artifact-reduced MR images, and outputting (360) the artifact-reduced MR image.

Description

Background Art

[0001] The field of the present disclosure relates generally to systems and methods for medical imaging, and more particularly, to systems and methods for artifact reduction in magnetic resonance (MR) images.

[0002] Magnetic resonance imaging (MRI) has proven useful in the diagnosis of many diseases. MRI provides detailed images of soft tissue, abnormal tissue (such as tumors), and other structures that cannot be easily imaged by other imaging modalities such as computed tomography (CT). In addition, MRI operates without exposing the patient to ionizing radiation experienced in modalities such as CT and X-ray.

[0003] Artifacts in MR images affect the image quality of the images and may interfere with diagnosis.Known methods are disadvantageous in several respects and improvements are needed. Summary of the invention

[0004] In one aspect, a computer-implemented method of reducing artifacts in a multi-channel magnetic resonance (MR) image is provided. The method includes receiving a plurality of MR image sets of a volume in a subject. The plurality of MR image sets are acquired by a radio frequency (RF) coil assembly having a plurality of channels, and each MR image set includes a plurality of slices of MR images of the volume acquired by one of the plurality of channels. The method also includes estimating a plurality of artifact sets in the plurality of MR image sets by inputting the plurality of MR image sets into a neural network model, wherein the neural network model is configured to estimate artifacts in the MR images input into the neural network model. Each artifact set corresponds to the one of the plurality of channels. The method also includes: reducing artifacts in the plurality of MR image sets based on the plurality of estimated artifact sets; deriving an artifact-reduced MR image by combining the plurality of artifact-reduced MR image sets; and outputting the artifact-reduced MR image.

[0005] In another aspect, a computer-implemented method for reducing artifacts in MR images is provided. The method includes receiving one or more MR image sets of a volume in a subject. The one or more MR image sets are acquired by an RF coil assembly having one or more channels, and each MR image set includes multiple slices of MR images of the volume acquired by one of the one or more channels. The method also includes estimating one or more artifact sets in the one or more MR image sets by inputting the one or more MR image sets into a neural network model, wherein the neural network model is configured to estimate artifacts in the MR images input into the neural network model. Each artifact set corresponds to the one of the one or more channels. The method also includes outputting the one or more estimated artifact sets.

[0006] In yet another aspect, a computer-implemented method for reducing artifacts in a multi-channel MR image is provided. The method includes receiving a plurality of first MR image sets of a volume in a subject, wherein the plurality of first MR image sets are acquired by an RF coil assembly having a plurality of channels. The method also includes reducing artifacts in the plurality of first MR image sets based on a plurality of estimated artifact sets. The plurality of estimated artifact sets are estimated by inputting a plurality of second MR image sets into a neural network model, wherein the neural network model is configured to estimate artifacts in the MR images input into the neural network model. The plurality of second MR image sets are MR images of a volume acquired by the RF coil assembly. Each artifact set corresponds to the one of the plurality of channels. The method also includes deriving an artifact-reduced first MR image by combining the plurality of first MR image sets with artifact reduction, and outputting the artifact-reduced first MR image. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a schematic diagram of an exemplary magnetic resonance imaging (MRI) system.

[0008] Figure 2 is a schematic diagram showing a combination of images acquired by channels of a radio frequency (RF) coil assembly.

[0009] Figure 3A is an exemplary artifact reduction system.

[0010] Figure 3B is a flow chart of an exemplary method of reducing artifacts.

[0011] Figure 4 It is shown Figure 3B Schematic diagram of the workflow of an exemplary embodiment of the method shown in .

[0012] Figure 5 is a schematic diagram showing the use of a neural network model to estimate artifacts.

[0013] Fig. 6A is a graph of artifact indicators for a channel with two clusters.

[0014] Figure 6B is a graph of artifact indicators for a channel with three clusters.

[0015] Figure 7 is a comparison of images reconstructed with and without the systems and methods described herein.

[0016] Fig. 8A It is a schematic diagram of the neural network model.

[0017] Figure 8B yes Fig. 8A Schematic diagram of neurons in the neural network model shown.

[0018] Fig. 9 is a block diagram of an exemplary computing device.

[0019] Fig.10 is a block diagram of an exemplary server computing device. DETAILED DESCRIPTION

[0020] The present disclosure includes systems and methods for reducing artifacts in magnetic resonance (MR) images of a subject. As used herein, a subject is a human, an animal, or a phantom, or a part of a human, an animal, or a phantom, such as an organ or tissue. Artifacts are visual abnormalities in medical images that do not represent the anatomical structures or functions of the subject, unlike image signals that represent the anatomical structures and functions of the subject. Reducing artifacts using the systems and methods described herein is channel-dependent, where artifacts are more prominent in images acquired by some channels of a radio frequency (RF) coil assembly of an MR system than in images acquired by other channels of an RF coil assembly. Reducing or removing artifacts is collectively referred to as reducing artifacts. Methodological aspects will be apparent in part and will be discussed explicitly in part in the following description.

[0021] In magnetic resonance imaging (MRI), a subject is placed in a magnet. When the subject is in a magnetic field generated by a magnet, the magnetic moments of nuclei such as protons attempt to align with the magnetic field, but precess around the magnetic field in a random order at the Larmor frequency of the nuclei. The magnetic field of the magnet is called B0 and extends in the longitudinal or z direction. During the acquisition of MRI images, a magnetic field in the xy plane and close to the Larmor frequency (called the excitation field B1) is generated by a radio frequency (RF) coil and can be used to rotate or "tilt" the net magnetic moment Mz of the nucleus from the z direction toward the transverse or xy plane. After the excitation signal B1 terminates, the nucleus emits a signal, which is called an MR signal. In order to use the MR signals to generate an image of the subject, magnetic field gradient pulses (Gx, Gy, and Gz) are used. Gradient pulses are used to scan through k-space, the space or distance of spatial frequency in reverse. There is a Fourier relationship between the acquired MR signals and the image of the subject, so the image of the subject can be derived by reconstructing the MR signals.

[0022] Figure 1A schematic diagram of an exemplary MRI system 10 is shown. In an exemplary embodiment, the MRI system 10 includes a workstation 12 having a display 14 and a keyboard 16. The workstation 12 includes a processor 18, such as a commercially available programmable machine running a commercially available operating system. The workstation 12 provides an operator interface that allows a scan protocol to be input into the MRI system 10. The workstation 12 is coupled to a pulse sequence server 20, a data acquisition server 22, a data processing server 24, and a data storage server 26. The workstation 12 and each of the servers 20, 22, 24, and 26 communicate with each other.

[0023] In an exemplary embodiment, the pulse sequence server 20 operates a gradient system 28 and a radio frequency ("RF") system 30 in response to instructions downloaded from the workstation 12. The instructions are used to generate gradient waveforms and RF waveforms in an MR pulse sequence. The RF coil 38 and the gradient coil assembly 32 are used to perform the prescribed MR pulse sequence. The RF coil 38 is shown as a whole body RF coil. The RF coil 38 can also be a local coil that can be placed near the anatomical structure to be imaged, or a coil array including multiple coils.

[0024] In the exemplary embodiment, gradient waveforms for performing a prescribed scan are generated and applied to the gradient system 28, which excites the gradient coils in the gradient coil assembly 32 to generate magnetic field gradients G for position encoding the MR signals. x , G y and G z The gradient coil assembly 32 forms part of a magnet assembly 34 , which also includes a polarizing magnet 36 and an RF coil 38 .

[0025] In an exemplary embodiment, the RF system 30 includes an RF transmitter for generating RF pulses used in an MR pulse sequence. The RF transmitter responds to the scanning scheme and direction from the pulse sequence server 20 to generate RF pulses with a desired frequency, phase, and pulse amplitude waveform. The generated RF pulses can be applied to the RF coil 38 by the RF system 30. The response MR signal detected by the RF coil 38 is received by the RF system 30 and amplified, demodulated, filtered, and digitized under the instruction of the command generated by the pulse sequence server 20. The RF coil 38 is described as a transmitting and receiving coil, so that the RF coil 38 transmits RF pulses and detects MR signals. In one embodiment, the MRI system 10 may include a transmitting RF coil that transmits RF pulses and a separate receiving coil that detects MR signals. The transmission channel of the RF system 30 can be connected to the RF transmitting coil, and the receiver channel can be connected to the separate RF receiving coil. Typically, the transmission channel is connected to the entire body RF coil 38, and each receiver section is connected to a separate local RF coil.

[0026] In an exemplary embodiment, the RF system 30 also includes one or more RF receiver channels. Each RF receiver channel includes an RF amplifier that amplifies the MR signals received by the RF coil 38 to which the channel is connected; and a detector that detects and digitizes the I quadrature component and the Q quadrature component of the received MR signal. The magnitude of the received MR signal can then be determined as the square root of the sum of the squares of the I component and the Q component, as shown in the following equation (1):

[0027]

[0028] And the phase of the received MR signal can also be determined as shown in the following equation (2):

[0029]

[0030] In an exemplary embodiment, the digitized MR signal samples generated by the RF system 30 are received by the data acquisition server 22. The data acquisition server 22 can operate in response to instructions downloaded from the workstation 12 to receive real-time MR data and provide a buffer memory so that no data is lost due to data overflow. In some scans, the data acquisition server 22 only passes the acquired MR data to the data processing server 24. However, in scans where information derived from the acquired MR data is needed to control further execution of the scan, the data acquisition server 22 is programmed to generate the required information and transmit it to the pulse sequence server 20. For example, during a pre-scan, MR data is acquired and used to calibrate the pulse sequence executed by the pulse sequence server 20. In addition, navigator signals can be acquired during the scan and used to adjust the operating parameters of the RF system 30 or the gradient system 28, or to control the view order in which k-space is sampled.

[0031] In an exemplary embodiment, the data processing server 24 receives MR data from the data acquisition server 22 and processes the MR data according to instructions downloaded from the workstation 12. Such processing may include, for example, Fourier transforming the raw k-space MR data to produce a two-dimensional image or a three-dimensional image, applying filters to the reconstructed image, performing back-projection image reconstruction on the acquired MR data, generating functional MR images, and calculating motion images or flow images.

[0032] In an exemplary embodiment, the image reconstructed by the data processing server 24 is transmitted back to the workstation 12 and stored there. In some embodiments, the real-time image is stored in a database memory cache ( Figure 1In the example shown in Figure 1, the real-time image can be output from the database memory cache to the operator display 14 or a display 46 located near the magnet assembly 34 for use by the attending physician. Batch mode images or selected real-time images can be stored in a disk storage device 48 or a host database on the cloud. When such images have been reconstructed and transferred to the storage device, the data processing server 24 notifies the data storage server 26. The operator can use the workstation 12 to archive images, generate films, or send images to other facilities via the network.

[0033] Figure 2 The MR image 202 is shown to be formed based on MR signals acquired by the multi-channel RF coil assembly 38. The RF coil assembly 38 is a local RF coil. The multi-channel RF coil assembly includes multiple coils and multiple channels. A channel can correspond to one or more coils. The MR signals acquired by the channel 204 are used to reconstruct the MR image corresponding to the channel 204. As used herein, a channel may include a physical coil and / or a virtual coil. The signal from the virtual coil is generated based on the signal acquired by the physical coil. For example, the signal S from the virtual coil 虚拟 The user is the k-space signal S from the physical coil 物理 The conjugate symmetric signal of is expressed in the following equation.

[0034]

[0035] Where * represents the complex conjugate, and k represents the position in k-space. An MR image corresponding to a particular channel may be referred to as a channel image. When the RF coil assembly 38 is placed close to the subject 102 (see Figure 1 ), the coil can be closer to a specific anatomical structure of the subject than other coils. As a result, although the images 206 of the individual channels 204 or channel images 206 are abdominal images of the same slice of the subject, the channel images 206 are different for different channels. The output image 202 is a combined image of the channel images 206.

[0036] Radial acquisitions such as star pile or PROPELLER (periodically rotated overlapping parallel lines with enhanced reconstruction) are attractive in MR imaging such as free-breathing MR imaging because radial acquisitions are relatively insensitive to motion. In radial imaging or PROPELLER imaging, k-space is sampled along radial lines. However, radial acquisitions suffer from streak artifacts that impair image and diagnostic quality. Figure 2, images are acquired using radial acquisition. Output image 202 includes streak artifacts 208. Streak artifacts are more pronounced in the arm-down position than in the arm-up position because streak artifacts are caused by aliasing of relatively bright signals from anatomical structures (e.g., arm 210) outside the field of view (FOV) due to field inhomogeneities, high receiver sensitivity, gradient nonlinearity, transient motion, and incomplete fat suppression. For example, during imaging, arm 210 is close to bore 104 of magnet assembly 34 (see Figure 1 ) at the edge of the aperture 104, where the magnetic field is less uniform than at the center of the aperture 104. In addition, because the arm 210 is positioned close to the RF coil assembly, the signal-to-noise ratio (SNR) from the arm 210 is relatively high, and the signal from the arm 210 is aliased into the FOV and produces artifacts 208. Because the various coils are in different spatial relationships to the anatomical structure being imaged, the artifacts 208 are more significant in some channels than in other channels. The streak artifact is used as an example for illustrative purposes only. The systems and methods described herein can be applied to determine and reduce other artifacts, such as peripheral signal artifacts, motion artifacts, and flow artifacts. Peripheral signal artifacts are caused by stimulated echoes of signals outside the FOV. Motion artifacts and flow artifacts are caused by motion or flow, respectively. Streak artifacts, peripheral signal artifacts, motion artifacts, and flow artifacts are channel-dependent, where artifacts in some channels are more significant than artifacts in other channels, or artifacts in some locations are more significant than artifacts in other locations due to the spatial position of the coils. For example, because some coils are closer to the artifact source than other coils, the artifact is more noticeable in some coils / channels than in other coils / channels. If the RF coil assembly is single channel, the artifact at some locations is more noticeable than at other locations.

[0037] In known methods for removing channel-related artifacts, a low-pass filter is applied to the channel data, and the difference between the low-resolution channel image / sinogram and the high-resolution channel image / sinogram is calculated. These differences are used to estimate the artifacts in each channel. Channels with significant artifacts in the channel data are removed during image reconstruction. However, the estimation of artifacts is highly dependent on the selection of parameters of the low-pass filter. The application of low-pass filtering to estimate and reduce artifacts is based on the observation that streak artifacts have a relatively high spatial frequency. Edges in images also have a relatively high spatial frequency. The known methods therefore distort the estimation of streak artifacts and mistakenly remove or reduce image signals. In addition, the difference between the low-resolution channel data and the high-resolution channel data includes mixed signals (such as noise and truncation artifacts) that are not dependent on the channel, thereby reducing the accuracy of the estimation of channel-dependent artifacts. In addition, low-pass filtering does not reduce other channel-related artifacts such as motion artifacts or peripheral signal artifacts, because motion artifacts or peripheral signal artifacts do not necessarily have a relatively high spatial frequency. Other known methods for reducing streak artifacts do not address channel correlation. Instead, known methods attempt to identify a region of interest (ROI) comprising artifacts. Furthermore, known methods require manual intervention to extract the ROI, thereby increasing the complexity of the workflow and reducing the robustness of the known methods.

[0038] In known methods that use neural network models to reduce artifacts or noise, the artifacts or noise are not channel-related. Images from different channels have been combined before being analyzed by the neural network model. Therefore, known methods that use neural network models are not suitable for removing channel-related artifacts.

[0039] In contrast, in the systems and methods described herein, the accuracy of estimating and reducing channel-related artifacts is increased, thereby improving the diagnostic quality of images and the robustness of non-Cartesian imaging and Cartesian imaging. Unlike known methods, the systems and methods described herein do not rely on the assumption that artifacts have relatively high spatial frequencies. The systems and methods described herein use a neural network model trained to estimate channel-related artifacts to estimate and reduce artifacts. The systems and methods described herein are not limited to streak artifacts, and can be used to reduce other channel-related artifacts, such as peripheral signal artifacts, motion artifacts, and flow artifacts, thereby further improving image quality and diagnostic quality. The systems and methods described herein reduce artifacts that are relatively significant in certain imaging positions (such as the arm-down position), thereby relaxing restrictions on imaging positions. As a result, the comfort and cooperation of the subject are increased, and restrictions on scan time during imaging from uncomfortable positions are avoided. The systems and methods described herein enable free breathing by reducing streak artifacts in radial acquisitions, thereby improving the imaging experience of the subject.

[0040] Figure 3Ais a schematic diagram of an exemplary artifact reduction system 300. In an exemplary embodiment, the system 300 includes an artifact reduction computing device 302-1 configured to reduce artifacts in an MR image. The computing device 302-1 also includes a neural network model 304. The system 300 may include a second artifact reduction computing device 302-2. The second artifact reduction computing device 302-2 can be used to train the neural network model 304, and then the artifact reduction computing device 302-1 can use the trained neural network model 304. The second artifact reduction computing device 302-2 can be the same computing device as the artifact reduction computing device 302-1, so that the training and use of the neural network model 304 are performed on one computing device. Alternatively, the second artifact reduction computing device 302-2 can be a computing device separate from the artifact reduction computing device 302-1, so that the training and use of the neural network model 304 are performed on a separate computing device. The artifact reduction computing device 302 may be included in the workstation 12 of the MRI system 10, or may be included on a separate computing device that communicates with the workstation 12. In one example, the artifact reduction computing device 302 is a server computing device, and may be cloud-based.

[0041] In an exemplary embodiment, the neural network model 304 can be trained by computer-generated training images. The training images can include natural images and computer-generated natural images with streak artifacts. For example, when generating computer-simulated training images, natural images are enhanced in random areas. Radial acquisition is used to simulate the randomly enhanced natural images to generate k-space data. Images with streak artifacts are generated after reconstructing the simulated k-space data. When trained with computer-generated training images, the neural network model 304 is trained to identify and estimate streak artifacts in the input images to the neural network model 304. Training images with other channel-related artifacts can be similarly generated by the computer based on the causes of channel-related artifacts. For example, motion artifacts can be generated by generating k-space data using selected acquisition, Cartesian acquisition, or non-Cartesian acquisition, introducing motion during computer-simulated acquisition, and generating natural images with motion artifacts using k-space data simulated by motion reconstruction.

[0042] Figure 3B 350 is a flow chart of an exemplary method 350. The method 350 may be implemented on the artifact reduction system 300. In an exemplary embodiment, the method 350 includes receiving 352 a subject 102 (see Figure 1) in the subject 102. The MR image set is acquired by the RF coil assembly 38. The RF coil assembly 38 includes one or more channels. Each MR image set corresponds to a channel, wherein the MR image set is acquired by the channel. For example, if the RF coil assembly 38 includes 120 channels, 120 MR image sets are acquired. Each set corresponds to one of the 120 channels, wherein each set includes an MR image acquired by one of the 120 channels of the same volume in the subject 102. The volume is covered by one or more slices. An exemplary volume is a three-dimensional (3D) abdominal region. The MR image set is a slice of the MR image of the volume acquired by the channel.

[0043] In an exemplary embodiment, the method 350 also includes estimating 354 artifact sets in the MR image set. Each artifact set includes artifacts in the MR image set of each channel and corresponds to each channel. The artifacts can be estimated by inputting the MR image set into the neural network model 304. The neural network model 304 is configured to estimate the artifacts in the MR images input to the neural network model 304. As a result, the neural network model 304 estimates the artifact set in the MR image set, wherein the artifact set corresponds to the channel that acquired the MR image set. For example, if 120 abdominal MR image sets are input into the neural network model 304, the neural network model 304 is configured to estimate 120 artifact sets in each of the 120 abdominal MR image sets. Each artifact set corresponds to an artifact in the abdominal MR image acquired by one of the 120 channels. Artifacts of some channels may be more significant than artifacts of other channels. The method 350 also includes reducing 356 artifacts in the MR image set based on the estimated artifact set. Channels with relatively significant artifacts can be discarded or reduced in weight in image reconstruction to reduce artifacts in the final image.

[0044] In an exemplary embodiment, method 350 includes deriving 358 an artifact-reduced MR image by combining a set of artifact-reduced MR images. The artifact-reduced MR image can be combined by adding the set of artifact-reduced MR images and dividing the sum by the number of sets. MRI is unique compared to other imaging modalities because MRI signals are represented by complex numbers, rather than by scalars or real numbers. The image value of each image pixel includes magnitude and phase. The combination is performed on the complex numbers. For example, the addition of a set of MR images is the addition of complex numbers. The sum divided by the number of sets is the complex number divided by the number of sets. For example, for a set of 100 MR images with artifact reduction, the combined MR image is derived by combining 100 complex numbers for each pixel in the volume. In addition, method 350 includes outputting 360 the artifact-reduced MR image.

[0045] Figure 4An exemplary workflow of method 350 is shown. In an exemplary embodiment, images of a volume in a subject 102 are acquired. Channels may be compressed 402 to reduce the amount of data to be processed. For example, 120 channels may be compressed into 100 channels. Channel images are reconstructed 404. Each channel has a set of images of the volume. The volume is covered by slices of the image. For example, if the volume includes 50 slices, the set of MR images corresponding to one of the channels includes 50 MR image slices. The total number of sets is the total number of channels, depending on whether the channels have been compressed. For example, if the channels are not compressed, then in the above example, a set of 120 images of the volume is reconstructed. If the channels are compressed, then a set of 100 images of the volume is reconstructed.

[0046] In an exemplary embodiment, a MR image set is projected 406 along a slice direction. Projecting an MR image set along a slice direction may be referred to as a volume projection. The MR image set may be represented as a complex 3D image set at (x, y, z) positions, where z is a slice direction and represents a slice position in the slice direction, and (x, y) is a position in a transverse plane perpendicular to the slice direction. The projection may be a weighted combination of images along the slice direction, where the image values ​​of pixels at (x, y) of all slices z are combined with the weights of slice z. In one example, the projection may be an image addition along the slice direction, where the image values ​​of pixels at (x, y) of all slices z are added. Alternatively, the projection is a maximum intensity projection, where the pixel at (x, y) with the maximum intensity (such as the maximum magnitude) in all slice positions z is selected as the image value of the pixel at (x, y) in the projection image. According to the above example, 100 image sets of an abdominal volume of 50 slices are provided. The slice direction is up and down. The transverse plane is axial. Each set includes 50 axial images arranged in the up and down directions. z is the slice position in the superior-inferior direction. (x, y) is the position in the transverse plane. Volume projection of the set of 100 abdominal images is performed in the superior-inferior direction by operating on 50 complex numbers for each (x, y) position in the transverse plane according to the algorithm selected for volume projection.

[0047] Volume projection is advantageous in reducing the amount of data to be input into the neural network model 304 and reducing the number of inferences required from the neural network model 304. For example, if the number of slices is 50, the amount of data required and the number of inferences are reduced to one-fiftieth, thereby greatly reducing the computational load and increasing the computational speed. Volume projection is also advantageous in enhancing the accuracy of estimating artifacts because, if artifacts are present, they tend to appear in multiple slices, and volume projection preserves artifacts and enhances the detection of artifacts while reducing both signal and noise. Volume projection is optional and may or may not be performed on the MR image set prior to input into the neural network model 304.

[0048] In an exemplary embodiment, a collection of images is input into the neural network model 304, and artifacts are estimated 408 by the neural network model 304. An artifact index may be calculated 410. Channels are clustered 412 based on the artifact index. Clusters and channels in the clusters having relatively low artifact indexes are selected 414, and the images of the selected channels are combined. As a result, an image with reduced artifacts is generated.

[0049] Figure 5 304. FIG. 304 is a schematic diagram showing the use of a neural network model to estimate artifacts. In an exemplary embodiment, channel images 206 of the same slice acquired by different channels 204 are input into a neural network model 304. The neural network model 304 estimates artifacts 208 in the channel images 206 and outputs artifacts in slices of different channels 204. The artifacts are presented as images of the artifacts for the slice of each channel 204. If a volume projection is performed, the input is the projection images 206 of the different channels 204, and each output image of the artifacts 208 corresponds to the artifacts in the projection image 206 of the channel 204.

[0050] Fig. 6A and Figure 6B A graph of an artifact index is shown. In an exemplary embodiment, the artifact index is determined as a ratio between an artifact and an artifact-reduced image or an artifact-reduced channel image. ref ) is used as a normalizer and is obtained by subtracting the estimated artifact (I 伪影 ) to export. Channel Image I 通道 is the set of images acquired by a particular channel before artifact reduction. Subtraction is performed on the complex numbers. The artifact index R is determined for each channel and represents the level of artifacts in the images acquired by that channel. The artifact index R can be calculated as the image intensity of the artifact (|I 伪影 |) and the image intensity of the artifact-reduced image (|I ref |), for example, R = |I 伪影 | / |Iref |. The image intensity can be calculated as the sum of the magnitudes of the individual pixels in the image. Alternatively, the artifact indicator can be the image intensity of the artifact (|I 伪影 |).

[0051] In an exemplary embodiment, the channels can be clustered based on the artifact indicator. The number of clusters can be 2, or other numbers, such as ten. Fig. 6A Two clusters are used in Figure 6B Three clusters are used in . Channels with relatively high artifact levels can be discarded from the reconstruction of the final image, while channels with relatively medium or low artifact levels are selected. Weights can be applied to the clusters. Different weights can be applied to different clusters. For example, in Figure 6B , channels in cluster 602-h having the highest artifact level in cluster 602 are discarded during image reconstruction, or a weight of zero is applied. Channels in cluster 602-m having medium artifact levels are weighted with a weight of 0.5. Channels in cluster 602-1 having relatively low artifact levels are weighted with a weight of 1. As a result, images from channels having relatively high artifact levels are discarded, and image values ​​from channels having medium artifact levels are reduced, while images from channels having relatively low artifact levels are retained. In some embodiments, the weighting may be inversely proportional to the artifact indicator 604. The selected or weighted images are combined to produce a final image, thereby reducing artifacts in the final image.

[0052] In some embodiments, the selection and combination of channels is determined by collective factors, such as artifact indicators, total signal intensity in the channel image, and / or intensity distribution, such as whether the signal is evenly distributed across the image or localized. The image intensity and / or intensity distribution can be used to adjust clustering and weighting, such as the number of clusters and weightings for each cluster. For example, if the total signal intensity is low, the number of clusters can be reduced and the weights can be increased to maintain the level of signal intensity. In another example, if the signal distribution across the image is relatively uniform, the number of clusters can be reduced.

[0053] Return to reference Figure 5 In some embodiments, the neural network model 304 is configured to output an artifact indicator 604 for each channel. One or more additional layers 504 (see below) Fig. 8A and Figure 8B ) is included in the neural network model 304 and is configured to determine the artifact indicator 604 based on the artifact 208 using the systems and methods as described above.

[0054] Figure 7702-wo is a comparison of image 702-wo reconstructed without the systems and methods described herein and image 702-w of the same slice using the systems and methods described herein. Compared to image 702-wo, the artifact 208 in image 702-w is greatly reduced (see the area pointed to by arrow 703).

[0055] The systems and methods described herein can be used to remove channels or reduce the signals acquired by those channels that have relatively high artifact levels in the image. The systems and methods described herein can be used to estimate and identify artifacts in the image. For example, if a single-channel RF coil assembly 38 is used, the image acquired by the single-channel RF coil assembly 38 can be input into the neural network model 304. The neural network model 304 outputs artifacts in the image. As described above, an artifact indicator can be determined based on the output artifacts. The artifact indicator can be determined by the neural network model 304 or an external neural network model 304. The user can be notified of the artifact level in the image based on the output artifact or the artifact indicator.

[0056] In some embodiments, artifact reduction can be performed on images from one imaging sequence using artifacts estimated based on images from a different imaging sequence of the same volume. For example, a calibration imaging sequence or a positioning imaging sequence is performed during a pre-scan. Calibration and / or position data can be used to estimate artifacts. The estimated artifacts are used to reduce artifacts in subsequent images acquired using the same type of pulse sequence. For example, if the calibration imaging sequence and the subsequent imaging sequence are both acquired using a fast spin echo (FSE) pulse sequence, the artifacts estimated using the calibration imaging sequence can be used to reduce artifacts in the subsequent imaging sequence.

[0057] Fig. 8A An exemplary artificial neural network model 304 is shown. The exemplary neural network model 304 includes neuron layers 502, 504-1 to 504-n, and 506, including an input layer 502, one or more hidden layers 504-1 to 504-n, and an output layer 506. Each layer may include any number of neurons, i.e., Fig. 8A Where q, r and n can be any positive integer. It should be understood that Fig. 8A The structures and configurations shown in the figure are different structures and configurations of neural networks to implement the methods and systems described herein.

[0058] In an exemplary embodiment, the input layer 502 may receive different input data. For example, the input layer 502 includes a first input a1 representing a training image, a second input a2 representing a pattern identified in the training image, a third input a3 representing an edge of the training image, etc. The input layer 502 may include several thousand or more inputs. In some embodiments, the number of elements used by the neural network model 304 changes during the training process, and if, for example, during the execution of the neural network, it is determined that some neurons have less relevance, then these neurons are bypassed or ignored.

[0059] In an exemplary embodiment, each neuron in the hidden layers 504-1 to 504-n processes one or more inputs from the input layer 502 and / or one or more outputs from a neuron in one of the previous hidden layers to generate a decision or output. The output layer 506 includes one or more outputs, each indicating a label, a confidence factor, a weight describing an input, and / or an output image. However, in some embodiments, in addition to or instead of the outputs from the output layer 506, the outputs of the neural network model 304 are obtained from the hidden layers 504-1 to 504-n.

[0060] In some embodiments, each layer has a discrete identifiable function with respect to the input data. For example, if n equals 3, the first layer analyzes the first dimension of the input, the second layer analyzes the second dimension of the input, and the last layer analyzes the third dimension of the input. The dimensions may correspond to aspects that are considered to be of strong certainty, then to those aspects that are considered to be of medium importance, and finally to those aspects that are considered to be less relevant.

[0061] In other embodiments, the layers are not clearly delineated in terms of the functionality they perform.For example, two or more of the hidden layers 504-1 through 504-n may share decisions related to labels, with no single layer making independent decisions regarding labels.

[0062] Figure 8B According to one embodiment, the corresponding Fig. 8A Example neuron 550 of a neuron labeled "1,1" in hidden layer 504-1 in FIG. Each input (e.g., Fig. 8A The input layer 502 in the input layer is weighted so that the input a1 to a p corresponds to the weights w1 to w2 determined during the training process of the neural network model 304 p .

[0063] In some embodiments, some inputs lack explicit weights, or have weights below a threshold. The weights are applied to a function α (labeled by reference numeral 510), which may be a summation and may produce a value z1 that is input to a function labeled f 1，1 (z1) is a function 520. Function 520 is any suitable linear or nonlinear function. Figure 8B As shown, function 520 generates a plurality of outputs that can be provided to neurons of subsequent layers or used as outputs of neural network model 304. For example, the outputs can correspond to index values ​​of a list of labels, or can be calculated values ​​used as inputs of subsequent functions.

[0064] It should be understood that the structure and function of the neural network model 304 and neurons 550 shown are for illustration purposes only, and other suitable configurations exist. For example, the output of any given neuron may depend not only on the values ​​determined by past neurons, but also on future neurons.

[0065] The neural network model 304 may include a convolutional neural network (CNN), a deep learning neural network, a reinforcement or enhanced learning module or program, or a combined learning module or program that learns in two or more areas of interest or aspects. Supervised machine learning techniques and unsupervised machine learning techniques can be used. In supervised machine learning, the processing element may be provided with an exemplary input and its associated output, and an attempt may be made to discover a general rule that maps the input to the output so that when a subsequent new input is provided, the processing element may accurately predict the correct output based on the discovered rule. The neural network model 304 may be trained using an unsupervised machine learning program. In unsupervised machine learning, the processing element may need to find its own structure in an unlabeled exemplary input. Machine learning may involve identifying and recognizing patterns in existing data so as to facilitate prediction of subsequent data. A model may be created based on an exemplary input so as to make effective and reliable predictions of new inputs.

[0066] In addition or alternatively, a machine learning program can be trained by inputting sample data sets or certain data such as images, object statistics and information into the program. The machine learning program can use a deep learning algorithm that can focus primarily on pattern recognition and can be trained after processing multiple examples. The machine learning program can include Bayesian program learning (BPL), speech recognition and synthesis, image or object recognition, optical character recognition, and / or natural language processing, alone or in combination. The machine learning program can also include natural language processing, semantic analysis, automated reasoning, and / or machine learning.

[0067] Based on these analyses, the neural network model 304 can learn how to recognize features and patterns that can then be applied to analyze image data, model data, and / or other data. For example, the model 304 can learn to recognize features in a series of data points.

[0068] The workstation 12 and artifact reduction computing device 302 described herein may be any suitable computing device 800 and software implemented therein. Fig. 9 800. In an exemplary embodiment, the computing device 800 includes a user interface 804 that receives at least one input from a user. The user interface 804 may include a keyboard 806 that enables the user to enter relevant information. The user interface 804 may also include, for example, a pointing device, a mouse, a stylus, a touch-sensitive panel (e.g., a touch pad and a touch screen), a gyroscope, an accelerometer, a position detector, and / or an audio input interface (e.g., including a microphone).

[0069] In addition, in an exemplary embodiment, the computing device 800 includes a presentation interface 817 for presenting information such as input events and / or verification results to a user. The presentation interface 817 may also include a display adapter 808 coupled to at least one display device 810. More specifically, in an exemplary embodiment, the display device 810 may be a visual display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), a light emitting diode (LED) display, and / or an "electronic ink" display. Alternatively, the presentation interface 817 may include an audio output device (e.g., an audio adapter and / or a speaker) and / or a printer.

[0070] The computing device 800 also includes a processor 814 and a memory device 818. The processor 814 is coupled to the user interface 804, the presentation interface 817, and the memory device 818 via the system bus 820. In an exemplary embodiment, the processor 814 communicates with the user, such as by prompting the user via the presentation interface 817 and / or by receiving user input via the user interface 804. The term "processor" generally refers to any programmable system, including system and microcontrollers, reduced instruction set computers (RISC), complex instruction set computers (CISC), application specific integrated circuits (ASICs), programmable logic circuits (PLCs), and any other circuits or processors capable of performing the functions described herein. The above examples are exemplary only and are therefore not intended to limit the definition and / or meaning of the term "processor" in any way.

[0071] In an exemplary embodiment, the memory device 818 includes one or more devices capable of storing and retrieving information such as executable instructions and / or other data. In addition, the memory device 818 includes one or more computer-readable media, such as, but not limited to, dynamic random access memory (DRAM), static random access memory (SRAM), solid-state disks, and / or hard disks. In an exemplary embodiment, the memory device 818 stores, but is not limited to, application source code, application object code, configuration data, additional input events, application states, assertion statements, verification results, and / or any other type of data. In an exemplary embodiment, the computing device 800 may also include a communication interface 830 coupled to the processor 814 via the system bus 820. In addition, the communication interface 830 is communicatively coupled to a data acquisition device.

[0072] In an exemplary embodiment, the processor 814 may be programmed by encoding operations using one or more executable instructions and providing the executable instructions in the memory device 818. In an exemplary embodiment, the processor 814 is programmed to select a plurality of measurements received from a data collection device.

[0073] In operation, a computer executes computer executable instructions embodied in one or more computer executable components stored on one or more computer readable media to implement various aspects of the present invention described and / or shown herein. Unless otherwise specified, the execution or implementation order of the operations in the embodiments of the present invention shown and described herein is not required. That is, unless otherwise specified, these operations can be performed in any order, and embodiments of the present invention may include more or fewer operations than those disclosed herein. For example, it is conceivable that performing or implementing a particular operation before, simultaneously with, or after another operation is within the scope of various aspects of the present invention.

[0074] Fig.10 An exemplary configuration of a server computer device 1001, such as computing device 302, is shown. Server computer device 1001 also includes a processor 1005 for executing instructions. For example, the instructions may be stored in memory area 1030. Processor 1005 may include one or more processing units (e.g., in a multi-core configuration).

[0075] The processor 1005 is operatively coupled to a communication interface 1015, enabling the server computer device 1001 to communicate with a remote device or another server computer device 1001. For example, the communication interface 1015 may receive data from the workstation 12 via the Internet.

[0076] The processor 1005 may also be operatively coupled to a storage device 1034. The storage device 1034 is any computer-operated hardware suitable for storing and / or retrieving data. In some embodiments, the storage device 1034 is integrated in the server computer device 1001. For example, the server computer device 1001 may include one or more hard disk drives as the storage device 1034. In other embodiments, the storage device 1034 is external to the server computer device 1001 and can be accessed by multiple server computer devices 1001. For example, the storage device 1034 may include multiple storage units, such as hard disks and / or solid-state disks in a redundant array of independent disks (RAID) configuration. The storage device 1034 may include a storage area network (SAN) and / or a network attached storage (NAS) system.

[0077] In some embodiments, the processor 1005 is operatively coupled to the storage device 1034 via the storage interface 1020. The storage interface 1020 is any component capable of providing the processor 1005 with access to the storage device 1034. The storage interface 1020 may include, for example, an Advanced Technology Attachment (ATA) adapter, a Serial ATA (SATA) adapter, a Small Computer System Interface (SCSI) adapter, a RAID controller, a SAN adapter, a network adapter, and / or any component that provides the processor 1005 with access to the storage device 1034.

[0078] At least one technical effect of the systems and methods described herein includes (a) reducing channel-related artifacts; (b) improving the accuracy of artifact reduction; and (c) improving image quality and subject comfort and experience by reducing artifacts associated with specific positioning in radial acquisitions.

[0079] Exemplary embodiments of systems and methods for artifact reduction are described in detail above. These systems and methods are not limited to the specific embodiments described herein, but rather the components of the systems and / or operations of the methods may be used independently and separately from other components and / or operations described herein. In addition, the components and / or operations described may also be defined in other systems, methods, and / or devices, or used in conjunction with other systems, methods, and / or devices, and are not limited to practice with only the systems described herein.

[0080] Although specific features of various embodiments of the invention may be shown in some drawings and not in others, this is for convenience only. In accordance with the principles of the invention, any feature of a drawing may be referenced and / or claimed in combination with any feature of any other drawing.

[0081] This written description uses examples to disclose the invention, including the best mode, and also to enable any person skilled in the art to practice the invention, including making and using any devices or systems and performing any included methods. The patentable scope of the invention is defined by the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insignificant differences from the literal language of the claims.

Claims

1. A computer-implemented method (350) of reducing (356) artifacts in a magnetic resonance (MR) image, the method comprising: receiving (352) one or more MR image sets of a volume in a subject (102), wherein the one or more MR image sets are acquired by a radio frequency (RF) coil assembly (38) having one or more channels, and each MR image set includes a plurality of slices of MR images of the volume acquired by one of the one or more channels; One or more artifact sets in the one or more MR image sets are estimated (354) by: inputting the one or more MR image sets into a neural network model (304), wherein the neural network model (304) is configured to estimate artifacts in the MR images input into the neural network model (304), wherein each artifact set corresponds to said one of said one or more channels; and The one or more estimated artifact sets are output (360).

2. The method (350) of claim 1, wherein estimating (354) one or more artifact sets further comprises: For each MR image set, The projection MR image is exported (358) by the following steps: projecting the MR image set along a slice direction, wherein the projected MR image corresponds to the one channel of the one or more channels; and The one or more MR image sets are input by: The projected MR image is input into the neural network model (304).

3. The method (350) of claim 1, wherein the method (350) further comprises: For each MR image set, An artifact indicator in the set of MR images is determined based on a set of estimated artifacts corresponding to the set of MR images.

4. The method (350) of claim 3, wherein determining an artifact indicator further comprises: The artifact indicator is determined using the neural network model (304), wherein the neural network model (304) is configured to output the artifact indicator.

5. The method (350) of claim 1, wherein the neural network model (304) is configured to estimate streak artifacts, peripheral signal artifacts, and / or motion artifacts in the MR image input into the neural network model (304), and estimating (354) one or more artifact sets further comprises: One or more sets of streak artifacts, peripheral signal artifacts, and / or motion artifacts in the one or more sets of MR images are estimated (354) using the neural network model (304).

6. The method (350) of claim 1, wherein: Receiving (352) one or more MR image sets further comprises receiving (352) a plurality of MR image sets of a volume in the subject (102), wherein the RF coil assembly (38) comprises a plurality of channels, the plurality of MR image sets are acquired by the RF coil assembly (38), and each MR image set comprises a plurality of slices of MR images of the volume acquired by one of the plurality of channels; Estimating (354) one or more artifact sets further includes estimating (354) a plurality of artifact sets in the plurality of MR image sets by: Inputting the one or more MR image sets also includes inputting the plurality of MR image sets into a neural network model (304), wherein each artifact set corresponds to said one channel of said plurality of channels; and The method (350) further includes: reducing (356) artifacts in the plurality of MR image sets based on the plurality of estimated artifact sets; by combining the plurality of artifact-reduced MR image sets to derive (358) Artifact-reduced MR images; and The MR image with reduced artifacts is output (360).

7. The method (350) of claim 6, wherein estimating (354) a plurality of artifact sets further comprises: For each MR image set, The projection MR images are exported (358) by the following steps: projecting the set of MR images along a slice direction, wherein the projected MR image corresponds to the one channel of the plurality of channels; and The plurality of MR image sets are input by the following steps: The projected MR image is input into the neural network model (304).

8. The method (350) of claim 7, wherein projecting the set of MR images further comprises: The projection MR image is derived (358) by combining the set of MR images along the slice direction as a weighted combination.

9. The method (350) of claim 7, wherein projecting the set of MR images further comprises: The projection MR image is derived (358) by adding the set of MR images along the slice direction.

10. The method (350) of claim 7, wherein projecting the set of MR images further comprises: The projection MR image is derived (358) using maximum intensity projection in the slice direction.

11. The method (350) of claim 6, wherein: Reducing (356) artifacts also includes: For each MR image set, determining an artifact indicator in the set of MR images based on an estimated artifact set corresponding to the set of MR images; and, optionally, The artifact indicators also include: The artifact indicator is determined using the neural network model (304), wherein the neural network model (304) is configured to output the artifact indicator.

12. The method (350) of claim 6, wherein: Reducing (356) artifacts also includes: clustering the plurality of MR image sets based on the plurality of estimated artifact sets; and applying weighting to the plurality of MR image sets based on the cluster to which the MR image sets belong; and Exporting (358) the MR image also includes: combining the plurality of weighted MR image sets; and, optionally Reducing artifacts further includes determining the number of clusters and / or the weighting based on an artifact indicator, a signal intensity and / or a signal intensity distribution of the MR image.

13. The method (350) of claim 6, wherein: Reducing (356) artifacts also includes: clustering the plurality of MR image sets based on the plurality of estimated artifact sets; and One or more MR image sets in the cluster having the highest artifact levels are removed.

14. The method (350) of claim 6, wherein reducing (356) artifacts further comprises: Weightings are applied to the plurality of MR image sets based on the plurality of estimated artifact sets.

15. The method (350) of claim 6, wherein the neural network model (304) is configured to estimate streak artifacts, peripheral signal artifacts, and / or motion artifacts in the MR image input into the neural network model (304), and estimating (354) a plurality of artifact sets further comprises: A plurality of streak artifacts, peripheral signal artifacts, and / or motion artifact sets in the plurality of MR image sets are estimated (354) using the neural network model (304).

Images