Motion correction using low-resolution magnetic resonance images

Abstract

Described herein is a medical system 100, 300 comprising a memory 110 storing machine executable instructions 120 and an upsampling neural network 122. In response to receiving a preliminary magnetic resonance image 126 at a first resolution lower than a second resolution, the upsampling neural network is configured to output an upsampled magnetic resonance image 130 at a second resolution. By executing the machine executable instructions, a computing system 104 receives (200) preliminary k-space data 124, reconstructs (202) the preliminary magnetic resonance image from the preliminary k-space data, receives (204) clinical k-space data 204, receives (206) an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image to the upsampling neural network, and provides (208) a motion corrected magnetic resonance image 132 using the upsampled magnetic resonance image and the clinical k-space data.

Description

[Technical field]

[0001] The present invention relates to magnetic resonance imaging, and in particular to motion correction during magnetic resonance imaging. [Background technology]

[0002] Magnetic resonance imaging (MRI) scanners use a large static magnetic field to align atomic nuclear spins as part of the procedure to generate images of the patient's body. This large static magnetic field is called the B0 field, or main magnetic field. MRI can be used to measure various quantities or characteristics of a subject spatially. Spatial encoding in magnetic resonance imaging is performed using a combination of a radio frequency (RF) waveform (or RF pulse) used to control the transmit coil of the MRI scanner, and multiple spatially selective gradient pulse waveforms (gradient pulses). A difficulty in performing magnetic resonance imaging is that it can take a significant amount of time to acquire enough data to reconstruct an image. MRI techniques are generally sensitive to subject motion.

[0003] US Patent Application Publication No. 20130278263A1 discloses magnetic resonance (MR) calibration data acquired using multiple radio frequency receiving coils, and both a coil sensitivity map and a reference projection vector are generated based on the MR calibration data. During imaging, additional navigator projection vectors can be acquired or a portion of the imaging data can be used as the navigator projection vector. Partial parallel imaging (PPI) can be performed to augment navigation information. The navigator projection vector and the reference projection vector are sensitivity weighted using the coil sensitivity map to generate a navigator sensitivity weighted projection vector (navigator SWPV) and a reference sensitivity weighted projection vector (reference SWPV), respectively, which are compared to generate subject position information. Subject motion is compensated for predictively or retroactively using the generated subject position information. Motion compensation is performed predictively by adjusting the imaging volume of the PPI based on the subject position information. Summary of the Invention

[0004] The present invention provides a medical system, a computer program and a method as set forth in the independent claims. Embodiments are set forth in the dependent claims.

[0005] The embodiments provide a means of compensating for subject motion by using lower resolution preliminary magnetic resonance images (such as scout images) to enable the provision of motion-corrected magnetic resonance images. This is achieved by converting the preliminary magnetic resonance images into upsampled magnetic resonance images using an upsampling neural network (super-resolution neural network). The upsampled magnetic resonance images can then be used in various motion correction schemes to provide motion-corrected magnetic resonance images from clinical k-space data.

[0006] The invention in one aspect provides a medical system comprising a memory storing machine executable instructions and also storing an upsampling neural network, the upsampling neural network being configured to, in response to receiving a preliminary magnetic resonance image at a first resolution, output an upsampled magnetic resonance image at a second resolution, the second resolution being higher than the first resolution, the upsampling neural network thus providing an upsampled magnetic resonance image at a higher resolution than the preliminary magnetic resonance image input to the upsampling neural network.

[0007] An upsampling neural network is a neural network configured for image processing. In some instances, an upsampling neural network is known as an upsampler or a super-resolution neural network. Methods for training an upsampling neural network are known. In some cases, a generative adversarial network may be used. Upsampling by a neural network is done using a convolutional neural network configured for image processing. An upsampling neural network can be trained by taking images at a second resolution and then transforming these images at the second resolution to the first resolution to create a training set of data. The set of pairs of images at the first resolution and the second resolution can then be used, for example, for deep learning. A subpixel transformation neural network was used in a proof-of-concept (POC) study described below. The subpixel transformation neural network is suitable for implementing an upsampling neural network.

[0008] The medical system further comprises a computing system, the computing system executing the machine executable instructions to receive preliminary k-space data at a first resolution representative of a region of interest of the subject. The computing system executing the machine executable instructions further reconstructs a preliminary magnetic resonance image from the preliminary k-space data. The computing system executing the machine executable instructions further receives clinical k-space data at a second resolution representative of the region of interest of the subject. The computing system executing the machine executable instructions further receives an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image to the upsampling neural network. The computing system finally executing the machine executable instructions further provides a motion corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data.

[0009] The upsampled magnetic resonance images are used in various ways to provide motion-corrected magnetic resonance images. The upsampled magnetic resonance images can be used, for example, to transform back to k-space and locate regions of corrupted k-space in clinical k-space data. Once identified, these corrupted k-space regions can be used, for example, to control a medical system to reacquire the k-space data or to perform actions to correct the corrupted k-space data.

[0010] In another embodiment, the medical system further comprises a magnetic resonance imaging system. The memory further comprises preliminary pulse sequence commands and clinical pulse sequence commands. The preliminary pulse sequence commands are configured to control the magnetic resonance imaging system to acquire preliminary k-space data. The clinical pulse sequence commands are configured to control the magnetic resonance imaging system to acquire clinical k-space data. The computing system executes the machine-executable instructions to further control the magnetic resonance imaging system with the preliminary pulse sequence commands to acquire the preliminary k-space data.

[0011] The computing system executes the machine executable instructions to further control the magnetic resonance imaging system with the clinical pulse sequence commands to acquire clinical k-space data. This embodiment is advantageous because the medical system includes a means for providing motion corrected magnetic resonance images. The upsampled magnetic resonance images may be useful, for example, for both predictive and retrospective motion correction of the clinical k-space data.

[0012] In another embodiment, the preliminary k-space data is acquired using a first magnetic resonance imaging modality. As used herein, magnetic resonance imaging modality encompasses the concept of a particular type of protocol used to acquire a magnetic resonance image. For example, there are various weighting options, such as T1, T2, or other weighting factors that are coded when the magnetic resonance image is acquired. Various repetition times are also changed, which affect the contrast or other image characteristics. Magnetic resonance imaging modalities are often referred to in the art or generally as contrasts. Clinical k-space data is acquired using a second magnetic resonance imaging modality. Providing a motion-corrected magnetic resonance image includes providing a simulated magnetic resonance image using an upsampled magnetic resonance image. The simulated magnetic resonance image has a second resolution and a second magnetic resonance imaging modality. This embodiment is highly beneficial because it not only increases the resolution, but also provides an upsampled magnetic resonance image using a simulated magnetic resonance image, even when the contrast or configuration is completely different when acquiring the k-space data twice. This can be useful, for example, by helping to achieve motion correction of clinical magnetic resonance images even when using different imaging modalities, i.e. contrasts, a first type of preliminary scan.

[0013] In another embodiment, the first magnetic resonance imaging modality is the same as the second magnetic resonance imaging modality, in which case there is no need to switch between the various imaging modalities, i.e. contrasts.

[0014] In another embodiment, the simulated magnetic resonance image is provided by an upsampling neural network configured to output an upsampled magnetic resonance image as the simulated magnetic resonance image, where for example the scaling neural network performs both the upsampling and the imaging modality conversion, i.e. contrast conversion.

[0015] In another embodiment, the memory further comprises a second resolution modality transfer neural network configured to output a simulated magnetic resonance image in response to receiving the upsampled magnetic resonance image. The second resolution modality transfer neural network is a modality transfer neural network configured to transfer the modality, i.e., "contrast", of the magnetic resonance image having the second resolution. The computing system further receives the simulated magnetic resonance image in response to inputting the upsampled magnetic resonance image to the second resolution modality transfer neural network by executing the machine executable instructions. Neural networks such as U-Net, F-Net, or other image processing neural networks are suitable for implementing the second resolution modality transfer neural network. The second resolution modality transfer neural network is trained by acquiring two images from the same subject using two different modalities. In this example, the preliminary magnetic resonance image is first upsampled, and then image transfer to the second modality is performed.

[0016] The memory, in another embodiment, further comprises a first resolution modality conversion neural network (similarly implemented and trained) configured to convert the preliminary magnetic resonance image from the first magnetic resonance imaging modality to the second magnetic resonance imaging modality. The computing system further executes the machine executable instructions to receive a converted preliminary magnetic resonance image in response to inputting the preliminary magnetic resonance image to the first resolution modality conversion neural network. The upsampling neural network is configured to output the upsampled magnetic resonance image as a simulated magnetic resonance image in response to receiving the converted preliminary magnetic resonance image as an input. In this example, first a contrast conversion, i.e. modality conversion, is performed and then the image is upsampled.

[0017] The first resolution modality-transferred neural network and the second resolution modality-transferred neural network have a similar structure and are trained in a similar manner.

[0018] In another embodiment, providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data includes performing a motion compensated reconstruction of the motion-corrected magnetic resonance image using the upsampled magnetic resonance image or the simulated magnetic resonance image.

[0019] In another embodiment, the motion compensated reconstruction is performed as an optimization using an upsampled or simulated magnetic resonance image as an intermediate estimate of the motion-free image to be reconstructed. This embodiment is advantageous because the use of an upsampled or simulated magnetic resonance image as the motion-free image not only improves the quality of the optimization, but also provides a means to reduce the computation time.

[0020] In another embodiment, the motion-compensated reconstruction includes determining the phase of the motion-free image to be reconstructed using a phase map determined at least in part from the clinical k-space data. This embodiment is beneficial because the phase of the k-space data derived from the upsampled magnetic resonance image is inaccurate. Therefore, improved motion correction can be performed by obtaining the amplitude from the upsampled image and the phase from the original clinical k-space data.

[0021] The computing system, in another embodiment, executes the machine-executable instructions to further calculate simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data or the upsampled magnetic resonance images. In the case of parallel imaging techniques, this involves multiplying the simulated imaging data by the coil sensitivity maps and Fourier transforming these images, which can then be directly compared to the k-space data acquired by each of the coil elements.

[0022] The computing system executes the machine executable instructions to further detect motion corrupted k-space data by comparing the simulated k-space data to clinical k-space data. The computing system executes the machine executable instructions to further optimize only the motion corrupted k-space data. In this embodiment, the regions of the clinical k-space data can be estimated using the simulated k-space data. The speed and quality of the optimization is significantly improved by optimizing only the regions detected as having corrupted k-space data.

[0023] In another embodiment, providing a motion corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data includes calculating simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data. As mentioned above, in parallel imaging, this involves multiplying the simulated imaging data by the coil sensitivity maps and Fourier transforming these images, which can then be directly compared to the k-space data acquired by each of the coil elements.

[0024] This embodiment further includes detecting motion corrupted k-space data by comparing the simulated k-space data with the clinical k-space data, and then finally reacquiring the motion corrupted k-space data and / or adjusting the acquisition of the clinical k-space data to compensate for the subject's motion. In this embodiment, the simulated k-space data is used to detect motion corrupted k-space data and then reacquire this data. This allows for superior reconstruction of motion corrected magnetic resonance images when the k-space data is so corrupted that it is impossible or undesirable to reconstruct.

[0025] In another embodiment, providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data includes calculating simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data, determining motion parameters by comparing the simulated k-space data to the clinical k-space data, and reconstructing a motion-corrected magnetic resonance image from the clinical k-space data using a motion correction algorithm that receives the motion parameters as input, the motion correction algorithm, for example, shifting or translating at least a portion of the clinical k-space data using the motion parameters before reconstructing the motion-corrected magnetic resonance image.

[0026] The computing system, in another embodiment, further executes the machine executable instructions to reconstruct a preliminary image from the clinical k-space data. The computing system, in another embodiment, further executes the machine executable instructions to determine an image registration between the preliminary image and the simulated magnetic resonance imaging data. The computing system, in another embodiment, further executes the machine executable instructions to use this image registration when comparing the simulated k-space data with the clinical k-space data. This is used, for example, to compensate or correct for subject motion between the preliminary magnetic resonance imaging acquisition and the clinical magnetic resonance imaging acquisition. This is useful, for example, in detecting which parts of the clinical k-space data are motion corrupted. This can be done by aligning the simulated k-space data with the motion corrupted image, the preliminary image, which is likely to align the simulated image with the predominant motion condition of the corrupted image. (On the other hand, if the motion parts are considered to be known and an upsampled image is used to estimate the motion parameters, this step is not important and the estimated motion parameters will simply differ.)

[0027] The second resolution, in another embodiment, is at least 1.5 times greater than the first resolution.

[0028] The second resolution, in another embodiment, is at least two times greater than the first resolution.

[0029] The second resolution, in another embodiment, is at least three times greater than the first resolution.

[0030] The second resolution, in another embodiment, is at least four times greater than the first resolution.

[0031] In another embodiment, the preliminary k-space data is at least partially coil calibration k-space data obtained from a plurality of magnetic resonance imaging coil elements. The computing system executes the machine executable instructions to further reconstruct a coil image for each of the plurality of magnetic resonance imaging coil elements. The computing system executes the machine executable instructions to further construct the preliminary magnetic resonance image by combining at least the coil images for each of the plurality of magnetic resonance imaging coil elements. This embodiment is beneficial because it provides a means for improving parallel imaging (e.g., sense imaging) reconstruction of magnetic resonance images.

[0032] In another embodiment, the preliminary k-space data is acquired, at least in part, from a body coil, which is advantageous because it allows for motion correction of higher resolution clinical images using the lower resolution images typically acquired with a body coil.

[0033] The preliminary k-space data is in another embodiment k-space data from magnetic resonance fingerprinting. The preliminary magnetic resonance image is a quantitative magnetic resonance image. This embodiment is particularly advantageous since it is easy to convert the magnetic resonance fingerprinting data into a pseudo magnetic resonance image that can be used for motion correction.

[0034] In another aspect, the present invention provides a computer program having machine executable instructions for execution by a computing system and an upsampling neural network also for execution by the computing system. The computer program may be stored on a non-transitory storage medium and may be a computer program product. The upsampling neural network is configured to output an upsampled magnetic resonance image of a second resolution in response to receiving a preliminary magnetic resonance image of a first resolution. The second resolution is higher than the first resolution.

[0035] The computing system further executes the machine-executable instructions to receive preliminary k-space data at a first resolution representative of a region of interest of the subject. The computing system further executes the machine-executable instructions to reconstruct a preliminary magnetic resonance image from the preliminary k-space data. The computing system further executes the machine-executable instructions to receive clinical k-space data at a second resolution representative of a region of interest of the subject. The computing system further executes the machine-executable instructions to receive an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image to an upsampling neural network. The computing system further executes the machine-executable instructions to provide a motion corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data.

[0036] In another aspect, the present invention provides a medical imaging method, comprising receiving preliminary k-space data of a first resolution representative of a region of interest of a subject. The method further comprises reconstructing a preliminary magnetic resonance image from the preliminary k-space data. The method further comprises receiving clinical k-space data of a second resolution representative of a region of interest of the subject. The second resolution is higher than the first resolution. The method further comprises receiving an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image to an upsampling neural network. The upsampling neural network is configured to output an upsampled magnetic resonance image of the second resolution in response to receiving the preliminary magnetic resonance image of the first resolution. The method further comprises providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data.

[0037] It should be understood that one or more of the above-mentioned embodiments of the present invention may be combined as long as the combined embodiments are not mutually exclusive.

[0038] Aspects of the invention may be embodied as an apparatus, method, or computer program product, as will be appreciated by one of ordinary skill in the art. Aspects of the invention may therefore 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 generally referred to herein as a "circuit," "module," or "system." Aspects of the invention may also take the form of a computer program product embodied in one or more computer readable medium(s) having computer executable code embodied thereon.

[0039] Any combination of one or more computer readable media may be utilized. A computer readable medium may be a computer readable signal medium or a computer readable storage medium. As used herein, a "computer readable storage medium" encompasses any tangible storage medium that stores instructions executable by a processor or computing system of a computing device. A computer readable storage medium may also be referred to as a computer readable non-transitory storage medium. A computer readable storage medium may also be referred to as a tangible computer readable medium. A computer readable storage medium may also store data that can be accessed by a computing system of a computing device in some embodiments. 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 register files of a computing system. 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 recording media that can be accessed by a computing device over a network or communication link. Data may be retrieved, for example, via a modem, via the Internet, or via a local area network. The computer executable code embodied on the computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wired, fiber optic cable, RF, etc., or any suitable combination thereof.

[0040] A computer-readable signal medium comprises a propagated data signal, e.g., in baseband or as part of a carrier wave, containing computer-executable code embodied therein. 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 is any computer-readable medium that can communicate, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device.

[0041] "Computer memory" or "memory" is an example of a computer-readable storage medium. Computer memory is any memory directly accessible to a computing system. "Computer storage" or "storage" is a further example of a computer-readable storage medium. Computer storage is any non-volatile computer-readable storage medium. Computer storage may be computer memory in some embodiments, or vice versa.

[0042] 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 "computing system," should be interpreted as including multiple computing systems or processing cores, as the case may be. A computing system is, for example, a multi-core processor. A computing system also refers to a collection of computing systems, either within a single computer system or distributed among multiple computer systems. The term computing system should also be interpreted as referring to a collection or network of computer processing devices, each of which includes a processor or computing system, as the case may be. Machine-executable code or instructions may be executed by multiple computing systems or processors, which may be within the same computer processing device or may be distributed across multiple computer processing devices.

[0043] Machine executable instructions or computer executable code comprise instructions or programs that cause a processor or other computing system to perform an aspect of the present invention. Computer executable code for carrying out the operations of an aspect of the present invention is written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, Smarttalk, C++, and traditional procedural programming languages ​​such as the "C" programming language or similar programming languages, and compiled into machine executable instructions. The computer executable code is sometimes in the form of a high-level language or in a pre-compiled form, and is used in conjunction with an interpreter that generates the machine executable instructions on the fly. The machine executable instructions or computer executable code is in another example a form of programming for a programmable logic gate array.

[0044] The computer executable code may execute completely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the scenario of execution on a remote computer or server, the remote computer may be connected to the user's computer via any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet Service Provider).

[0045] Aspects of the present invention are described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It is to be understood that each block or part of the blocks of the flowcharts, illustrations, and / or block diagrams, where applicable, can be implemented by computer program instructions in the form of computer executable code. It is further to be understood that combinations of blocks of different flowcharts, illustrations, and / or block diagrams may be combined, if not mutually exclusive. These computer program instructions are supplied to a computing system, such as a general purpose computer, a special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions executed by the computing system of the computer or other programmable data processing apparatus create means for performing the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0046] Such machine-executable instructions or computer program instructions may also be stored on a computer-readable medium that can direct a computer, other programmable data processing apparatus, or other device to function in a particular manner, such that the instructions stored on the computer-readable medium create an article of manufacture having instructions that implement the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0047] The machine-executable instructions or computer program instructions may also be loaded into a computer, other programmable data processing apparatus, or other device and cause the computer, other programmable apparatus, or other device to execute a series of operational steps to produce a computer-implemented process, such that the instructions executing on the computer or other programmable apparatus generate a process for performing the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0048] A "user interface" as used herein is an interface that allows a user or operator to interact with a computer or computer system. A "user interface" is also referred to as a "human interface device." A user interface provides information or data to an operator and / or receives information or data from an operator. A user interface allows a computer to receive input from an operator and provides output from the computer to a user. A user interface, in other words, allows an operator to control or manipulate a computer, and an interface allows a computer to display the results 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. A keyboard, a mouse, a trackball, a touchpad, a pointing stick, a graphics tablet, a joystick, a gamepad, a webcam, a headset, pedals, wired gloves, a remote control, and receiving data via an accelerometer are all examples of user interface components that allow for the reception of information or data from an operator.

[0049] As used herein, a "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 devices. A hardware interface allows a computing system to send control signals or commands to external computing devices and / or devices. A hardware interface also allows a computing system to exchange data with external computing devices and / or devices. 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.

[0050] As used herein, a "display" or "display device" encompasses an output device or user interface adapted to display images or data. A display outputs visual, audio, and / or tactile data. Examples of displays include, but are not limited to, computer monitors, television screens, touch screens, tactile electronic displays, Braille screens, cathode ray tubes (CRTs), storage tubes, bi-stable displays, ePaper, vector displays, flat panel displays, vacuum 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.

[0051] K-space data is defined herein as the recorded measurements 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 tomographic medical image data.

[0052] A Magnetic Resonance Imaging (MRI) image or MR image is defined herein as a reconstructed, two- or three-dimensional visualization of the anatomical data contained within the k-space data, which visualization can be performed using a computer.

[0053] Preferred embodiments of the invention will now be described, by way of example only, with reference to the drawings in which: [Brief description of the drawings]

[0054] [Figure 1] FIG. 1 illustrates an example of a medical device. [Diagram 2] 2 is a flow chart showing a method of using the medical device of FIG. 1. [Diagram 3] FIG. 13 is a diagram illustrating another example of a medical device. [Figure 4] 4 is a flow chart showing a method of using the medical device of FIG. 3. [Diagram 5] FIG. 13 illustrates another example of the method. [Figure 6] FIG. 1 shows the results of a proof-of-concept study. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0055] In these figures, like numbered elements are either equivalent elements or perform the same function. If functionally equivalent, an element that has been previously discussed will not necessarily be discussed in a subsequent figure.

[0056] FIG. 1 shows an example of a medical system 100. The medical system 100 is illustrated as comprising a computer 102. The computer 102 is intended to represent one or more computer systems, either co-located or distributed. The computer 102 is illustrated as comprising a computing system in communication with an optional hardware interface 106, an optional user interface 108, and a memory 110. The computing system 104 is intended to represent one or more computing systems, either in one or more locations. The computing system 104 may be integrated into a control system of a magnetic resonance imaging system, on a radiology computer system, or may also be available as a web-based or cloud-based service. The hardware interface 106 allows the computing system 104 to communicate with and / or control other components of the medical system 100. The hardware interface 106 may, for example, allow the computing system 104 to control a magnetic resonance imaging system, if one is present. The user interface 108 provides a means for an operator or user to control the operation and functionality of the medical system 100 .

[0057] Memory 110 is intended to represent various types of memory available or accessible to computing system 104. Memory 110 is illustrated as having machine-executable instructions 120. Machine-executable instructions 120 are instructions that enable computing system 104 to perform various tasks, such as image processing, numerical computations, and control of other components. Memory 110 is further illustrated as including an upsampling neural network 122 configured to receive and upsample a preliminary magnetic resonance image 126 to an upsampled magnetic resonance image 130.

[0058] The memory 110 is further illustrated as having preliminary k-space data 124, which may be retrieved from a storage device, for example, via a network interface, or may be directly acquired by a magnetic resonance imaging system. The memory 110 is further illustrated as having a preliminary magnetic resonance image 126, reconstructed from the preliminary k-space data 124. The memory 110 is further illustrated as having clinical k-space data 128. The clinical k-space data 128 is representative of a region of interest of a subject and has a second resolution. The preliminary k-space data 124 is representative of the same region of interest and has a first resolution. The second resolution is higher than the first resolution.

[0059] The memory 110 is illustrated as having an upsampled magnetic resonance image 130 obtained from the upsampling neural network 122 by inputting the preliminary magnetic resonance image 126. The upsampled magnetic resonance image 130 has a second resolution. The memory is further illustrated as having a motion-corrected magnetic resonance image 132 reconstructed using the clinical k-space data 128 and the upsampled magnetic resonance image 130. The upsampled magnetic resonance image 130 is used in various ways to assist in motion correction when reconstructing the clinical k-space data 128 into the motion-corrected magnetic resonance image 132. The upsampled magnetic resonance image 130 is, for example, transformed into k-space, which is used to locate or determine corrupted portions of the clinical k-space data 128. The upsampled magnetic resonance image 130 is, in another example, used as a reference image for reconstructing the motion-corrected magnetic resonance image 132.

[0060] Fig. 2 shows a flow diagram illustrating a method of operating the medical system 100 of Fig. 1. First, in step 200, preliminary k-space data 124 is received. Next, in step 202, a preliminary magnetic resonance image 126 is reconstructed from the preliminary k-space data 124. Then, in step 204, clinical k-space data 128 is received. Next, in step 206, an upsampled magnetic resonance image 130 is received in response to inputting the preliminary magnetic resonance image to the upsampling neural network 122. Then, finally, in step 208, a motion-corrected magnetic resonance image 132 is generated using the upsampled magnetic resonance image 130 and the clinical k-space data 128.

[0061] 3 illustrates another example of a medical system 300. The medical system 300 illustrated in FIG. 3 is similar to the medical system 100 illustrated in FIG. 1, except that the medical system 300 further includes a magnetic resonance imaging system 302.

[0062] The magnetic resonance imaging system 302 comprises a magnet 304. The magnet 304 is a superconducting cylindrical magnet with a bore 306 passing through the magnet. It is also possible to use different types of magnets, for example both split cylindrical magnets and so-called open magnets. Split cylindrical magnets are similar to standard cylindrical magnets, except that the cryostat is split into two parts to allow access to the same surface of the magnet, such magnets are used, for example, in conjunction with charged particle beam therapy. Open magnets have two magnet parts, one above the other, with a space between them large enough to accommodate a subject. The arrangement of the areas of the two parts is similar to that of a Helmholtz coil. Open magnets are popular because the subject is less confined. Inside the cryostat of the cylindrical magnet, a superconducting coil is assembled.

[0063] Within the bore 306 of the cylindrical magnet 304 is an imaging zone 308 where the magnetic field is strong enough and sufficiently uniform to perform magnetic resonance imaging. A field of view 309 is illustrated within the imaging zone 308. Acquired magnetic resonance data is typically acquired with respect to the field of view 309. A region of interest may be identical to the field of view 309 or may be a partial volume of the field of view 309. A subject 318 is illustrated as being supported on a subject support 320 such that at least a portion of the subject 318 is within the imaging zone 308 and the field of view 309.

[0064] Also within the magnet bore 306 is a set of magnetic field gradient coils 310 used for preliminary magnetic resonance data acquisition to spatially encode magnetic spins within an imaging zone 308 of the magnet 304. The magnetic field gradient coils 310 are connected to a magnetic field gradient coil power supply 312. The magnetic field gradient coils 310 are intended to be representative. The magnetic field gradient coils 310 typically comprise 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 is controlled as a function of time and may be ramped or pulsed.

[0065] Adjacent to the imaging zone 308 is a radio frequency coil 314 for manipulating the orientation of the magnetic spins in the imaging zone 308 and for receiving radio transmissions by the spins, also in the imaging zone 308. The radio frequency antenna comprises multiple coil elements. The radio frequency antenna is also referred to as a channel or an 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. The transceiver 316 may represent separate transmitters and receivers as well. The radio frequency coil 314 may also comprise multiple receive / transmit elements, and the radio frequency transceiver 316 may comprise multiple receive / transmit channels.

[0066] The transceiver 316 and gradient controller 312 are shown as being connected to the hardware interface 106 of the computer system 102. Both of these components, as well as other components such as the subject support providing position data, provide sensor data 126.

[0067] The memory 110 is shown to include preliminary pulse sequence commands 330 and clinical pulse sequence commands 332 configured to control the magnetic resonance imaging system 302 to acquire the preliminary k-space data 124 and clinical k-space data 128, respectively. The memory is further shown to include a modality transfer neural network 334 used to transfer the modality, i.e., contrast, of the magnetic resonance images. The modality transfer neural network may be, for example, a first resolution modality transfer neural network or a second resolution modality transfer neural network.

[0068] The preliminary k-space data 124 and the clinical k-space data 128 are acquired using different contrasts. The preliminary k-space data 124 is, for example, a simple proton density image acquired at a relatively low resolution as a scout or pilot image. The clinical k-space data 128 may have a different contrast, such as a T1 or T2 weighted image. The modality switching neural network 334 is used together with the upsampling neural network 122. The two neural networks may be configured in various ways, for example. In one example, the upsampling neural network 122 is used first, and then the modality switching neural network 334 is used (in which case the modality switching neural network 344 is a modality switching neural network of a second resolution). In another configuration, the modality-switching neural network 334 is used first, followed by the upsampling neural network 122 (in this case the modality-switching neural network is a modality-switching neural network of the first resolution). In either case, the use of the two neural networks 334, 122 results in a simulated magnetic resonance image 336 having the same modality, i.e. contrast, and the same second resolution as the clinical k-space data 128. The simulated magnetic resonance image 336 or the upsampled magnetic resonance image 130 is then used to obtain a motion-corrected magnetic resonance image using the clinical k-space data. This is performed, for example, by performing a motion-compensated reconstruction of the motion-corrected magnetic resonance image 132. The motion-compensated reconstruction performs, for example, an optimization using the upsampled magnetic resonance image 130 or the simulated magnetic resonance image 336 as a motion-free image during this reconstruction.

[0069] Fig. 4 shows a flow diagram illustrating another method of operating the medical system 300 of Fig. 3. First, in step 400, preliminary k-space data is acquired by controlling the magnetic resonance imaging system 302 with a preliminary pulse sequence command 330. Then, in step 402, clinical k-space data 128 is acquired by controlling the magnetic resonance imaging system 302 with a clinical pulse sequence command 332. After step 402, steps 200, 202, 204, 206, and 208 shown in Fig. 1 are performed.

[0070] The example presents a method to convert a low-resolution pre-scan (preliminary magnetic resonance image 126) to a high-resolution estimate of a subsequent anatomical scan (T1w, T2w, etc.) using at least one dedicated neural network (upsampling neural network 122 and possibly modality conversion neural network 334). In a first step, a super-resolution network (upsampling neural network 122) upsamples the pre-scan data to a target resolution (second resolution). In a second step, a dedicated network converts the data to a target contrast.

[0071] If the patient moves during one of the anatomical scans, this translated pre-scan is used in one example as a motion-free reference to identify corrupted portions of k-space and estimate associated motion parameters. This information is used in the motion-compensated reconstruction below. Identification of corrupted k-space profiles can alternatively be performed in real-time during data acquisition of the anatomical scan, thereby guiding reacquisition of appropriate data and reducing motion artifacts.

[0072] Image quality degradation due to patient motion is one of the most frequent problems in clinical applications of MRI. Many patients have difficulty remaining calm throughout the entire scan. Retrospective correction of motion artifacts relies on accurate identification of corrupted parts of k-space and / or estimation of a set of parameters describing the patient motion. Previously proposed methods for this task either lack robustness or require excessively long reconstruction times. Improved methods for identifying corrupted shots and estimating motion parameters could significantly reduce the computational time of identification and estimation.

[0073] In examples, pre-scan data such as a Sense reference scan (SenseRefScan) is used to obtain an estimate of the next anatomical scan. This is accomplished in some examples using two dedicated networks for super-resolution (upsampling neural network) and contrast inversion (modality inversion neural network). The resulting inverted pre-scan data is then used to identify corrupted portions of k-space and / or estimate associated motion parameters as part of a motion compensated reconstruction.

[0074] An overview of an exemplary method is shown in FIG. 5. FIG. 5 shows a flow chart that diagrammatically represents an imaging method. In this example, a preliminary magnetic resonance image 126 or pre-scan is fed to an upsampling neural network 122, also called a super-resolution network. This results in an upsampled magnetic resonance image 130, also called an upsampled pre-scan. The pre-upsampled magnetic resonance image 130 is then fed to a modality transfer neural network 334 or a contrast transfer network. This results in a simulated magnetic resonance image 336, also called a transferred pre-scan image. The example of FIG. 5 shows just one possibility. If the images are of the same modality, the modality transfer neural network 334 can be omitted since there is no need to create a simulated magnetic resonance image 336. The contrast transfer network 334 is applied before the super-resolution network 122 in a further example. In this case, the transferred image 336 is then upsampled to the upsampled pre-scan 130. In either case, the transformed pre-scan 336 is fed into a motion correction algorithm along with the clinical k-space data 128 to calculate a motion corrected magnetic resonance image.

[0075] In a first step (122), the pre-scan data (preliminary magnetic resonance images 126) are transformed to a target resolution (second resolution) using a super-resolution network (upsampling neural network 122). For this task, various network architectures and training settings can be envisaged. In a proof-of-concept (POC) study, a subpixel convolutional neural network was trained on a dataset of high-resolution natural images downsampled by a factor of four.

[0076] In a second step, the upsampled pre-scan data (upsampled magnetic resonance images 130) are transformed into the desired MR contrast using a dedicated contrast transformation network (modality transformation neural network 334). This network can use various inter-image architectures such as U-Net, F-Net, etc. The creation of a suitable dataset can be achieved in several ways.

[0077] Within the clinical database, artifact-free scan pairs containing identical geometry are identified and, if necessary, registration of the two scans is used to create the database.

[0078] A quantitative data set containing tissue parameter maps is acquired to allow forward simulation of any MR contrast, i.e. hydrogen nuclear density, T1 and T2 maps. Additional tissue parameters such as diffusion, perfusion, etc. would be useful to extend this method to functional MR sequences.

[0079] If matching scan pairs containing identical geometry are not available, large datasets of (unpaired) scans are also used. In this case, the cycleGAN network architecture can be used. To avoid the need to train a dedicated contrast transformation network for every modification of scan parameter settings (e.g., changing TE and TR), the transformation network can be designed 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.

[0080] The converted pre-scan data (hereinafter referred to as x p The vectors 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 199, 199, 199, 199, 199, 199, 199, 199, 199, 200, 200

[0081] Identification of corrupted portions of k-space can be achieved by subtracting the Fourier transformed and transformed pre-scan from the acquired anatomical data. Large differences between these data sets are assumed to correspond to portions of k-space corrupted by motion and can be identified, for example, by thresholding.

[0082] In real-time motion artifact reduction techniques aiming at reacquiring the corrupted parts of k-space, the resulting information can be used as prior information in motion compensated reconstruction (see below) to simplify and speed up the computations and guide iterative reconstruction techniques including the rejection of motion corrupted data. Here, one can take advantage of the fact that all processing steps of the pre-scan data set can be performed before the anatomical scan starts, thus allowing very fast processing of the incoming anatomical data.

[0083] Retrospective correction of motion artifacts can be performed using motion compensated reconstruction and can be expressed as follows:

number

[0084] A complication of this method is the accurate prediction of the phase of the anatomical scan. Since the detection of the corrupted profile is performed in k-space, complex data is required. To address this problem, several approaches are possible.

[0085] For minor motion artifacts, the phase of the acquired motion corrupted data can simply be applied to the transformed pre-scan amplitude data, and the corrupted profile can still be detected.

[0086] Motion artifacts in the phase of the acquired motion corrupted data are corrected before combining the motion corrupted data with the pre-scan amplitude data. For minor artifacts, a simple polynomial fit of the phase may be sufficient. For more severe artifacts, a dedicated image-to-image network has been found to produce highly accurate artifact-free phase images.

[0087] x p Topological map of

number

number

number

[0088] To demonstrate the feasibility of the present invention, a POC experiment was performed using 2D high-resolution brain data obtained from a volunteer, and the results are shown in FIG.

[0089] FIG. 6 shows images used in a simulation to demonstrate the effectiveness of motion correction. Image 600 shows a high-resolution 2D brain scan. Image 602 is the same image as 600 downsampled by a factor of 4. Image 604 is an image of the same resolution as image 600, but constructed by upsampling image 602 with a super-resolution neural network. Image 606 shows a single-shot, motion-corrupted image with translation-induced errors. Image 608 shows the difference between the motion-corrupted data and the output of the super-resolution neural network in k-space, where the bright vertical lines correspond to the motion-corrupted shots. Image 610 shows the result of applying a Gaussian filter vertically to difference image 608. Image 612 shows a single-shot, motion-corrupted image with rotation-induced errors. Image 614 shows the difference between the motion-corrupted data and the output of the super-resolution neural network in k-space, where the bright vertical lines correspond to the motion-corrupted shots. Image 616 shows the result of applying a Gaussian filter vertically to difference image 614 .

[0090] The high resolution input data 600 was first downsampled by a factor of four (image 602). This low resolution "pre-scan" image x p was then upsampled using a dedicated super-resolution network trained on natural images to obtain an estimate of the input data (

number

[0091] Additional features may be considered in relation to the design and application of the described systems and methods.

[0092] Instead of using the SENSE reference scan, we use a dedicated fast pre-scan that produces the same contrast as the anatomical scan, albeit with lower resolution, thus eliminating the contrast conversion step and potentially reducing the associated errors, i.e. increasing the sensitivity of the method to very weak motion.

[0093] At the beginning of the examination, a pre-scan with a single low-resolution magnetic resonance fingerprinting (MRF) can be performed. The resulting quantitative tissue parameter maps (hydrogen nuclear density, T1, T2, ~) can then be used to obtain low-resolution estimates for all subsequent anatomical scans. Again, in this scenario, the contrast conversion step of Figure 5 can be omitted.

[0094] 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 representative and not restrictive. The invention is not limited to the disclosed embodiments.

[0095] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprises" does not exclude other elements or steps, nor does the singular element exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage. The computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, provided integrally with or as part of other hardware, or distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be interpreted as limiting the scope. [Explanation of symbols]

[0096] 100 Medical Systems 102 Computer 104 Computing Systems 106 Optional Hardware Interface 108 Optional User Interface 110 Memory 120 Machine Executable Instructions 122 Upsampling Neural Networks 124 Preliminary k-space data 126 Preliminary Magnetic Resonance Data 128 Clinical k-space data 130 Upsampled Magnetic Resonance Images 132 Motion-corrected magnetic resonance images 200 receiving preliminary k-space data at a first resolution representing a region of interest of a subject; 202 Preliminary magnetic resonance images are reconstructed from preliminary k-space data 204 receiving clinical k-space data at a second resolution representing a region of interest of the subject; 206 receiving an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image to the upsampling neural network. 208 Providing motion-corrected magnetic resonance images using upsampled magnetic resonance images and clinical k-space data 300 Medical Systems 302 Magnetic Resonance Imaging System 304 Magnet 306 Magnet Bore 308 Imaging Zone 309 Field of view 310 Magnetic field gradient coil 312 Power supply for magnetic field gradient coil 314 Radio Frequency Coil 316 Transmitter / Receiver 318 Target 320 Subject Support 330 Preliminary Pulse Sequence Command 332 Clinical Pulse Sequence Commands 334 Modality-switching Neural Network 336 Simulated Magnetic Resonance Imaging 400 A preliminary pulse sequence command is used to control the magnetic resonance imaging system to acquire preliminary k-space data. 402 Clinical pulse sequence commands are used to control the magnetic resonance imaging system to acquire clinical k-space data. 500 Image Reconstruction Algorithms 600 high resolution reference images 602 Downsampled Image 604 upsampled image 606 Simulated motion corrupted images (translation) 608 Motion Artifacts in k-Space 610 Motion Artifacts in Image Space 612 Image damaged by motion (rotation) 614 Motion Artifacts in k-Space 616 Motion Artifacts in Image Space

Claims

1. a memory storing machine-executable instructions and an upsampling neural network, the upsampling neural network responsive to receiving a preliminary magnetic resonance image at a first resolution to output an upsampled magnetic resonance image at a second resolution higher than the first resolution; Computing systems and 10. A medical system comprising: receiving preliminary k-space data at the first resolution representing a region of interest of a subject; reconstructing the preliminary magnetic resonance image from the preliminary k-space data; receiving clinical k-space data at the second resolution representing the region of interest of the subject; receiving the upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image into the upsampling neural network; providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data; Healthcare system.

2. The medical system further comprises a magnetic resonance imaging system, the memory further having preliminary pulse sequence commands and clinical pulse sequence commands, the preliminary pulse sequence commands controlling the magnetic resonance imaging system to acquire the preliminary k-space data, and the clinical pulse sequence commands controlling the magnetic resonance imaging system to acquire the clinical k-space data, and by executing the machine-executable instructions, the computing system further controlling the magnetic resonance imaging system using the preliminary pulse sequence commands to acquire the preliminary k-space data; acquiring the clinical k-space data by controlling the magnetic resonance imaging system using the clinical pulse sequence commands; The medical system of claim 1 .

3. 3. The medical system of claim 2, wherein the preliminary k-space data is acquired using a first magnetic resonance imaging modality, the clinical k-space data is acquired using a second magnetic resonance imaging modality, and providing the motion-corrected magnetic resonance image includes providing a simulated magnetic resonance image using the upsampled magnetic resonance image, the simulated magnetic resonance image having the second resolution and the second magnetic resonance imaging modality.

4. The simulated magnetic resonance image is the first magnetic resonance imaging modality is the same as the second magnetic resonance imaging modality; the upsampling neural network outputting the upsampled magnetic resonance image as the simulated magnetic resonance image; the memory further comprising a second resolution modality transfer neural network that outputs the simulated magnetic resonance image in response to receiving the upsampled magnetic resonance image, and execution of the machine-executable instructions further causes the computing system to receive the simulated magnetic resonance image in response to inputting the upsampled magnetic resonance image into the second resolution modality transfer neural network; the memory further comprising a first resolution modality transfer neural network that transfers the preliminary magnetic resonance image from the first magnetic resonance imaging modality to the second magnetic resonance imaging modality, and executing the machine-executable instructions further causes the computing system to: in response to inputting the preliminary magnetic resonance image to the first resolution modality transfer neural network, receive a transferred preliminary magnetic resonance image; and in response to the upsampling neural network receiving the transferred preliminary magnetic resonance image as an input, output the upsampled magnetic resonance image as the simulated magnetic resonance image. The medical system of claim 3 , provided by any one of

5. 5. The medical system of claim 1, wherein providing the motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data comprises performing motion-compensated reconstruction of the motion-corrected magnetic resonance image using the upsampled magnetic resonance image or the simulated magnetic resonance image.

6. The medical system of claim 5 , wherein the motion-compensated reconstruction using the upsampled magnetic resonance image or the simulated magnetic resonance image as a motion-free image to be reconstructed is performed as an optimization.

7. 7. The medical system of claim 6, wherein the motion compensated reconstruction includes determining a phase of the motion-free image to be reconstructed, at least in part, using a phase map determined from the clinical k-space data.

8. Execution of the machine-executable instructions further causes the computing system to: calculating simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data or the upsampled magnetic resonance image; detecting motion-corrupted k-space data by comparing the simulated k-space data with the clinical k-space data; restricting the optimization to the motion-corrupted k-space data; The medical system of claim 6.

9. using the upsampled magnetic resonance image and the clinical k-space data to provide the motion corrected magnetic resonance image; calculating simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data; detecting motion-corrupted k-space data by comparing the simulated k-space data with the acquired clinical k-space data; and Reacquiring the motion-corrupted k-space data and / or adjusting the acquisition of the clinical k-space data to adjust for motion of the subject. The medical system according to claim 3 or 4, comprising:

10. using the upsampled magnetic resonance image and the clinical k-space data to provide the motion corrected magnetic resonance image; calculating simulated k-space data by performing a Fourier transform of the simulated magnetic resonance imaging data; determining motion parameters by comparing the simulated k-space data with the clinical k-space data; and reconstructing the motion-corrected magnetic resonance image from the clinical k-space data using a motion correction algorithm that receives the motion parameters as input. The medical system according to claim 3 or 4, comprising:

11. the preliminary k-space data being coil calibration k-space data acquired, at least in part, from a plurality of magnetic resonance imaging coil elements, and executing the machine-executable instructions causes the computing system to further: reconstructing a coil image for each of the plurality of magnetic resonance imaging coil elements from the coil calibration k-space data; constructing the preliminary magnetic resonance image by combining at least the coil images for each of the plurality of magnetic resonance imaging coil elements; The medical system according to any one of claims 1 to 4.

12. The medical system of claim 1 , wherein the preliminary k-space data is acquired, at least in part, from a body coil.

13. The medical system of claim 1 , wherein the preliminary k-space data is k-space data obtained by magnetic resonance fingerprinting, and the preliminary magnetic resonance image is a quantitative magnetic resonance image.

14. 1. A computer program having machine-executable instructions for execution by a computing system and an upsampling neural network, the upsampling neural network responsive to receiving a preliminary magnetic resonance image at a first resolution to output an upsampled magnetic resonance image at a second resolution higher than the first resolution, the computer program causing the computing system to: receiving preliminary k-space data at a first resolution representative of a region of interest of a subject; reconstructing the preliminary magnetic resonance image from the preliminary k-space data; receiving clinical k-space data at a second resolution representing the region of interest of the subject; receiving the upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image into the upsampling neural network; providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data; Computer program.

15. receiving preliminary k-space data at a first resolution representing a region of interest of a subject; reconstructing a preliminary magnetic resonance image from the preliminary k-space data; receiving clinical k-space data at a second resolution greater than the first resolution, the second resolution representing the region of interest of the subject, the second resolution being greater than the first resolution; receiving an upsampled magnetic resonance image in response to inputting the preliminary magnetic resonance image into an upsampling neural network, the upsampling neural network outputting the upsampled magnetic resonance image at the second resolution in response to receiving the preliminary magnetic resonance image at the first resolution; providing a motion-corrected magnetic resonance image using the upsampled magnetic resonance image and the clinical k-space data; 1. A method of medical imaging comprising:

Images