Correction of magnetic resonance images using multiple magnetic resonance imaging system configurations

Abstract

Disclosed is a medical system (100, 300) comprising a memory (110) storing machine-executable instructions (120) and an image generation neural network (122). The image generation neural network is configured to receive reference magnetic resonance image data (126) as input and output synthetic magnetic resonance image data (128). The synthetic magnetic resonance image data is a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system. Execution of the machine-executable instructions causes a computing system (106) to receive (200) measured k-space data (124) acquired according to the first configuration of the magnetic resonance imaging system, receive (202) the reference magnetic resonance image data, input the reference magnetic resonance image data to the image generation neural network to receive (204) the synthetic magnetic resonance image data, and reconstruct (206) corrected magnetic resonance image data (132) from the measured k-space data and the synthetic magnetic resonance image data.

Description

[Technical Field]

[0001] The present invention relates to magnetic resonance imaging, and more particularly to reducing artifacts in magnetic resonance images. [Background technology]

[0002] Magnetic resonance imaging (MRI) scanners use a large static magnetic field to align the nuclear spins of atoms as part of a procedure to generate images of a patient's internal body. This large static magnetic field is called the B0 field or main magnetic field. MRI can be used to spatially measure various quantities or properties of a subject. For example, MRI can be used to study various electrical and tissue properties of a subject. A drawback of MRI is that it can take several minutes to acquire enough k-space data to reconstruct a magnetic resonance image. Subject movement and reception of spurious RF signals can introduce artifacts and corrupt the magnetic resonance image.

[0003] U.S. Patent Application Publication US20190377047A1 discloses using deep learning to train an image-to-image neural network to generate artifact-reduced images for a magnetic resonance imaging system. The image-to-image network can be applied in real time. To address diverse image processing situations, the image-to-image network can (a) use an auxiliary map as input in conjunction with MR data input from a patient, (b) use sequence metadata as a controller for the image-to-image network's encoder, and / or (c) be trained to generate contrast-invariant features in the encoder using a discriminator that receives the encoder features.

[0004] International application WO2019 / 224800 relates to simulating and constructing actual MRI images in a second modality from source MRI images taken in a first modality. Summary of the Invention

[0005] The present invention provides a medical system, a computer program product, and a magnetic resonance imaging method according to the independent claims. Embodiments are described in the dependent claims. Accordingly, the medical system comprises a memory storing machine-executable instructions and access to an image-generating neural network. The image-generating neural network may be integrated into the medical system, or the medical system may be configured to control access via a data link to a potentially remotely located image-generating neural network. The image-generating neural network receives reference magnetic resonance image data as input and outputs synthetic magnetic resonance image data, and the image-generating neural network generates the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system. A computing system is configured to control the medical system, and by executing the machine-executable instructions, the computing system: accessing measured k-space data acquired according to a first configuration of a magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; accessing reference magnetic resonance image data representative of a region of interest of the subject; generating access to synthetic magnetic resonance image data by inputting reference magnetic resonance image data into an image generation neural network; and arranging for reconstructing corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data.

[0006] Access to the measured k-space data and the reference magnetic resonance image data may be implemented in such a way that the medical system receives these data and forwards them to input to an image generation neural network that may be remote from the medical system or integrated into the medical system. Access to the measured k-space data and the reference magnetic resonance image data may also be implemented in such a way that these data are remotely controlled and input to and output from the image generation neural network, respectively. Access to the composite magnetic resonance image data may be generated from a remotely located image generation neural network by remotely controlling the composite magnetic resonance image data to be applied to reconstruction software. Access may be generated in such a way that the medical system receives the composite magnetic resonance image data and forwards these data to reconstruction software or applies these data to reconstruction software integrated into the medical system.

[0007] Subject motion, spurious RF signals, or other defects can create artifacts or corrupt magnetic resonance images. Embodiments may provide means for reducing artifacts or image corruption and / or accelerating image acquisition. An image generation neural network may be trained to receive reference magnetic resonance image data acquired using a second configuration of the magnetic resonance imaging system and output synthetic magnetic resonance image data. The synthetic magnetic resonance image data is a simulation of the magnetic resonance image data acquired for a first configuration of the magnetic resonance imaging system.

[0008] The synthetic magnetic resonance image data can then be used to improve the reconstruction of a corrected magnetic resonance image from measured k-space data acquired using the first configuration of the magnetic resonance imaging system. In one example, the synthetic magnetic resonance image data can provide prior knowledge that can be used in a regularization term during reconstruction. In another example, the synthetic magnetic resonance image data can be used to calculate synthetic k-space data that can be used, for example, to correct, supplement, correct, or replace a portion of the measured k-space data.

[0009] The magnetic resonance imaging system is configured to arrange for the reconstruction of a set of magnetic resonance images from the echo signals, with reconstruction software installed on the computing system of the magnetic resonance examination system or accessible to a remote reconstruction facility. The reconstruction software may be installed on a remote server, such as at a medical institution where the magnetic resonance imaging system is installed, or in some cases, accessible to a data network, such that the reconstruction software is available in the "cloud." In these remote configurations, the computing system is capable of arranging the reconstruction of the set of magnetic resonance images at a remote reconstruction facility. Furthermore, the reconstruction of the magnetic resonance images may be performed by machine learning, for example, by a trained neural network embedded in the computing system, or may be accessible and transferable for reconstruction from a remote location.

[0010] In one aspect, the present invention provides a medical system including a memory storing machine-executable instructions and an image generation neural network configured to receive as input reference magnetic resonance image data and output a synthetic magnetic resonance image, the image generation neural network configured to generate the synthetic magnetic resonance image as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system.

[0011] In other words, the image-generating neural network takes reference magnetic resonance image data acquired according to a second configuration and generates a synthetic magnetic resonance image that simulates a magnetic resonance image acquired according to a first configuration of the magnetic resonance imaging system. The first and second configurations may, for example, be different types of pulse sequences used to control the magnetic resonance imaging system to generate a particular MR contrast. In another example, the difference between the first and second configurations may be a change in the configuration of a similar pulse sequence. For example, the TE or TR values ​​may be changed. In another example, the same pulse sequence may be used with a different resolution. Often, even when using different magnetic resonance imaging protocols, much of the data is redundant. This allows for a fairly high degree of accuracy in the output of synthetic magnetic resonance image data.

[0012] Image-generating neural networks can be trained in a simple manner. For example, a magnetic resonance imaging system can be used to acquire training images using a second configuration of magnetic resonance images, and before or after acquiring ground truth images acquired with a first configuration of the magnetic resonance imaging system. This can be done once to provide one training data pair. This process can be repeated with different subjects and different configurations as appropriate. This training data can then be used, for example, using backpropagation or deep learning algorithms to train the image-generating neural network.

[0013] The medical system further comprises a computing system for controlling the medical system. The computing system may take different forms in different examples. In one example, the computing system may be a workstation used by, for example, a radiologist. In another example, the computing system may be a remote computing system or a cloud computing system that provides an imaging surface. In another example, the computing system may be a computing system that controls the operation and functionality of a magnetic resonance imaging system.

[0014] Execution of the machine-executable instructions causes the computing system to receive measured k-space data acquired according to a first configuration of the magnetic resonance imaging system, the measured k-space data representing a region of interest of the subject. Execution of the machine-executable instructions also causes the computing system to receive reference magnetic resonance image data, the reference magnetic resonance image data representing the region of interest of the subject. Execution of the machine-executable instructions also causes the computing system to receive synthesized magnetic resonance image data by inputting the reference magnetic resonance image data into an image-generating neural network.

[0015] And finally, execution of the machine-executable instructions causes the computing system to reconstruct corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image, the composite magnetic resonance image being consistent with the first configuration used to acquire the measured k-space data, such that the composite magnetic resonance image can be used to assist in the reconstruction of the corrected magnetic resonance image data.

[0016] As used herein, magnetic resonance image data encompasses data that can be used to render or construct one or more magnetic resonance images. For example, the reference magnetic resonance image data may be one or more magnetic resonance images in one example, or an averaged magnetic resonance image in some other examples. In another example, the reference magnetic resonance image data may be an image or mapping generated from a magnetic resonance fingerprinting protocol. Similarly, the composite magnetic resonance image data may take different forms in different examples. The composite magnetic resonance image data may be data for constructing one or more magnetic resonance images, or may be a three-dimensional magnetic resonance imaging mapping or image dataset. The composite magnetic resonance image data may be the result of different magnetic resonance fingerprinting protocols.

[0017] In some instances, the reference magnetic resonance image data is a single magnetic resonance image or image dataset.

[0018] In other examples, the reference magnetic resonance image data includes multiple magnetic resonance images. In some cases, these multiple magnetic resonance images are acquired using multiple configurations or contrasts. In this case, the second configuration of the magnetic resonance imaging system is a collection or bundle of configurations, one configuration for each image or image data set that constitutes the reference magnetic resonance imaging data. As a specific example, three, four, or more magnetic resonance images acquired with different contrasts may be grouped to form the reference magnetic resonance image data.

[0019] In other embodiments, the image generation neural network is configured to receive reference magnetic resonance image data according to a predetermined image format. For example, this may be the format of the images used to train the image generation neural network. Execution of the machine-executable instructions further causes the computing system to convert the reference magnetic resonance image data into the predetermined image format before inputting the reference magnetic resonance image data into the image generation neural network. For example, the region of interest and voxel size may be changed using standard image transformation techniques. Execution of the machine-executable instructions further causes the computing system to spatially align the composite magnetic resonance image with the measured k-space data before reconstructing the corrected magnetic resonance image data. This may include changing the view within the image or the positioning of the image. Using these basic image transformation techniques, the composite magnetic resonance image data can be formatted to match the first configuration of the magnetic resonance imaging system.

[0020] For example, the image generation neural network can be configured to receive the composite magnetic resonance image according to a predetermined output format, and the computing system can adapt the predetermined output format to match a first configuration of the magnetic resonance imaging system.

[0021] In other embodiments, the measured k-space data and the synthetic magnetic resonance image data are spatially coincident, which may, for example, allow for better comparison of the k-space data.

[0022] In some examples, the image generation neural network may have an input vector specifying a first configuration of the magnetic resonance imaging system and a second configuration of the magnetic resonance imaging system, in which case the neural network may automatically match the reference magnetic resonance image data and the synthetic magnetic resonance image data, although this requires more training of the image generation neural network.

[0023] Another embodiment of the synthetic magnetic resonance image data provides prior knowledge during reconstruction of the corrected magnetic resonance image data. Macroscopic structures, such as the location of organs or other anatomical structures, may be present in the synthetic magnetic resonance image data. This may be useful, for example, to replace or correct various portions of the measured k-space data. The synthetic magnetic resonance image data may also be used, for example, as a regularization term during reconstruction to improve the quality of the corrected magnetic resonance image data.

[0024] In another embodiment, execution of the machine-executable instructions further causes the computing system to reconstruct synthetic k-space data from the reference magnetic resonance image data. The measured k-space data is divided into multiple k-space data groups. Corrected magnetic resonance image data is reconstructed using the synthetic k-space data to correct at least some of the k-space data groups. Standard techniques can be used to return from the image space of the synthetic magnetic resonance image data to the k-space data. The first configuration of the magnetic resonance imaging system can be used, for example, to perform a backward calculation to simulate what the k-space data would look like if the k-space data were used to generate the synthetic magnetic resonance image data.

[0025] For example, in parallel imaging techniques, the coil sensitivities may even be used to generate simulated images of each coil or acquisition channel, which may then be used to simulate k-space data acquired from the individual coils or channels. This may be beneficial because it may allow for correction of noise or other errors in acquiring the measured k-space data.

[0026] In another embodiment, execution of the machine-executable instructions further causes the computing system to determine a rigid body transformation of one or more of the groups of k-space data using the composite k-space data. Execution of the machine-executable instructions further causes the computing system to perform phase and amplitude correction of the one or more groups of k-space data using the rigid body transformation. This embodiment may be advantageous because it may provide a simple way to reduce the effects of rigid body motion by the subject.

[0027] In another embodiment, execution of the machine-executable instructions further causes the computing system to use the synthetic k-space data to determine a configuration for a predefined motion model. Execution of the machine-executable instructions further causes the computing system to perform correction of one or more of the k-space data groups using the predefined motion model. For example, there may be a motion model that can be used to represent affine and / or non-rigid transformations or motions of a subject. The predefined motion model can be used to define how the k-space data is modified as the subject moves according to the predefined motion model. This may be beneficial because it may enable correction of measured k-space data.

[0028] In other embodiments, the predefined motion model is configured to provide a transformation of the synthetic k-space data that corresponds to an affine or elastic transformation in image space.

[0029] In another embodiment, execution of the machine-executable instructions further causes the computing system to detect at least one incomplete k-space sampling region in the measured k-space data. For example, a portion of the measured k-space data may be incomplete, corrupted, or missing. Execution of the machine-executable instructions further causes the computing system to fill the incomplete k-space sampling region in the measured k-space data with synthetic k-space data. This may be beneficial because it may improve the quality or enable the use of measured k-space data that would otherwise have to be discarded and reacquired. One situation in which this may be beneficial is when subject movement is monitored using a navigator or an external motion measurement system such as a camera or breathing belt. This may enable automatic detection of corrupted k-space data. If the corrupted k-space data is discarded, the incomplete k-space sampling region may be filled with synthetic k-space data.

[0030] In another embodiment, execution of the machine-executable instructions further causes the computer system to receive a motion signal representing the subject's movement. Execution of the machine-executable instructions further causes the computer system to reconstruct corrected magnetic resonance image data using k-space data groups having motion signals that fall within a predetermined range. In this example, the motion signal provided may be provided, for example, from a magnetic resonance navigator or a system that measures changes in the subject's position or movement. For example, a breathing belt and camera may be used. The motion signal is then essentially used to gate which k-space data to use.

[0031] In another embodiment, execution of the machine-executable instructions causes the computing system to further calculate a motion signal as a composite motion signal by comparing the composite k-space data with each k-space data group. For example, each k-space data group may be directly compared with the composite k-space data and fitting may be performed. This may correspond to a change in phase and / or amplitude of the sample points. This may enable calculation of a motion signal that may correspond to a navigator. This may enable gating which k-space data to use for a particular motion signal. This may be useful, for example, to generate magnetic resonance images of cardiac or respiratory phases.

[0032] The resultant motion signal may be calculated in either k-space or image space, depending, for example, on the size of the k-space data group.

[0033] In another embodiment, the memory further includes an image quality assessment module configured to output an image quality metric. Execution of the machine-executable instructions further causes the computing system to generate a plurality of k-space data sets by systematically replacing combinations of k-space data groups with portions of the composite k-space data. Execution of the machine-executable instructions further causes the computing system to generate a plurality of trial magnetic resonance image data sets by reconstructing each set of the plurality of k-space data sets. Execution of the machine-executable instructions further causes the computing system to select corrected magnetic resonance image data sets from the plurality of trial magnetic resonance image data sets by optimizing the image quality metric output of the image quality assessment module.

[0034] For example, when executing this algorithm, it can determine how many portions of the composite k-space data can be used to replace a group of k-space data. An iterative algorithm can be run to systematically replace all or many combinations for an optimization process. This embodiment can be beneficial because it allows for data correction when there is no other way to correct data corrupted by, for example, noise, spurious signals, or complex involuntary movements.

[0035] In another embodiment, the image quality metric is determined by using registration between the composite magnetic resonance image data and one of the trial magnetic resonance image data. The composite magnetic resonance image data must be similar to or very close to the format that the desired corrected magnetic resonance image data should be. Standard registration techniques can be used to calculate the registration or mapping between the two image data sets. This metric can then be used to provide an image quality metric. For example, the similarity between the positions of various anatomical landmarks can be measured.

[0036] In other embodiments, the image quality metric is determined using the output from a trained neural network that outputs an image quality metric when one of a plurality of trial magnetic resonance images is input. For example, the trained neural network may be trained by acquiring a complete set of magnetic resonance imaging data and then corrupting or introducing spurious motion artifacts into the data. This may then be used to assign a classification or metric that can be used in the optimization process.

[0037] In another embodiment, the image quality metric is determined by calculating the total image gradient for each of the plurality of trial magnetic resonance image data.

[0038] In another embodiment, the image quality metric is determined by calculating the image entropy of each of a plurality of trial magnetic resonance images.

[0039] In another embodiment, execution of the machine-executable instructions further causes the computing system to reconstruct a plurality of corrected magnetic resonance image data sets. Execution of the machine-executable instructions further causes the computing system to perform one of: providing the corrected magnetic resonance image data as an average of the plurality of corrected magnetic resonance image data sets; and providing the corrected magnetic resonance image as a selection from the plurality of corrected magnetic resonance images. For example, the corrected magnetic resonance image data can be generated using one or more of the above methods. All of these images may be averaged to provide a better estimate.

[0040] In another embodiment, reconstructing the corrected magnetic resonance image data from the measured k-space data and the synthetic magnetic resonance image is formulated as an optimization problem that assigns a weighting factor to each k-space data group. Execution of the machine-executable instructions further causes the computing system to identify at least one corrupted k-space data group selected from the plurality of k-space data groups. This identification can be performed in various ways. In some cases, an external navigator or other signal can be used to identify the corrupted k-space data. In another example, the corrupted k-space data can be identified by comparing the k-space data with the synthetic k-space data.

[0041] Execution of the machine-executable instructions causes the computing system to further correct at least one corrupted k-space data group using the composite k-space data. Execution of the machine-executable instructions also causes the computing system to assign a weighting factor to each of the k-space data groups. The at least one corrupted k-space data group is assigned a weighting factor of a reduced value. This can be beneficial because, during reconstruction, the remaining measured k-space data is given a higher weight for reconstructing the corrected magnetic resonance image. Assigning a reduced weighting factor to the corrected corrupted k-space data group allows that group to participate in the reconstruction of the corrected magnetic resonance image, but with a reduced influence.

[0042] In other embodiments, at least one corrupted k-space data group selected from the k-space data groups is detected by using any one of an external navigator signal, detecting missing k-space data, or comparison with synthetic k-space data, and combinations thereof.

[0043] In other embodiments, correcting the at least one corrupted k-space data group using synthetic k-space data is performed using any one of: replacing the at least one corrupted k-space data group with synthetic k-space data; modifying or shifting the at least one corrupted k-space data group; adding synthetic k-space data to the at least one corrupted k-space data group; and combinations thereof.

[0044] In the above embodiment, a soft gating process is described. This may be a data integrity term that includes a weighting factor that reflects how reliable each measurement result is. The weight may be, for example, any positive number. One possibility is that a gating process uses weights of either 0 or 1. It is also possible to replace the description with a more general soft gating formulation in which the weighting factor w is a positive number that depends on the value of the navigator signal.

[0045] In another embodiment, the corrected magnetic resonance image data is reconstructed according to a compressed sensing image reconstruction algorithm, which can be beneficial because the use of a synthetic magnetic resonance image can reduce the amount of data that needs to be sampled to reconstruct the corrected magnetic resonance image data.

[0046] In other embodiments, the compressed sensing image reconstruction algorithm is an iterative algorithm that iteratively generates intermediate magnetic resonance images. The compressed sensing image reconstruction algorithm includes denoising the intermediate magnetic resonance images using the composite magnetic resonance image data.

[0047] In another embodiment, the compressed sensing image reconstruction algorithm is configured to generate the intermediate magnetic resonance images by solving an optimization problem, the optimization problem including a regularization term, the regularization term being a function of the composite magnetic resonance image data, and the composite magnetic resonance image data being used to denoise the intermediate magnetic resonance images.

[0048] In another embodiment, the memory further includes an image denoising neural network configured to receive as input the intermediate magnetic resonance image data and the composite magnetic resonance image data and output denoised magnetic resonance image data. Execution of the machine-executable instructions further causes the processor to receive filtered magnetic resonance image data by inputting the intermediate magnetic resonance image data and the composite magnetic resonance image data to the image denoising neural network. The denoised magnetic resonance image data is used as input to an iterative algorithm to iteratively generate intermediate magnetic resonance image data. In this embodiment, the denoising neural network is configured as a filtering network. The filter depends on the value of the composite magnetic resonance image data.

[0049] In another embodiment, the image generation neural network is further configured to receive a configuration vector as an input. The configuration vector may specify a first configuration of the magnetic resonance imaging system and a second configuration of the magnetic resonance imaging system. In this embodiment, the input generation neural network is configured by the configuration vector to control the input and output formats. Using the configuration vector may allow training of a single network that can accommodate a variety of configuration pairs, but may require more training.

[0050] In other embodiments, the medical system further includes at least one magnetic resonance imaging system. For example, the first configuration may be for a first magnetic resonance imaging system and the second configuration may be for a second magnetic resonance imaging system. In other cases, there is only one magnetic resonance imaging system, and both the measurement k-space data and the reference magnetic resonance image data are acquired on the same magnetic resonance imaging system. Various software-implemented functions of the magnetic resonance imaging system, such as the image generation neural network and reconstruction, may be remotely accessible or located within a computing system that controls the magnetic resonance imaging system.

[0051] The memory further includes first pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire measured k-space data. The memory further includes second pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire reference k-space data. Execution of the machine-executable instructions causes the computing system to further acquire the reference k-space data by controlling the magnetic resonance imaging system with the second pulse sequence commands. Execution of the machine-executable instructions also causes the computing system to reconstruct a reference magnetic resonance image from the reference k-space data. Execution of the machine-executable instructions also causes the processor to further acquire the measured k-space data by controlling the magnetic resonance imaging system with the first pulse sequence commands.

[0052] In other embodiments, execution of the machine-executable instructions further causes the computing system to construct synthetic k-space data using a reference magnetic resonance image. Execution of the machine-executable instructions further causes the computing system to use the synthetic k-space data to control the acquisition of measured k-space data. For example, as measured k-space data is acquired in groups or shots, the acquired measured k-space data may be compared directly to the synthetic k-space data and used to control or modify the acquisition of further measured k-space data.

[0053] In another embodiment, execution of the machine-executable instructions causes the computing system to control the acquisition of measured k-space data by using the composite k-space data to select a k-space data sampling pattern for a first pulse sequence command. Signals in k-space have non-uniform power density. By examining the composite k-space data, it is possible to infer important portions of k-space to select as the sampling pattern when sampling the measured k-space data. For example, an algorithm can look at the composite k-space data to see where the power density is highest and modify the sampling pattern of the k-space data for sampling accordingly.

[0054] In another embodiment, the first pulse sequence command is configured to control the magnetic resonance imaging system to acquire measurement k-space data in the k-space data group. Execution of the machine-executable instructions further causes the computing system to calculate a comparison metric between the composite k-space data and each k-space data group. Execution of the machine-executable instructions further causes the computing system to perform a predetermined action if the comparison metric is outside a predetermined range of values. For example, the comparison metric may calculate a similarity between the composite k-space data and the group of acquired k-space data or perform a pattern matching operation. If the match falls below a predetermined amount, the predetermined action is triggered. In another embodiment, the predetermined action is reacquiring at least a portion of the k-space data group, stopping acquisition of measurement k-space data, or a combination thereof.

[0055] In other embodiments, the corrected magnetic resonance image is reconstructed according to a parallel imaging magnetic resonance imaging reconstruction algorithm, which may be combined with, for example, compressed sensing.

[0056] In another aspect, the present invention provides a method of operating a medical system. The method includes receiving measured k-space data acquired according to a first configuration of a magnetic resonance imaging system. The measured k-space data represents a region of interest of a subject. The method further includes receiving reference magnetic resonance image data. The reference magnetic resonance image data represents the region of interest of the subject. The reference magnetic resonance image data is acquired according to a second configuration of the magnetic resonance imaging system. The method further includes receiving synthetic magnetic resonance image data by inputting the reference magnetic resonance image data to an image generation neural network. The image generation neural network is configured to output synthetic magnetic resonance image data upon receiving the reference magnetic resonance image data as input.

[0057] The image generation neural network is configured to generate a synthetic magnetic resonance image as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system. The method further includes reconstructing corrected magnetic resonance image data from the measured k-space data and the synthetic magnetic resonance image data.

[0058] In another aspect, the present invention provides a computer program comprising machine-executable instructions executed by a computing system for controlling a medical system, the computer program further comprising an image generation neural network configured to receive as input reference magnetic resonance image data and output synthetic magnetic resonance image data, the image generation neural network being configured to generate the synthetic magnetic resonance image as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system.

[0059] Execution of the machine-executable instructions causes the computing system to receive measured k-space data acquired according to a first configuration of the magnetic resonance imaging system. The measured k-space data represents a region of interest of a subject. Reference magnetic resonance image data is acquired according to a second configuration of the magnetic resonance imaging system. The reference magnetic resonance image data represents the region of interest of the subject. Execution of the machine-executable instructions also causes the computing system to receive synthesized magnetic resonance image data by inputting the reference magnetic resonance image data into an image-generating neural network. Execution of the machine-executable instructions also causes the computing system to reconstruct corrected magnetic resonance image data from the measured k-space data and the synthesized magnetic resonance image data.

[0060] In another aspect, the present invention provides a magnetic resonance imaging system including a memory storing machine-executable instructions and an image generation neural network configured to receive reference magnetic resonance image data as input and output synthetic magnetic resonance image data, and the image generation neural network configured to generate the synthetic magnetic resonance image as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system.

[0061] The memory further includes first pulse sequence commands configured to control the magnetic resonance imaging system to acquire the measurement k-space data. The memory further includes second pulse sequence commands configured to control the magnetic resonance imaging system to acquire the reference k-space data. The magnetic resonance imaging system further includes a computing system configured to control the medical system.

[0062] Execution of the machine-executable instructions causes the computing system to acquire reference k-space data by controlling the magnetic resonance imaging system with second pulse sequence commands. Execution of the machine-executable instructions also causes the computing system to reconstruct reference magnetic resonance image data from the reference k-space data. Execution of the machine-executable instructions also causes the computing system to construct composite k-space data using the reference magnetic resonance image data. Execution of the machine-executable instructions also causes the computing system to control the acquisition of measurement k-space data using the first pulse sequence commands and the composite k-space data. For example, as measurement k-space data groups or shots are measured, the composite k-space data can be compared to measurement k-space data groups or shots and used to adapt the acquisition of further measurement k-space data in real time.

[0063] In another embodiment, execution of the machine-executable instructions causes the computing system to control the acquisition of measured k-space data by using the synthetic k-space data to select a k-space data sampling pattern for a first pulse sequence command. For example, the synthetic k-space data may be used to select a k-space data sampling pattern or to modify the first pulse sequence command before executing the first pulse sequence command. In this embodiment, first synthetic k-space data is calculated first. The synthetic k-space data is then used to adjust the sampling pattern. Because k-space is sparse, the synthetic k-space data is used to predict where more samples are needed.

[0064] The first pulse sequence command is configured to control the magnetic resonance imaging system to acquire measurement k-space data in the k-space data group. Execution of the machine-executable instructions further causes the computing system to calculate a comparison metric between the composite k-space data and each k-space data group. Execution of the machine-executable instructions further causes the computing system to perform a predetermined action if the comparison metric is outside a predetermined range of values.

[0065] In other embodiments, the predetermined action is reacquiring at least a portion of the k-space data group, stopping acquisition of the measurement k-space data, or any combination thereof.

[0066] Various embodiments may be represented by one or more of the following numbered clauses:

[0067] Clause 1. A medical system, comprising: a memory storing machine-executable instructions and access to an image generation neural network, the image generation neural network receiving as input reference magnetic resonance image data and outputting synthetic magnetic resonance image data, the image generation neural network generating the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system; a computing system for controlling the medical system, wherein execution of the machine-executable instructions causes the computing system to: receiving measured k-space data acquired according to a first configuration of a magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; receiving reference magnetic resonance image data representative of a region of interest of a subject; receiving synthetic magnetic resonance image data by inputting reference magnetic resonance image data into an image generation neural network; and reconstructing corrected magnetic resonance image data from the measured k-space data and the synthetic magnetic resonance image data.

[0068] Clause 2. The image generating neural network is configured to receive reference magnetic resonance image data according to a predetermined input format, and execution of the machine executable instructions causes the computing system to further: converting the reference magnetic resonance image data into a predetermined input format before inputting the reference magnetic resonance image data into the image generation neural network; 2. The medical system of claim 1, further comprising: spatially matching the synthetic magnetic resonance image data to the measured k-space data before reconstructing the corrected magnetic resonance image data.

[0069] Clause 3. A medical system according to clause 1 or 2, wherein the measured k-space data and the synthesized magnetic resonance image data are spatially coincident.

[0070] Clause 4. The medical system of clause 1, 2, or 3, wherein the synthetic magnetic resonance image data provides prior knowledge during reconstruction of the corrected magnetic resonance image data.

[0071] Clause 5. The medical system of clauses 1 to 4, wherein upon execution of the machine-executable instructions, the computing system further reconstructs synthetic k-space data from the synthetic magnetic resonance image data, the measured k-space data being divided into a plurality of k-space data groups, and the corrected magnetic resonance image data being reconstructed by correcting at least some of the k-space data groups using the synthetic k-space data.

[0072] Clause 6. By executing the machine-executable instructions, the computing system further: determining a rigid body transformation or higher order transformation (HOT) of one or more of the groups of k-space data using the composite k-space data; 6. The medical system of clause 5, further comprising: performing phase and amplitude correction of one or more of the k-space data groups using a rigid body transformation or HOT.

[0073] Clause 7. By executing the machine-executable instructions, the computing system further: determining a configuration of a predefined motion model using the synthetic k-space data; 7. The medical system of clause 5 or 6, further comprising: performing correction of one or more of the k-space data groups using a predefined motion model.

[0074] Clause 8. The medical system of clause 7, wherein the predefined motion model is configured to provide a transformation of the synthetic k-space data that corresponds to an affine or elastic transformation in image space.

[0075] Clause 9. By executing the machine-executable instructions, the computing system further: Detecting at least one incomplete k-space sampling region in the measured k-space data; and filling incomplete k-space sampling regions in the measured k-space data with synthetic k-space data.

[0076] Clause 10. By executing the machine-executable instructions, the computing system further: receiving a motion signal representative of movement of the subject; A medical system described in any one of clauses 5 to 9, further comprising: reconstructing corrected magnetic resonance image data using k-space data groups having motion signals that fall within a predetermined range.

[0077] Clause 11. By executing the machine-executable instructions, the computing system further: 11. The medical system of claim 10, wherein the motion signal is calculated as a composite motion signal by comparing the composite k-space data with each k-space data group.

[0078] Clause 12. The memory further includes a picture quality assessment module that outputs a picture quality metric, and execution of the machine-executable instructions further causes the computing system to: generating a plurality of k-space data sets by systematically replacing combinations of k-space data groups with portions of the synthetic k-space data; generating a plurality of trial magnetic resonance image data sets by reconstructing each set of the plurality of k-space data sets; A medical system described in any one of clauses 5 to 11, further comprising: selecting corrected magnetic resonance image data from a plurality of trial magnetic resonance image data by optimizing the image quality metric output by the image quality assessment module.

[0079] Article 13. Image Quality Metrics registration between the composite magnetic resonance image data and one of the plurality of trial magnetic resonance image data; an output from a trained neural network that outputs an image quality metric when one of the plurality of trial magnetic resonance image data is input; By calculating the total image gradient, and 13. The medical system of claim 12, wherein the image entropy is determined using any of:

[0080] Clause 14. By executing the machine-executable instructions, the computing system further: reconstructing a plurality of corrected magnetic resonance image data; A medical system as described in any one of clauses 5 to 13, performing any one of providing corrected magnetic resonance image data as an average of a plurality of corrected magnetic resonance image data sets, and providing corrected magnetic resonance image data as a selection from a plurality of corrected magnetic resonance image data sets.

[0081] Clause 15. The reconstruction of corrected magnetic resonance image data from the measured k-space data and the synthetic magnetic resonance image data is formulated as an optimization problem of assigning weighting factors to each of the k-space data groups, and by executing the machine-executable instructions, the computing system further: identifying at least one corrupted k-space data group selected from the k-space data groups; correcting at least one corrupted group of k-space data using the synthetic k-space data; 15. A medical system as described in any one of clauses 5 to 14, wherein the medical system performs the steps of: assigning a weighting factor to each of the k-space data groups, and wherein at least one corrupted k-space data group is assigned a weighting factor of a reduced value.

[0082] Clause 16. The medical system of clause 15, wherein at least one corrupted k-space data group selected from the plurality of k-space data groups is detected by any one of using an external navigator signal, detecting missing k-space data, comparing with synthetic k-space data, and combinations thereof.

[0083] Clause 17. The medical system of clause 15 or 16, wherein correcting at least one corrupted k-space data group using synthetic k-space data is performed using any one of replacing at least one corrupted k-space data group with synthetic k-space data, modifying or shifting at least one corrupted k-space data group, adding synthetic k-space data to at least one corrupted k-space data group, and combinations thereof.

[0084] Clause 18. A medical system according to any one of clauses 1 to 17, wherein the corrected magnetic resonance image data is reconstructed according to a compressed sensing image reconstruction algorithm.

[0085] Clause 19. The medical system of clause 18, wherein the compressed sensing image reconstruction algorithm is an iterative algorithm that repeatedly generates intermediate magnetic resonance images, and the compressed sensing image reconstruction algorithm includes denoising the intermediate magnetic resonance images using the composite magnetic resonance image data.

[0086] Clause 20. The medical system of clause 19, wherein the compressed sensing image reconstruction algorithm is configured to generate intermediate magnetic resonance image data by solving an optimization problem, the optimization problem including a regularization term, the regularization term being a function of the synthetic magnetic resonance image data, and the synthetic magnetic resonance image data is used to denoise the intermediate magnetic resonance image data.

[0087] Clause 21. The medical system of clause 19, wherein the memory further includes an image denoising neural network configured to receive as input the intermediate magnetic resonance image data and the composite magnetic resonance image data and output denoised magnetic resonance image data, and execution of the machine-executable instructions causes the processor to further receive filtered magnetic resonance image data by inputting the intermediate magnetic resonance image data and the composite magnetic resonance image data to the image denoising neural network, and the denoised magnetic resonance image data is used as input to an iterative algorithm that iteratively generates the intermediate magnetic resonance image data.

[0088] Clause 22. A medical system as described in any one of clauses 1 to 21, wherein the image generation neural network is further configured to receive as input a configuration vector, the configuration vector specifying a first configuration of the magnetic resonance imaging system and a second configuration of the magnetic resonance imaging system.

[0089] Clause 23. The medical system further comprises at least one magnetic resonance imaging system, the memory further comprising first pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire measurement k-space data, and the memory further comprising second pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire reference k-space data, and execution of the machine-executable instructions causes the computing system to further: acquiring reference k-space data by controlling the magnetic resonance imaging system with a second pulse sequence command; reconstructing reference magnetic resonance image data from the reference k-space data; A medical system as described in any one of clauses 1 to 22, further comprising: acquiring measured k-space data by controlling a magnetic resonance imaging system using a first pulse sequence command.

[0090] Clause 24. By executing the machine-executable instructions, the computing system further: constructing synthetic k-space data using the synthetic magnetic resonance image data; 24. The medical system of claim 23, further comprising: controlling acquisition of measured k-space data using the synthetic k-space data.

[0091] Clause 25. The medical system of clause 24, wherein execution of the machine-executable instructions causes the computing system to control acquisition of measured k-space data by selecting a k-space sampling pattern for a first pulse sequence command using the composite k-space data.

[0092] Clause 26. The first pulse sequence command is configured to control the magnetic resonance imaging system to acquire measurement k-space data in the k-space data group, and execution of the machine-executable instructions causes the computing system to further: calculating a comparison metric between the composite k-space data and each group of k-space data; 26. The medical system of clause 24 or 25, wherein if the comparison metric is outside a predetermined range, performing a predetermined action.

[0093] Clause 27. The medical system of clause 26, wherein the predetermined action is reacquiring at least a portion of the k-space data group, stopping acquisition of the measurement k-space data, or a combination thereof.

[0094] Clause 28. A medical system according to any one of clauses 1 to 27, wherein the corrected magnetic resonance image data is reconstructed according to a parallel imaging magnetic resonance image reconstruction algorithm.

[0095] Clause 29. A method of operating a medical device system, the method comprising: receiving measured k-space data acquired according to a first configuration of a magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; receiving reference magnetic resonance image data acquired according to a second configuration of the magnetic resonance imaging system, the reference magnetic resonance image data representing a region of interest of the subject; receiving synthetic magnetic resonance image data by inputting reference magnetic resonance image data into an image generation neural network, wherein the image generation neural network outputs synthetic magnetic resonance image data when it receives the reference magnetic resonance image data as input, and the image generation neural network generates the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image was acquired according to a second configuration of the magnetic resonance imaging system; and reconstructing corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data.

[0096] Clause 30. A computer program comprising machine-executable instructions executed by a computing system for controlling a medical system, the computer program further comprising an image generation neural network, the image generation neural network receiving as input reference magnetic resonance image data and outputting synthetic magnetic resonance image data, the image generation neural network generating the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system, the computer program, upon execution of the machine-executable instructions, causing the computing system to: receiving measured k-space data acquired according to a first configuration of a magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; receiving reference magnetic resonance image data representative of a region of interest of a subject; receiving synthetic magnetic resonance image data by inputting reference magnetic resonance image data into an image generation neural network; and reconstructing corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data.

[0097] Clause 31. A magnetic resonance imaging system, comprising: a memory storing machine-executable instructions and an image generation neural network, the image generation neural network outputting synthetic magnetic resonance image data when receiving reference magnetic resonance image data as input, the image generation neural network generating the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system, the memory further including first pulse sequence commands configured to control the magnetic resonance imaging system to acquire measured k-space data, and the memory further including second pulse sequence commands configured to control the magnetic resonance imaging system to acquire reference k-space data; a computing system for controlling the medical system, wherein execution of the machine-executable instructions causes the computing system to: acquiring reference k-space data by controlling the magnetic resonance imaging system with a second pulse sequence command; reconstructing reference magnetic resonance image data from the reference k-space data; receiving synthetic magnetic resonance image data by inputting reference magnetic resonance image data into an image generation neural network; constructing synthetic k-space data using the synthetic magnetic resonance image data; and controlling acquisition of measured k-space data using the first pulse sequence command and the synthesized k-space data.

[0098] Clause 32. The magnetic resonance imaging system of clause 31, wherein execution of the machine-executable instructions causes the computing system to control acquisition of measured k-space data by selecting a k-space sampling pattern for a first pulse sequence command using the composite k-space data.

[0099] Clause 33. The first pulse sequence command is configured to control the magnetic resonance imaging system to acquire measurement k-space data in the k-space data group, and execution of the machine-executable instructions causes the computing system to further: calculating a comparison metric between the composite k-space data and each group of k-space data; 33. The magnetic resonance imaging system of clause 31 or 32, wherein if the comparison metric is outside a predetermined range of values, performing a predetermined action.

[0100] Article 34. 34. The magnetic resonance imaging system of claim 33, wherein the predetermined action is reacquiring at least a portion of the k-space data group, stopping acquisition of the measurement k-space data, and any combination thereof.

[0101] One or more of the above embodiments of the present invention may be combined, unless the combined embodiments are inconsistent.

[0102] As will be appreciated by those skilled in the art, aspects of the present invention may be embodied as an apparatus, a method, or a computer program product. Accordingly, aspects of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects (all of which may be referred to generically herein as a "circuit," "module," or "system"). Furthermore, aspects of the present invention may take the form of a computer program product embodied by one or more computer-readable medium(s) having computer-executable code embodied thereon.

[0103] Any combination of one or more computer-readable media may be utilized. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. As used herein, the term "computer-readable storage medium" may encompass any tangible storage medium capable of storing instructions executable by a processor or a computing system of a computing device. The computer-readable storage medium may also be referred to as a computer-readable non-transitory storage medium. The computer-readable storage medium may also be referred to as a tangible computer-readable medium. In some embodiments, the computer-readable storage medium may be capable of storing data accessible by the computing system of a computing device. Examples of computer-readable storage media include, but are not limited to, floppy disks, magnetic hard disk drives, solid-state hard disks, flash memory, USB thumb drives, random access memory (RAM), read-only memory (ROM), optical disks, magneto-optical disks, and computing system register files. Examples of optical disks include compact disks (CDs) and digital versatile disks (DVDs), such as CD-ROM, CD-RW, CD-R, DVD-ROM, DVD-RW, or DVD-R disks. The term computer-readable storage medium also refers to various types of storage media that a computing device can access over a network or communications link. For example, data may be retrieved over a modem, the Internet, or a local area network. Computer-executable code embodied in a computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, fiber optic cable, RF, etc., or any suitable combination thereof.

[0104] A computer-readable signal medium may include a propagated data signal that contains computer-executable code (e.g., in baseband or as part of a carrier wave). Such a propagated signal may take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium is not a computer-readable storage medium, but may be any computer-readable medium that can communicate, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0105] "Computer memory" or "memory" is one example of a computer-readable storage medium. Computer memory is any memory directly accessible by a computing system. "Computer storage" or "storage" is another example of a computer-readable storage medium. Computer storage is any non-volatile computer-readable storage medium. In some embodiments, computer storage can also be computer memory, and vice versa.

[0106] As used herein, a "computing system" encompasses electronic components capable of executing programs, machine-executable instructions, or computer-executable code. References to a computing system, including examples of a "computing system," should be interpreted as potentially including multiple computing systems or processing cores. A computing system may be, for example, a multi-core processor. A computing system may also refer to a collection of multiple processors aggregated in a single computing system or distributed among multiple computing systems. The term computing system should also be interpreted as meaning a collection or network of multiple computing devices, each containing one or more processors or computing systems. Machine-executable code or instructions may be executed by multiple computing systems or processors aggregated on the same computing device or distributed across multiple computing devices.

[0107] Machine-executable instructions or computer-executable code may include instructions or programs that cause a processor or other computing system to perform aspects of the present invention. Computer-executable code for carrying out operations of aspects of the present invention may be written in any combination of one or more of object-oriented programming languages ​​such as Java, Smalltalk®, C++, and conventional procedural programming languages ​​such as C, or similar programming languages, and compiled into machine-executable instructions. In some cases, the computer-executable code may be in the form of a high-level language or pre-compiled, and may be used in conjunction with an interpreter that generates machine-executable instructions on the fly. In other examples, the machine-executable instructions or computer-executable code may be in the form of programming for a programmable logic gate array.

[0108] The computer executable code may run entirely on the user computer, partially on the user computer, as a standalone software package, partially on the user computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or a connection may be established to an external computer (e.g., via the Internet using an Internet Service Provider).

[0109] Aspects of the present invention will be described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block or group of blocks in the flowcharts, illustrations, and / or block diagrams may, where appropriate, be implemented by computer program instructions in the form of computer-executable code. It will also be understood that blocks in different flowcharts, illustrations, and / or block diagrams may be combined where not mutually inconsistent. These computer program instructions may be provided to a computing system, such as a general-purpose computer, a special-purpose computer, or other programmable data processing device, to form a machine. The instructions, executed via the computing system of the computer or other programmable data processing device, create means for implementing the functions / acts identified in one or more blocks of the flowcharts and / or block diagrams.

[0110] These machine-executable instructions or computer program instructions may be stored on a computer-readable medium that can cause a computer, other programmable data processing apparatus, or other device to function in a particular manner. The instructions stored on the computer-readable medium create an article of manufacture including instructions that implement the functions / acts identified in one or more blocks of the flowcharts and / or block diagrams.

[0111] The machine-executable instructions or computer program instructions may be loaded onto a computer, other programmable data processing apparatus, or other device and cause the computer, other programmable apparatus, or other device to perform a series of operational steps to produce a computer-implemented process. The instructions executing on the computer or other programmable apparatus provide a process for implementing the functions / operations identified in one or more blocks of the flowcharts and / or block diagrams.

[0112] As used herein, a "user interface" is an interface that allows a user or operator to interact with a computer or computer system. A "user interface" may also be referred to as a "human interface device." A user interface may provide information or data to an operator and / or receive information or data from an operator. A user interface may allow a computer to receive input from an operator and may provide output from the computer to a user. In other words, a user interface may allow an operator to control or manipulate a computer, and an interface may allow a computer to display the effects of the operator's control or manipulation. Displaying data or information on a display or graphical user interface is an example of providing information to an operator. Receiving data via a keyboard, mouse, trackball, touchpad, pointing stick, graphics tablet, joystick, gamepad, webcam, headset, pedals, wired gloves, remote control, and accelerometer are all examples of user interface elements that allow receiving information or data from an operator.

[0113] As used herein, the term "hardware interface" encompasses an interface that allows a computing system of a computer system to interact with and / or control external computing devices and / or equipment. A hardware interface may allow a computing system to send control signals or instructions to external computing devices and / or equipment. A hardware interface may also allow a computing system to exchange data with external computing devices and / or equipment. Examples of hardware interfaces include, but are not limited to, a universal serial bus, an IEEE 1394 port, a parallel port, an IEEE 1284 port, a serial port, an RS-232 port, an IEEE-488 port, a Bluetooth connection, a wireless local area network connection, a TCP / IP connection, an Ethernet connection, a control voltage interface, a MIDI interface, an analog input interface, and a digital input interface.

[0114] As used herein, "display" or "display device" encompasses an output device or user interface adapted to display images or data. A display may output visual, auditory, or tactile data. Examples of displays include, but are not limited to, computer monitors, television screens, touch screens, tactile electronic displays, Braille screens, These include cathode ray tubes (CRTs), storage tubes, bistable displays, electronic paper, vector displays, flat panel displays, fluorescent displays (VFs), light emitting diode (LED) displays, electroluminescent displays (ELDs), plasma display panels (PDPs), liquid crystal displays (LCDs), organic light emitting diode displays (OLEDs), projectors, and head-mounted displays.

[0115] As used herein, k-space data is defined as the measurement of radio frequency signals emitted by atomic spins using the antenna of a magnetic resonance machine during a magnetic resonance imaging scan. Magnetic resonance data is an example of medical tomographic image data.

[0116] As used herein, a magnetic resonance imaging (MRI) image, MR image, or magnetic resonance imaging data is defined as a reconstructed two-dimensional or three-dimensional visualization of anatomical data contained within magnetic resonance imaging data, which visualization may be performed using a computer. [Brief explanation of the drawings]

[0117] Preferred embodiments of the present invention will now be described, by way of example only, with reference to the following drawings: [Figure 1] FIG. 1 shows an example of a medical system. [Figure 2] FIG. 2 shows a flow chart illustrating a method of operation of the medical system of FIG. [Figure 3] FIG. 3 shows a further example of a medical system. [Figure 4] FIG. 4 shows a flow chart illustrating a method of operation of the medical system of FIG. [Figure 5] FIG. 5 shows a further example of a medical system. [Figure 6] FIG. 6 shows a flow chart illustrating a method of operation of the medical system of FIG. [Figure 7] FIG. 7 shows an example of a fully sampled MR image. [Figure 8] FIG. 8 shows an example of measured k-space data. [Figure 9] FIG. 9 shows a compressed sensing reconstruction of the measured k-space data of FIG. [Figure 10] FIG. 10 shows an example of reference magnetic resonance image data. [Figure 11] FIG. 11 shows an example of the composite magnetic resonance image data 128. [Figure 12]FIG. 12 shows an example of a corrected magnetic resonance image 132. [Figure 13] FIG. 13 shows an example of the method. [Figure 14] FIG. 14 shows an example of reference magnetic resonance image data. [Figure 15] FIG. 15 shows an example of the composite magnetic resonance image data. [Figure 16] FIG. 16 shows a magnetic resonance image with motion artifacts caused by intentionally corrupting some lines of k-space data. [Figure 17] FIG. 17 shows an example of corrected magnetic resonance image data. [Figure 18] FIG. 18 shows the correct magnetic resonance image. [Figure 19] FIG. 19 shows the relative change in estimated artifact level after replacing individual shots with the contrast transformation data shown in FIG. [Figure 20] FIG. 20 shows a flow chart illustrating various methods for reconstructing corrected magnetic resonance image data. DETAILED DESCRIPTION OF THE INVENTION

[0118] Elements with like numbers in the figures are equivalent elements or perform the same function. An element described earlier is not necessarily described in a later figure if the function is equivalent.

[0119] FIG. 1 illustrates an example medical system 100. The medical system of FIG. 1 is shown to include a computer having a computing system 106. The computing system 106 is intended to represent one or more computing systems, such as processors or cores located in one or more locations. The computing system 106 is shown to be connected to an optional interface 104. If other components of the medical system 100, such as a magnetic resonance imaging system, are present, the computing system 106 may be used by the hardware interface 104 to communicate with and control the other components. The medical system 100 is further shown to include an optional user interface 108 that may allow an operator to use and control the medical system 100. The medical system 100 is further shown to include a memory 110 also connected to the computing system 106. The memory 110 is intended to represent any memory or storage connected to the computing system 106.

[0120] The memory 110 is shown as including machine-executable instructions 120. The machine-executable instructions 120 may enable the processor 106 to perform various image processing, data processing, and control functions. The memory 110 is further shown as including an image generation neural network. The image generation neural network 122 is configured to receive a reference magnetic resonance image and then output synthetic magnetic resonance image data 128. The reference magnetic resonance image data 126 is acquired or configured according to a second configuration of the magnetic resonance imaging system, and the synthetic magnetic resonance image data 128 is a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system.

[0121] Thus, the image generation neural network 122 may enable the use of previously acquired data to control or improve the generation of corrected magnetic resonance image data. The memory 110 is further shown as including examples of reference magnetic resonance image data 126 and output composite magnetic resonance image data 128. Once acquired, the composite magnetic resonance image data 128 may optionally be used to calculate composite k-space data 130. For example, knowledge of a first configuration of the magnetic resonance imaging system may enable the calculation of composite k-space data 130 that is sampled in a manner similar to the sampling of the measured k-space data 124. Measured k-space data 124 acquired by the magnetic resonance imaging system using the first configuration is also shown as stored in the memory 110.

[0122] The memory 110 is further shown as including corrected magnetic resonance imaging data 132, which may be calculated, for example, using the measured k-space data 124 and either synthetic k-space data 130 or synthetic magnetic resonance image data 128. The synthetic k-space data 130 may be used to modify or replace portions of the measured k-space data 124. In other examples, the synthetic magnetic resonance image data 128 may be used as prior knowledge to improve the reconstruction of the corrected magnetic resonance image data 132 from the measured k-space data 124.

[0123] The memory 110 is also shown as including an optional image processing module 134. This module may be used, for example, to condition the reference magnetic resonance image data 126 so that it has a predetermined image format before being input to the image generation neural network 122. Similarly, the image processing module 134 may also be used to configure or modify the composite magnetic resonance image data 128 to spatially match the measured k-space data 124.

[0124] 2 shows a flowchart illustrating a method of operation of the medical system 100 of FIG. 1. First, in step 200, k-space data 124 is received. Next, in step 202, magnetic resonance image data 126 is also received. Next, in step 204, synthetic magnetic resonance image data 128 is obtained by inputting the reference magnetic resonance image data 126 into the image generation neural network 122. Step 200 may be performed after steps 202 or 204. The synthetic magnetic resonance image data 128 or synthetic k-space data 130 is then used in step 206. In step 206, corrected magnetic resonance image data 132 is reconstructed using the measured k-space data 124 and the synthetic magnetic resonance image data 128, or alternatively, the synthetic k-space data 130.

[0125] Figure 3 shows a further example of a medical system 300. The medical system shown in Figure 3 is similar to the medical system 100 of Figure 1, except that it additionally includes a magnetic resonance imaging system 302.

[0126] The magnetic resonance imaging system 302 includes a magnet 304. The magnet 304 is a superconducting cylindrical magnet with a bore 306 extending therethrough. Different types of magnets can be used. For example, both split cylindrical magnets and so-called open magnets can be used. Split cylindrical magnets are similar to standard cylindrical magnets, except that the cryostat is split into two sections to allow access to the magnet's isoplane. Such magnets can be used in conjunction with charged particle beam therapy, for example. Open magnets have two magnet sections, one positioned above the other to provide sufficient space to accommodate a subject between them, an arrangement similar to that of a Helmholtz coil. Open magnets are popular because they provide a less confined space for the subject. Inside the cryostat of the cylindrical magnet is a collection of superconducting coils.

[0127] Within the bore 306 of the cylindrical magnet 304 is an imaging zone 308, where a magnetic field exists that is strong enough and uniform to perform magnetic resonance imaging. A region of interest 309 is shown within the imaging zone 308. Magnetic resonance data is typically acquired about the region of interest. A subject 318 is shown supported by a subject support 320 such that at least a portion of the subject 318 is within the imaging zone 308 and the region of interest 309.

[0128] Also present within the magnet bore 306 are a set of magnetic field gradient coils 310 used for preliminary magnetic resonance data acquisition to spatially encode magnetic spins within the imaging zone 308 of the magnet 304. The magnetic field gradient coils 310 are connected to a magnetic field gradient coil power supply 312. It should be understood that the magnetic field gradient coils 310 are representative. Typically, the magnetic field gradient coils 310 include three separate coil sets for spatial encoding in three orthogonal spatial directions. The magnetic field gradient power supply supplies current to the magnetic field gradient coils 310. The current supplied to the magnetic field gradient coils 310 can be controlled as a function of time and can be ramped or pulsed.

[0129] Adjacent to the imaging zone 308 is a radio frequency coil 314 for manipulating the orientation of magnetic spins within the imaging zone 308 and for receiving radio signals from the spins within the imaging zone 308. A radio frequency antenna may include multiple coil elements. A radio frequency antenna may also be referred to as a channel or antenna. The radio frequency coil 314 is connected to a radio frequency transceiver 316. The radio frequency coil 314 and the radio frequency transceiver 316 may be replaced by separate transmit and receive coils and separate transmitters and receivers. It should be understood that the radio frequency coil 314 and the radio frequency transceiver 316 are representative. The radio frequency coil 314 is also intended to represent a dedicated transmit antenna and a dedicated receive antenna. Similarly, the transceiver 316 may also represent separate transmitters and receivers. The radio frequency coil 314 may also have multiple receive / transmit elements, and the radio frequency transceiver 316 may have multiple receive / transmit channels. For example, if a parallel imaging technique such as SENSE is performed, the radio frequency coil 314 has multiple coil elements.

[0130] The transceiver 316 and gradient controller 312 are shown as connected to the hardware interface 106 of the computer system 102 .

[0131] The memory 110 is further shown as including a first pulse sequence command 330 configured to acquire measured k-space data 124 while the magnetic resonance imaging system 302 is in a first configuration. A second pulse sequence command 332 is configured to acquire reference k-space data 334 when the magnetic resonance imaging system 302 is in a second configuration. The memory 110 is further shown as including reference k-space data 334 acquired when the second pulse sequence command 332 is executed. The measured k-space data 124 may be acquired when the first pulse sequence command 330 is acquired.

[0132] In some examples, the reference k-space data 334 and the measured k-space data 124 may be acquired for the same subject 318 at different times, or possibly on different magnetic resonance imaging systems 302. In this example, both are acquired during the same examination. For example, both may be acquired for the same region of interest 309 and may be acquired to be spatially coincident.

[0133] 4 shows a flowchart illustrating a method of operation of the medical system 300 of FIG. 3. First, in step 400, reference k-space data 334 is acquired by controlling the magnetic resonance imaging system with second pulse sequence commands 332. Next, in step 402, reference magnetic resonance image data 126 is reconstructed from the reference k-space data 334. Next, in step 404, measured k-space data 124 is acquired by controlling the magnetic resonance imaging system 302 with first pulse sequence commands 330. After step 404, the method proceeds to and executes steps 200, 202, 204, and 206, as shown in FIG. 2.

[0134] 5 shows an example of a magnetic resonance imaging system 500. The magnetic resonance imaging system 500 is similar to the medical system 300 of FIG. 3, except for the contents of the memory 110. In this example, the machine-executable instructions 120 are configured such that the synthetic k-space data 130 is used to correct the acquisition of the measured k-space data 124. This may be useful, for example, to compensate for movement of the subject 318, imperfections in various channels of the radio frequency system, or noise received by the RF antenna 314.

[0135] The memory 110 is shown as containing a corrected magnetic resonance image 502 constructed from the measured k-space data 124. The composite k-space data 130 can be used in several different ways to correct the measured k-space data 124. For example, the composite k-space data 130 can be used to select a sampling pattern for the first pulse sequence command 330, which effectively selects sample locations for the measured k-space data 124. In another example, the composite k-space data 130 can be compared to a shot or group of measured k-space data 124 as they are being acquired and used to correct the acquisition or adjust the measured k-space data 124. This can be done on the fly or after all of the measured k-space data 124 has been acquired. The features of FIG. 5 can be combined with those of FIGS. 1 and 3.

[0136] FIG. 6 shows a flowchart illustrating a method of operation of the magnetic resonance imaging system 500 of FIG. 5. First, in step 600, reference k-space data 334 is acquired by controlling the magnetic resonance imaging system 500 with the second pulse sequence commands 332. Next, in step 602, reference magnetic resonance image data 126 is reconstructed from the reference k-space data 334. Next, in step 604, synthetic magnetic resonance image data 128 is acquired by inputting the reference magnetic resonance image data 126 into the image generation neural network 122. Next, in step 606, synthetic k-space data 130 is constructed from the synthetic magnetic resonance image data 128. This may be constructed using, for example, an inverse Fourier transform. Finally, in step 608, the magnetic resonance imaging system is controlled using the first pulse sequence commands 330 to acquire measured k-space data 124. In step 608, the acquisition is also controlled or adjusted using the synthetic k-space data 130.

[0137] Due to the inherent speed limitations of MRI acquisition, numerous image reconstruction techniques have been investigated to enable the acquisition of good image quality from undersampled k-space data. The most prominent of these techniques are parallel imaging (PI), compressed sensing (CS), and their combination, PI-CS. Recently, it has been shown that the use of neural networks and deep learning (DL) allows PI-CS to reach even higher acceleration rates while maintaining image quality. This is possible because neural networks can better capture the low-dimensional space of MRI images, learning from large datasets containing images of many other patients during training.

[0138] Some examples disclosed herein take advantage of the fact that in an MRI examination, multiple scans of the same anatomical structure are typically acquired with different contrasts (configurations of the magnetic resonance imaging system), where "contrast" as used herein refers to the configuration of the magnetic resonance imaging system used to acquire the k-space data.

[0139] These different contrasts contain common information (e.g., the same patient, the same pathology), and can be utilized in compressed sensing reconstruction if a good model is available that describes how the different contrasts are correlated. Such CS reconstruction is called multi-contrast CS. Previous multi-contrast CS methods have used simple analytical models to represent the correlation between contrasts. Here, we propose to use neural networks to learn this correlation from real data, allowing us to reach higher acceleration factors while maintaining high image quality.

[0140] The acceleration methods described above are based on the use of prior information. In parallel imaging, coil sensitivity can be considered a type of prior, while in compressed sensing, image sparsity is the prior (either general to all types of images (e.g., in the case of wavelet-based CS) or general to MRI images (facilitated by networks that can learn representations, e.g., trained with large amounts of MRI data, DL-CS). The performance advantage of DL-CS over CS is due to the fact that the prior information used is better adapted, since it is more specific to the type of image being reconstructed. A logical further step that would lead to better prior knowledge would be to include patient-specific information. This is the approach adopted in multi-contrast CS (MC-CS), in which multiple images of the same anatomical structure acquired with different contrasts are reconstructed simultaneously or sequentially, taking into account previous reconstructions. This is motivated by the fact that these different contrasts contain visible, correlated information (see Figures 7 and 10, discussed below).

[0141] The main problem with MC-CS is the difficulty of modeling the information shared between contrasts. In what follows, a Bayesian estimation setting is used to motivate the mathematical formulation of the problem. However, other approaches can also be adopted to justify the mathematical formulation of the MC-CS problem.

[0142] Given an image to be reconstructed, x, with undersampled measurements and an (undersampled multi-coil) measurement operator, A, the Bayesian CS is the estimate of x.

number

number

[0143] where p x (x) is the prior distribution, ∝ is the proportionality up to a constant, and in the case of additive white Gaussian noise with variance λ for the measurement results, the posterior distribution p y|x (y|x) can be given as proportional to a Gaussian distributed estimate of the residual |y-Ax| (data minus an undersampling operator applied to the inferred image) normalized to the noise λ, as follows:

number

[0144] Although the true prior distribution is not known for real MR images, assuming a Laplacian distribution on the wavelet transform Ψx of x gives good sparse results in wavelet space.

number

number

[0145] A corresponding formulation can be created for multi-contrast compressed sensing. Here, we consider a set of two images x and x' with different contrasts. We assume that a good reconstruction of x' is already available and we want to use it to reconstruct x from undersampled measurements y. Taking x' into account, we obtain the following posterior distribution:

number

[0146] and the corresponding minimization problem is obtained:

number

[0147] The difficulty is p x (x) and p x|x’ However, this Bayesian derivation encourages us to tackle the MC-CS problem by solving a minimization problem of the general form:

number

[0148] Note that this formulation can be directly extended to more than two contrasts, or joint reconstruction of more than two contrasts. In general, x and x' are not perfectly aligned with each other as a result of patient motion between scans or mismatches in scan resolution, field of view, and planning.

[0149] As a first approximation, we assume that the images are perfectly aligned with each other. One possible assumption is that x and x' are sparse in the same basis T, and that their supports in that basis overlap highly. This means that

number

[0150] For example, in a neural network, the distribution p x or product p x (x)p x|x’ One approach is to encode (x|x') and use it as a building block for iterative CS reconstruction that solves various minimization problems inspired by equation (1). (1) A contrast-to-contrast network N (image generating neural network 122) can be trained that takes as input an image with well-defined contrast A (reference magnetic resonance image data 126) and outputs an estimate of the corresponding image (synthetic magnetic resonance image data 128) with well-defined contrast B (first configuration of the magnetic resonance imaging system). This network can be trained using a training dataset of pairs of the same images with contrast A and B and a loss function such as MSE. Alternatively, it is possible to use a conditional CycleGAN to use unpaired datasets of images with contrast A and other images with contrast B from different patients. During multi-contrast reconstruction, the network is applied once to image x', and estimates

number

number

number

[0151] where Ψ and T are appropriately chosen sparsifying transformations, λ and μ are tunable regularization parameters, and p is either 1 or 2, resulting in a tractable equation. (2) Train one or more networks N that take as input two images stacked in two different channels (four channels for complex-valued images). The first image (x') is a clean image from well-defined contrast A, and the second image is an artifacted version of the corresponding image from well-defined contrast B. The networks are used at each iteration of the following type of iterative CS reconstruction:

number

[0152] There may be a single network N that is trained once, or there may be as many networks as there are iterations to be performed, and training is done by combining the ground truth image x and the final estimation result

number

number

number

[0153] The methods described in items (1) and (2), when enhanced as described in (3), can also be made robust to motion and / or changes in resolution or FOV between the two contrasts by using rigid or non-rigid motion estimation and transformations, as is done in some CS reconstructions with an additional time dimension. For the method described in (3), this motion estimation is performed on the undersampled image A T It may be performed once from y and the combined contrast N(x'), or estimate

number

[0154] The example application is the acceleration of a scanning protocol involving the acquisition of several different contrasts of the same anatomical structure. While the technique described here is limited to sequential reconstruction of contrasts (rather than joint simultaneous reconstruction of several undersampled contrasts), it can be easily extended to more than two contrasts, or even joint reconstruction of contrasts. The entire protocol can be optimally accelerated by starting with the acquisition of a high-SNR "fast" contrast, which is then used as a reference contrast in the reconstruction of subsequent contrasts, which are acquired more slowly but can be further accelerated thanks to MC-CS. The acceleration rate achievable using MC-CS is expected to be close to that achieved with dynamic CS (approximately two times higher than conventional CS).

[0155] Inter-scan motion issues can be addressed as described in (4) above. However, MC-CS is particularly suited to 3D scan sequences, as through-plane motion in multi-slice scans can be problematic in this manner. Another way to mitigate motion issues is to consider interleaved scans, which can also further reduce scan time. However, by reducing scan time, MC-CS already helps to improve patient comfort while reducing potential motion artifacts through shorter overall scan times.

[0156] Example 1: In one possible implementation of variant (1), the contrast-to-contrast network is trained with data for two well-defined contrasts, A and B, which may or may not be matched. The scan protocol includes one sequence to acquire contrast A and one sequence to acquire contrast B, or a single sequence to acquire k-space profiles of both contrasts in an interleaved manner. Contrast A is first reconstructed using a conventional method such as PI, CS, or CS-PI to obtain a high-quality image x'. This image x' is fed into the contrast-to-contrast network, which then estimates the contrast.

number

number

number

[0157] where Ψ is the wavelet transform and λ is an adjustable regularization parameter. As shown in Figure 2, the resulting MC-CS reconstruction has much better image quality than the standard CS reconstruction.

[0158] 7-12 are used to illustrate the use of an image generation neural network 122 to assist in compressed sensing reconstruction. In FIG. 7, a fully sampled image 700 is shown. In this example, a portion of the k-space data is used to reconstruct the image 700. FIG. 8 shows an example of undersampled measured k-space data 124. This is the portion of the k-space data used to reconstruct the image 700 of FIG. 7.

[0159] FIG. 9 illustrates a wavelet compressive sensing reconstruction 900. It can be seen that the undersampled k-space data 124 of FIG. 8 was insufficient to reconstruct a high-quality image. FIG. 10 illustrates different contrast, or reference, magnetic resonance image data 126. Comparing image 700 and image 126, it can be seen that the two images are of the same anatomical structure. FIG. 11 illustrates synthetic magnetic resonance image data 128 generated from the reference magnetic resonance image data 126 using the image generation neural network 122. The synthetic magnetic resonance image data 128 is then used with the measured k-space data 124 of FIG. 8 to reconstruct the corrected magnetic resonance image data 132 shown in FIG. 12. In this example, the synthetic magnetic resonance image data 128 of FIG. 11 was used in the regularization term for the compressive sensing reconstruction.

[0160] In other words, Figures 7-12 are illustrations of MC-CS using a contrast-to-contrast network. Figures 7 and 10 are images representing two contrasts, A and B. While the image in Figure 10 is available from a previous scan, Figure 7 represents an image reconstructed from the measured undersampled k-space data shown in Figure 8. Figure 7 includes an artificially brightened central region 702 to demonstrate the effectiveness of the algorithm. Without a reference contrast, the CS reconstruction shown in Figure 9, which uses sparsity in the wavelet basis, still contains strong undersampling artifacts. In the proposed approach, the trained contrast-to-contrast network starts from Figure 126 and generates the estimated result 128 shown in Figure 11 of the image 700 in Figure 7. Using this estimated result in the MC-CS reconstruction yields Figure 12, which has significantly improved image quality compared to Figure 9. Structures that appear in Figure 7 but not in Figure 10 (e.g., the artificial central region 702) are preserved in the reconstruction in Figure 12.

[0161] Example 2: The contrast-to-contrast network used in method (1) and the different networks used in method (2) can have a variety of architectures, including fully convolutional networks (FCNs) such as U-net and its variations. In the case of (1), training can be performed from a training dataset of unmatched contrast A and B images.

[0162] An additional topic is discussed below: the use of image-generating neural networks to reduce motion artifacts. Image degradation due to subject motion during acquisition is a persistent problem in clinical applications of magnetic resonance imaging (MRI). The associated artifacts typically appear in images as ghosting or blurring, often reducing image quality to the point where medical analysis is impossible. However, in many cases, only a subset of the total scans in an examination exhibit motion artifacts, and many patients exhibit various forms of movement during different parts of the examination. Additionally, some MR sequences are more sensitive to motion than others.

[0163] Many strategies for mitigating motion artifacts in MR are based on estimating the underlying motion trajectory. This typically involves applying a parameterized motion model, such as a rigid 3D model for brain scans. This can be problematic when the actual patient motion differs from this model, such as swallowing in a brain scan. Alternatively, portions of k-space can be rejected and the missing data points reconstructed using the data redundancy of multi-coil acquisitions (i.e., SENSE-based reconstruction). However, this inevitably incurs a noise penalty. The approach described in this disclosure avoids both drawbacks by leveraging information from an artifact-free second scan.

[0164] The invention described above can be applied when multiple scans are acquired during an examination and at least one of these scans is identified as motion artifact-free. While this identification of artifact-free scans can be performed manually by an operator, it is also possible to automate this step using a dedicated metric, for example, based on a neural network trained to estimate the level of motion artifacts in an image. The latter was implemented and tested as part of a proof-of-concept (POC) study included below.

[0165] FIG. 13 illustrates a method for reducing motion artifacts in corrected magnetic resonance image data 132 using synthetic k-space data 130. Block 126 represents reference magnetic resonance image data, in this case the first scan without motion. Block 122 is a contrast-to-contrast transformation U-Net neural network, which corresponds to the image generation neural network 122. The output of this neural network 122 is synthetic magnetic resonance image data 128, referred to in this figure as the transformed first scan. The second scan, corrupted by motion, corresponds to measured k-space data 124. This is used in an algorithm to replace the profile of k-space 1300. The result is corrected magnetic resonance image data 132. This may be performed multiple times. If this is performed multiple times, the fused image 132 is an intermediate image. This is then input to an artifact level estimator 1302 or image quality estimation module. Block 1304 represents an algorithm step in which the particular replaced k-space profile is retained if the artifact level is reduced. Step 1306 represents repeating this for all profiles, for a specific number of k-space profiles, or for a combination of k-space profiles.

[0166] In Figure 13, assuming that the first scan of the examination (reference magnetic resonance image data 126) is detected as a motionless artifact, this scan is transformed into a target contrast (synthetic magnetic resonance image data 128) using a dedicated contrast transformation network (image generation neural network 122). In the first POC study, a U-Net architecture was used to realize this image transformation module, but other architectures are possible. Creating a suitable dataset can be achieved in various ways, such as: - Identifying artifact-free scan pairs with identical geometry in the clinical database and creating a database using the alignment of both scans if necessary. - To enable forward simulation of arbitrary MR contrasts, a quantitative data set is acquired, including tissue parameter maps, i.e., proton density, T1, and T2 maps. To extend this method to functional MR sequences, additional tissue parameters such as diffusion and perfusion may be useful. - If matching scan pairs with identical geometry are not available, a large dataset of (non-corresponding) scans can also be used, in which case a CycleGAN network architecture can be used.

[0167] Using this trained contrast transformation network (image generation neural network 122), the first motionless scan is transformed into the target contrast, i.e., the contrast of the second scan, which is corrupted by motion artifacts. If the two scans do not have the same field of view and resolution, field of view adjustment and interpolation can be used to match the first contrast to the geometry of the second scan. If necessary, an image registration algorithm can be used to take into account possible patient motion between the two scans. In either case, a composite k-space of the contrast-transformed, aligned first scan is generated using a Fourier transform and coil sensitivity maps.

[0168] To reduce the artifact level of this second scan, certain k-space profiles from the second scan are then replaced with the corresponding k-space profiles from the transformed first scan. The choice of which profile to replace depends on the type of scan and the specific k-space acquisition method; in the case of a standard sequential Cartesian method, only a single profile may be replaced. In POC studies where an interleaved TSE-like acquisition was assumed, all profiles corresponding to a single TSE shot were replaced (this corresponds to the assumption of negligible motion during each shot).

[0169] After each k-space profile replacement, the resulting "fused" dataset is Fourier transformed to obtain a fused image in the image domain. The artifact level of the resulting fused image is estimated using a dedicated motion artifact level estimator. Various implementations of this module are possible, including traditional metrics such as total image gradients and image entropy. In the proof-of-concept study, a dedicated recurrent convolutional neural network (CNN) was trained to estimate the L2 norm of artifacts in the image. The generation of the associated training dataset was achieved based on T2w images of motionless volunteers and an artifact simulation pipeline. If the estimated artifact level of the fused image is significantly lower than that of the original images, the profile is considered motion-corrupted.

[0170] Once the entire k-space has been analyzed, any profiles deemed corrupted are replaced with their counterparts in the transformed motion-free data set. In the final Fourier transform and coil combination step, an artifact-corrected image is generated.

[0171] Figures 14-18 demonstrate the effectiveness of the method shown in Figure 13. Figure 14 shows an example of magnetic resonance image data 126 with a synthetic lesion 1400. Figure 15 shows an example of synthetic magnetic resonance image data 128 generated from image 126 of Figure 14. Figure 16 shows another magnetic resonance image with motion artifacts caused by intentionally corrupting some lines of k-space data. Figure 17 shows an example of corrected magnetic resonance image data 132 reconstructed using the method shown in Figure 13 using the measured k-space data 124 of Figure 16 and the synthetic magnetic resonance image data 128 used to replace some of the k-space data. This image is compared to Figure 18, which shows a ground truth image 1800 containing the same k-space data used to generate image 124 of Figure 16, except that the k-space lines have not been artificially corrupted. It can be seen that the images of Figures 17 and 18 match very well.

[0172] In general, contrast-to-contrast conversion using neural networks cannot be expected to be completely error-free, since the underlying tissue properties cannot be fully inferred from a single scan. As an illustrative example, the first PD-weighted contrast in Figure 14 contains a synthetic lesion. This lesion can also be seen in the network-based contrast conversion results shown in Figure 15. To demonstrate the robustness of the described method to errors during the contrast conversion process, the ground truth image for the second contrast did not contain a lesion (Figure 18). Based on this ground truth image, forward simulation was used to generate the artifact-corrupted image in Figure 16. Here, a TSE acquisition with 16 shots (256 profiles, TSE factor = 16) was assumed, with two shots corrupted by a 10° in-plane rotation.

[0173] Figure 19 below shows the relative change in estimated artifact level after replacement of individual shots with the contrast transformed data shown in Figure 15. Shots #0 and #7 are correctly identified as motion corrupted, as shown by the significant drop in estimated artifact level after replacement. Importantly, no synthetic lesions are visible in the artifact-corrected results shown in Figure 17, even though two shots (=12.5%) in k-space were replaced by transformed data.

[0174] In general, the effect of displaced k-space lines on this "error propagation" depends on the k-space trajectory, the number of displaced lines, etc. Empirical tests can be easily performed to determine an upper limit for k-space displacement.

[0175] Other features: For the design and application of the above described examples, additional features such as the following may be considered: To increase the accuracy of the contrast transformation, multiple scans can be used as input to the contrast transformation network, if available. To avoid training a dedicated contrast transformation network for each change in scan parameter settings (e.g., changes in TE and TR), the transformation network can be made to incorporate these scan settings as additional inputs. One possibility for such a design is to include an adaptive instance normalization (AdaIn) layer in the network. To avoid a brute-force search for corrupting locations, a determination of k-space locations affected by motion can be performed in advance, for example, using external sensors that track the patient's motion over time (respiratory belt, end-of-bore camera) or by exploiting discrepancies in the raw data, which are possible due to the data redundancy of multi-coil scans. The applicability of this method is not limited to Cartesian scans, but is valid for any combination of multiple k-space trajectories.

[0176] FIG. 20 shows a flowchart illustrating various methods for reconstructing corrected magnetic resonance image data 132. The steps are divided into several major steps. Step 2000 represents the acquisition of measured k-space data 124 and reference k-space data 334. In the next step, a standard reconstruction 2002 is performed. In this step, an unused or corrupted image may be reconstructed directly from the measured k-space data 124, but this is not employed in this method. Reference magnetic resonance image data 126 is reconstructed from the reference k-space data 334. The image 126 is then input to an image generation neural network 122 or a contrast-to-contrast network to generate an estimated image or synthetic magnetic resonance image data 128. Before inputting the reference magnetic resonance image data 126 into the neural network 122, geometry correction may be performed using a geometry correction module 134. Similarly, after the data is output from the neural network 122, geometry correction using the geometry correction module 134 may change the form of the estimated image or synthetic magnetic resonance image data 128. The estimated image 128 may also be used to generate estimated or synthetic k-space data 130. The steps within level 2006 represent several different pre-processing options.

[0177] For example, there may be a globally corrupted k-space, in which case there is a correction model that can be used. For example, in an EPI magnetic resonance image acquisition, even and odd echoes may be misaligned. This can be detected and corrected using estimated k-space data. Another pre-processing option may be locally corrupted k-space data. For example, if an error occurs, the k-space data may be discarded and simply filled, as shown in FIG. 13. Another pre-processing option is an incomplete k-space. In this example, there may be a choice to continue with an incomplete k-space or to perform a k-space fill. Depending on the pre-processing choice made in step 2006, there may also be a pre-processed k-space 2008.

[0178] In another example, the corrected k-space data may be generated as a hybrid k-space, which is a combination of both the estimated k-space 130 and the actually acquired k-space 124. The k-space may also be soft-gated, in which the replaced k-space data is given a lower weighting factor so that it has less impact on the final image. In another example, an image may be reconstructed using an incomplete k-space, or the profile may simply be discarded. Step 2010 represents multi-contrast reconstruction to generate the diagnostic image 132. This allows standard reconstruction to be used when the corrected k-space or hybrid k-space is used, and soft-gated reconstruction to be used when soft-gated weighting is used. If an incomplete k-space exists, it may be used as a regularization term for the estimated image 128.

[0179] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The invention is not limited to the disclosed embodiments.

[0180] Other variations of the disclosed embodiments can be understood and realized by those skilled in the art in practicing the claimed invention from the drawings, the disclosure, and the appended claims. In the claims, the terms "comprise" and "include" do not exclude other elements or steps, and the singular does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that several means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage. A computer program may be stored and / or distributed on a suitable medium, such as an optical storage medium or a solid-state medium, supplied together with or as part of other hardware, or in other forms, such as via the Internet or other wired or wireless telecommunications systems. Any reference signs in the claims should not be construed as limiting their scope. [Explanation of symbols]

[0181] 100 Medical Systems 102 Computer 104 Hardware Interface 106 Computing Systems 108 User Interface 110 memory 120 machine-executable instructions 122 Image Generation Neural Network 124 measured k-space data 126 Reference Magnetic Resonance Imaging Data 128 Synthetic Magnetic Resonance Image Data 130 synthetic k-space data 132 Corrected Magnetic Resonance Image Data 134 Image Processing Module 200 receiving measured k-space data acquired according to a first configuration of the magnetic resonance imaging system 202 receives reference magnetic resonance image data (the reference magnetic resonance image data represents a region of interest of the subject); 204 receiving synthetic magnetic resonance image data by inputting the reference magnetic resonance image data into an image generation neural network; 206 Reconstructing corrected magnetic resonance image data from measured k-space data and synthetic magnetic resonance image data 300 Medical Systems 302 Magnetic Resonance Imaging System 304 Magnet 306 Magnet Bore 308 Imaging Zone 309 Areas of Interest 310 Magnetic Gradient Coil 312 Magnetic field gradient coil power supply 314 Radio Frequency Coil 316 Transceiver 318 Subjects 320 Subject support platform 330 First Pulse Sequence Command 332 Second Pulse Sequence Command 334 Reference k-space data 400. Acquire reference k-space data by controlling the magnetic resonance imaging system with a second pulse sequence command. 402 Reconstructing reference magnetic resonance image data from reference k-space data 404. Acquire measured k-space data by controlling the magnetic resonance imaging system using a first pulse sequence command. 500 Magnetic Resonance Imaging System 502 Corrected Magnetic Resonance Images 600. Acquire reference k-space data by controlling the magnetic resonance imaging system with a second pulse sequence command. 602 Reconstructing reference magnetic resonance image data from reference k-space data 604 receives synthetic magnetic resonance image data by inputting the reference magnetic resonance image data into an image generation neural network. 606 Constructing synthetic k-space data using synthetic magnetic resonance image data 608 Control acquisition of measured k-space data using first pulse sequence command and synthesized k-space data 700 fully sampled images 900 Wavelet Reconstructed Images 1300 Image Quality Estimation Module 1400 Synthetic Lesions 1800 Correct Images 2000 acquired 2002 Standard Reconstruction 2004 contrast-to-contrast estimation 2006 Pretreatment Selection 2008 Preprocessed k-space Multi-contrast reconstruction of 2010 B 2016 Multi-contrast Reconstruction

Claims

1. 1. A medical system, comprising: a memory storing machine-executable instructions and access to an image generation neural network that outputs synthetic magnetic resonance image data when it receives reference magnetic resonance image data as input, and that generates the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system; a computing system that controls the medical system, wherein execution of the machine-executable instructions causes the computing system to: accessing measured k-space data acquired according to the first configuration of the magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; accessing the reference magnetic resonance image data representative of the region of interest of the subject; generating access to the synthetic magnetic resonance image data by inputting the reference magnetic resonance image data into the image generation neural network; and arranging for reconstructing corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data, Execution of the machine-executable instructions further causes the computing system to reconstruct synthetic k-space data from the synthetic magnetic resonance image data, the measured k-space data being divided into a plurality of k-space data groups, and the corrected magnetic resonance image data being reconstructed by modifying at least some of the k-space data groups using the synthetic k-space data.

2. The medical system of claim 1 , wherein the synthetic magnetic resonance image data provides prior knowledge during reconstruction of the corrected magnetic resonance image data.

3. Execution of the machine-executable instructions further causes the computing system to: determining a rigid body transformation of one or more of the groups of k-space data using the composite k-space data; The medical system of claim 1 , further comprising: performing phase and amplitude correction of one or more of the groups of k-space data using the rigid body transformation.

4. Execution of the machine-executable instructions further causes the computing system to: detecting at least one incomplete k-space sampling region in the measured k-space data; and filling the incomplete k-space sampling regions in the measured k-space data with the synthesized k-space data.

5. The memory further includes a picture quality assessment module that outputs a picture quality metric, and execution of the machine-executable instructions further causes the computing system to: generating a plurality of k-space data sets by systematically replacing combinations of the k-space data groups with portions of the composite k-space data; generating a plurality of trial magnetic resonance image data sets by reconstructing each set of the plurality of k-space data sets; The medical system of claim 1 , further comprising: selecting the corrected magnetic resonance image data from the plurality of trial magnetic resonance image data by optimizing the image quality metric output by the image quality assessment module.

6. The reconstruction of the corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data is formulated as an optimization problem that assigns weighting factors to each of the k-space data groups, and execution of the machine-executable instructions further causes the computing system to: identifying at least one corrupted k-space data group selected from the k-space data groups; correcting the at least one corrupted group of k-space data using the composite k-space data; and assigning the weighting factor to each of the k-space data groups, wherein the at least one corrupted k-space data group is assigned a weighting factor of a reduced value.

7. 2. The medical system of claim 1, wherein the corrected magnetic resonance image data is reconstructed according to a compressed sensing image reconstruction algorithm, the compressed sensing image reconstruction algorithm being an iterative algorithm that repeatedly generates intermediate magnetic resonance images, and the compressed sensing image reconstruction algorithm includes using the composite magnetic resonance image data to perform noise removal on the intermediate magnetic resonance images.

8. The medical system further comprises at least one magnetic resonance imaging system, the memory further comprising first pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire the measurement k-space data, and the memory further comprising second pulse sequence commands configured to control the at least one magnetic resonance imaging system to acquire reference k-space data, and execution of the machine-executable instructions further causes the computing system to: acquiring the reference k-space data by controlling the magnetic resonance imaging system using the second pulse sequence command; reconstructing the reference magnetic resonance image data from the reference k-space data; 2. The medical system of claim 1, further comprising: a step of: acquiring the measured k-space data by controlling the magnetic resonance imaging system using the first pulse sequence command.

9. Execution of the machine-executable instructions further causes the computing system to: constructing synthetic k-space data using the synthetic magnetic resonance image data; and using the synthetic k-space data to control acquisition of the measured k-space data.

10. 10. The medical system of claim 9, wherein execution of the machine-executable instructions causes the computing system to control the acquisition of the measured k-space data by using the composite k-space data to select a k-space sampling pattern for the first pulse sequence command.

11. The first pulse sequence command is configured to control the magnetic resonance imaging system to acquire the measurement k-space data in a k-space data group, and execution of the machine-executable instructions further causes the computing system to: calculating a comparison metric between the composite k-space data and each group of k-space data; 10. The medical system of claim 9, further comprising: performing a predetermined action if the comparison metric is outside a predetermined range of values.

12. The medical system of claim 1 , wherein the corrected magnetic resonance image data is reconstructed according to a parallel imaging magnetic resonance image reconstruction algorithm.

13. 1. A computer program comprising machine-executable instructions stored on a non-transitory computer-readable medium for execution by a computing system, the computer program further comprising an image generation neural network that outputs synthetic magnetic resonance image data upon receiving reference magnetic resonance image data as input, the image generation neural network generating the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system, the computer program, upon execution of the machine-executable instructions, causing the computing system to: accessing measured k-space data acquired according to the first configuration of the magnetic resonance imaging system, the measured k-space data representing a region of interest of a subject; accessing the reference magnetic resonance image data representative of the region of interest of the subject; generating access to the synthetic magnetic resonance image data by inputting the reference magnetic resonance image data into the image generation neural network; and arranging to reconstruct corrected magnetic resonance image data from the measured k-space data and the composite magnetic resonance image data, 10. The computer program product of claim 9, wherein execution of the machine-executable instructions causes the computing system to further reconstruct synthetic k-space data from the synthetic magnetic resonance image data, the measured k-space data being divided into a plurality of k-space data groups, and the corrected magnetic resonance image data being reconstructed by modifying at least some of the k-space data groups using the synthetic k-space data.

14. 1. A magnetic resonance imaging system, comprising: a memory storing machine-executable instructions and an image generation neural network, the image generation neural network outputting synthetic magnetic resonance image data when receiving reference magnetic resonance image data as input, the image generation neural network generating the synthetic magnetic resonance image data as a simulation of magnetic resonance image data acquired according to a first configuration of the magnetic resonance imaging system when the reference magnetic resonance image data was acquired according to a second configuration of the magnetic resonance imaging system, the memory further including first pulse sequence commands configured to control the magnetic resonance imaging system to acquire measured k-space data, and the memory further including second pulse sequence commands configured to control the magnetic resonance imaging system to acquire reference k-space data; a computing system, wherein execution of the machine-executable instructions causes the computing system to: acquiring the reference k-space data by controlling the magnetic resonance imaging system using the second pulse sequence command; arranging for said reference magnetic resonance image data to be reconstructed from said reference k-space data; accessing the synthetic magnetic resonance image data by inputting the reference magnetic resonance image data into the image generation neural network; arranging for constructing synthetic k-space data using the synthetic magnetic resonance image data; and controlling acquisition of the measured k-space data using the first pulse sequence command and the resultant k-space data.

Images