Motion-corrected tracer motion mapping using MRI.

Abstract

Disclosed herein is a medical system 100, 300, 500 comprising a memory 110 having stored thereon machine-executable instructions 120 and a magnetic resonance reconstruction module 122. The magnetic resonance reconstruction module is configured to reconstruct a motion-corrected tracer motion map 126 from measured k-space data 124. The measured k-space data is undersampled. The measured k-space data is T1-weighted. The measured k-space data is dynamic contrast-enhanced k-space data. The medical system further comprises a processor 104 configured to control the medical system. Execution of the machine-executable instructions causes the processor to receive the measured k-space data (200) and reconstruct a motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module (202). The magnetic resonance reconstruction module 122 is configured to reconstruct the motion-corrected tracer motion map as a direct model-based reconstruction from the measured k-space data 124.

Description

[Technical Field]

[0001] The present invention relates to magnetic resonance imaging, and in particular to dynamic contrast-enhanced magnetic resonance imaging. [Background technology]

[0002] A large static magnetic field is used in magnetic resonance imaging (MRI) scanners to align the nuclear spins of atoms as part of the procedure to generate images of a patient's body. This large static magnetic field is called the B0 field or main magnetic field. Various quantities or properties of a subject can be measured spatially using MRI. In some MRI techniques, a contrast agent such as gadolinium, which affects the T1 relaxation time, can be injected into the subject. Measurements obtained over time can be used to determine the transport of the contrast agent through the subject. So-called tracer kinetic models can then be fitted to the transport in the subject to determine quantities such as perfusion.

[0003] A 2017 journal article by Guo et al., "Direct estimation of tracer-kinetic parameter maps from highly undersampled brain dynamic contrast-enhanced MRI," Magn. Reson. Med, 78:1566-1578 doi:10.1002 / mrm.26540, discloses a reconstruction method that utilizes the Patlak TK model to solve a nonlinear least-squares optimization problem, including the explicit use of a full forward model to transform parameter maps into (k,t) space. The proposed approach is compared to an indirect method that uses parallel imaging and compressed sensing to create intermediate images before TK modeling. Summary of the Invention

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

[0005] The method discussed in Guo et al. has been used for dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) of the brain. When imaging the brain, it is straightforward to constrain the subject's head. For other DCE MRI techniques, such as first-pass perfused cardiac CDE MRI and abdominal DCE MRI, the technique described in Guo et al. (2007) does not work. Embodiments can provide a means for generating a tracer motion (TK) map by providing motion correction. This can be done, for example, in several different ways. In some embodiments, the optimization problem is modified to include an additional term that corrects for subject motion. In other embodiments, the trained neural network can be trained to compensate for subject motion, such as respiratory and / or cardiac motion. In yet other embodiments, measured k-space data is first used to calculate motion-corrected k-space data. The motion-corrected k-space data is then used to calculate a motion-corrected tracer motion map. References herein to k-space data are understood to refer to k-space data sampled as a function of both position within k-space and time.

[0006] In one aspect, the present invention provides a medical system including a memory storing machine-executable instructions and a magnetic resonance reconstruction module configured to reconstruct a motion-corrected tracer motion map from measured k-space data, the measured k-space data being undersampled, the measured k-space data being T1-weighted, and the measured k-space data being dynamic contrast-weighted k-space data.

[0007] As used herein, a medical system encompasses either a workstation or a computer configured to process medical image data and / or a system for acquiring this medical image data. For example, in one example, the medical system may be a workstation. In another example, the medical system may be a combination of a magnetic resonance imaging system and a medical system.

[0008] In dynamic contrast-enhanced (DCE) magnetic resonance imaging, DCE signal data can be determined using repeated measurements over a period of time, resulting in the D1 contrast of the measured magnetic resonance image. This DCE signal data can be converted to gadolinium concentration. Once the gadolinium concentration is determined as a function of time, a tissue model can be fitted to model the transport of gadolinium through the subject. This tissue model is called a tracer motion map. Various tracer motion maps exist. A common one is the Tofts model. In this model, there is a fractional plasma volume for a portion of the voxel, and there is also a fractional extravascular space. The transport of gadolinium from the fractional plasma volume to the extravascular extracellular space can be used to measure perfusion. In cardiac DCE magnetic resonance imaging, this measurement of perfusion can be a measurement of the health or health status of the subject. As used herein, a motion-corrected tracer motion map includes parameters fitted to a model of contrast agent or gadolinium transport.

[0009] The medical system further includes a processor configured to control the medical system. Execution of the machine-executable instructions causes the processor to receive the measured k-space data. The measured k-space data may be retrieved from a memory storage device or acquired over a network, for example. In another example, the machine-executable instructions may control a medical imaging system or a magnetic resonance imaging system to acquire the measured k-space data. Execution of the machine-executable instructions further causes the processor to reconstruct the motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module. This embodiment may be advantageous because the k-space data is undersampled. Undersampled means that the k-space data does not meet the Nyquist criterion.

[0010] However, there may be redundancy in the repeated measurements forming the measured k-space data, which allows the k-space data to be undersampled. The measured k-space data can be used directly to calculate a motion-corrected tracer motion map from, for example, the magnetic resonance reconstruction module. This can increase the speed at which the measured k-space data is acquired and improve the quality or accuracy of the motion-corrected tracer motion map.

[0011] In another embodiment, the magnetic resonance reconstruction module is configured to reconstruct a motion-corrected tracer motion map as a direct model-based reconstruction from measured k-space data.

[0012] In another embodiment, the magnetic resonance reconstruction module is configured to solve the motion-corrected tracer motion map as an optimization problem, the optimization problem including a motion compensation regularization term. This embodiment may be advantageous because it allows skipping intermediate steps typically used to calculate a tracer motion map. The use of an optimization problem allows the k-space data to be undersampled to a greater extent, which may allow, for example, the measurement k-space data to be acquired more quickly. The use of a regularization term may allow for greater motion compensation.

[0013] In another embodiment, the optimization problem is a single optimization problem that directly solves the motion-corrected tracer motion map from the measured k-space data.

[0014] In another embodiment, the motion compensation regularization term is formulated from a deformation map of a motion compensated tracer motion map.

[0015] In another embodiment, the deformation map has time and space dependence.

[0016] In another embodiment, the motion compensation regularization term is formulated as an accumulated energy function that depends on the deformation map.

[0017] In another embodiment, the motion compensation regularization term is formulated as a hyperelastic material model that depends on the deformation map.

[0018] In another embodiment, the motion compensation regularizer is formulated as a curvature-based regularizer that depends on the deformation map.

[0019] In another embodiment, the motion compensation regularization term is formulated as a freeform deformation model using a cubic B-spline model that depends on the deformation map.

[0020] In another embodiment, the motion compensation regularizer is formulated as an affine transformation model that depends on the deformation map.

[0021] In another embodiment, the motion compensation regularization term is a non-rigid motion compensation regularization term.

[0022] In another embodiment, the motion compensation regularization term is a rigid body motion compensation regularization term.

[0023] The motion compensation regularization term can be, for example, a rigid, affine, or non-rigid motion compensation regularization term. This can be done in various ways. In the case of a rigid transformation, a phase shift can occur in the measured k-space data. In other examples, a stored energy function of a hyperelastic material, a curvature-based regularization term, or even a free-form deformation model using a cubic B-spline model can be used. This can enable non-rigid motion compensation of the motion-compensated tracer motion map.

[0024] In another embodiment, the optimization problem is formulated as minimizing the motion-compensated regularization term plus the norm of the difference between the measured k-space data and a k-space model configured to map the undersampled k-space data. This embodiment may be advantageous because it may provide a means of computing a motion-compensated tracer motion map from the measured k-space data without an intermediate step. This reuses redundancy of k-space data measurements in the measured k-space data, while also allowing the measured k-space data to be more highly undersampled.

[0025] In another embodiment, the norm is the L2 norm. The use of the L2 norm has been shown to work well in numerical methods. However, other norms can also be used numerically.

[0026] In another embodiment, the optimization problem comprises the medical system of claim 2 or 3, wherein the optimization problem comprises:

number

[0027] The term TK is used to generally represent the term tracer motion map. The term norm is used to represent a global or general norm, which is a mathematical norm. In some cases, the norm may be an L2 norm.

[0028] Instead of TK(r) above, various tracer kinetic models can be used, for example, the Patlak TK model can be substituted into the equation.

number

[0029] In the above example, the L2 norm is optionally substituted for the generalized norm. Furthermore, the hyperelasticity-based regularization term R hyper allows for large, smooth deformations while maintaining elastic behavior. Other models can also be used, such as curvature-based regularization, affine transformations, free-form deformation (FFD) parameterized using cubic B-spline models, as well as others. This problem can be solved, for example, using an alternating minimization strategy.

[0030] The above embodiments and explanations can be modified so that a single optimization problem without a regularization term is formulated. This provides an additional embodiment. In one such embodiment, the single optimization problem is:

number

[0031] The deformation map M(r,t) still exists and can correct for subject motion. The other terms in the above equation are as described above.

[0032] In another similar embodiment, the optimization problem is formulated as minimizing the norm of the difference between the measured k-space data and a k-space model configured to map the motion-corrected tracer motion map onto the undersampled k-space data, which also has the advantage of motion correction without a regularization term.

[0033] In another embodiment, the magnetic resonance reconstruction module is a neural network that is trained to output the motion-corrected tracer motion map in response to input of the k-space data. This embodiment is advantageous because it can be very efficient to measure k-space data and then directly receive the motion-corrected tracer motion map.

[0034] In other embodiments, the neural network can be trained using k-space data for a particular volume of interest. For example, when generating motion-corrected tracer motion maps for a cardiac region, the training data can be k-space data and motion-corrected tracer motion maps of the cardiac region.

[0035] In another embodiment, the neural network is trained using a motion-corrected tracer motion map combined with simulated motion-corrected k-space data. For example, the motion-corrected tracer motion map can be obtained using various means. For example, the motion-corrected tracer motion map can be determined using fully sampled measured k-space data. Motion correction can then be performed using imaging techniques, such as deformable or non-deformable mapping between a series of images. Regardless of how the motion-corrected tracer motion map is reconstructed, the equations can be reversed to create simulated motion-corrected k-space data. For example, k-space data for a fully sampled motion-corrected tracer motion map can be calculated. A subset of this data can be referred to as undersampled k-space data. Motion artifacts can then be artificially added to this k-space data.

[0036] In another embodiment, execution of the machine-executable instructions further causes the processor to calculate motion-corrected k-space data using the measured k-space data. A tracer motion map is calculated by inputting the motion-corrected k-space data into the magnetic resonance reconstruction module. In this embodiment, the measured k-space data is motion-corrected before being input. This can be useful, for example, to correct for certain types of motion. For example, rigid body motion of the subject can be compensated for by changing the phase of the measured k-space data in the k-space data. This can be achieved, for example, in various ways. The measured k-space data can be self-navigation k-space data. For example, k-space measurements can be focused on a central region, and this data alone can be sufficient to detect rigid body displacements of the subject. In another example, subsequent acquisition of measured k-space data can be used to generate individual images, which are then used to calculate translations used to correct the phase in k-space.

[0037] Other methods may also include the use of image navigators, such as two-dimensional navigators additionally acquired using the magnetic resonance imaging system, and external navigators, such as breathing bellows, cameras or breathing belts.

[0038] In another embodiment, the magnetic resonance reconstruction module is configured to solve the motion-corrected tracer motion map as an optimization problem or using a trained convolutional neural network. This embodiment may be similar to the one described above, but motion correction may be performed before the k-space data is input to the magnetic resonance reconstruction module.

[0039] In other embodiments, the tracer motion map is a mapping of any one of relative extracellular volume, intravascular plasma volume, plasma flow, permeability-surface area product, tissue extracellular extravascular space, influx mask transfer rate of a contrast agent such as gadolinium, myocardial blood flow, and combinations thereof.

[0040] In another embodiment, the measured k-space data is first-pass perfused cardiac k-space data. This embodiment can be beneficial because undersampling can help reduce the prominence of motion artifacts. Furthermore, when performing first-pass perfused cardiac magnetic resonance imaging, respiratory and cardiac motion can also be present. Such motion compensation can help improve the quality of motion-corrected tracer motion maps.

[0041] In another embodiment, the measured k-space data is dynamic contrast-enhanced magnetic resonance imaging data of the abdomen, which may be beneficial because the motion correction may help to compensate for abdominal motion caused by the subject's breathing.

[0042] In other embodiments, the measured k-space data is multi-coil k-space data. For example, there may be multiple receive coils used to receive the measured k-space data. This configuration may be particularly beneficial because when multi-coil k-space data is acquired, the data is typically undersampled and then reconstructed using something similar to the SENSE magnetic resonance imaging protocol. The magnetic resonance reconstruction module may be configured to use the k-space data and combine the data using coil sensitivities.

[0043] In another embodiment, the measured k-space data is undersampled by at least five times.

[0044] In another embodiment, the measured k-space data is undersampled by at least 10 times.

[0045] In other embodiments, the measured k-space data is undersampled by more than 20 times.

[0046] In another embodiment, the measured k-space data is undersampled by at least 30 times.

[0047] In another embodiment, the measured k-space data is undersampled by at least 40 times.

[0048] In another embodiment, the measured k-space data is undersampled by at least 50 times.

[0049] In another embodiment, the measured k-space data is undersampled by at least 60 times.

[0050] In other embodiments, the measured k-space data is undersampled by a factor of less than 70.

[0051] In another embodiment, the measured k-space data is undersampled by at least 80 times.

[0052] In another embodiment, the medical system further includes a magnetic resonance imaging system configured to acquire the measured k-space data from an imaging zone. The memory further includes pulse sequence commands configured to acquire the measured k-space data according to a first-pass perfusion cardiac magnetic resonance imaging protocol or an abdominal dynamic contrast-enhanced magnetic resonance imaging protocol. Execution of the machine-executable instructions further causes the processor to control the magnetic resonance imaging system with the pulse sequence commands to acquire the measured k-space data. This embodiment may be advantageous because the medical system may provide a motion-corrected tracer motion map with a reduced amount of motion artifact.

[0053] In another embodiment, the pulse sequence command is configured to acquire k-space data using a self-navigation k-space sampling pattern. For example, the self-navigation k-space sampling pattern can be a so-called stack-of-stars. In a stack-of-stars, a core k-space sampling pattern is rotated in k-space. A central region of k-space is sampled in each measurement, and the central k-space data can be used to perform self-navigation.

[0054] In another aspect, the present invention provides a method of operating a medical system, the method including receiving measured k-space data. The method further includes reconstructing a motion-corrected tracer motion map by inputting the measured k-space data into a magnetic resonance reconstruction module. The magnetic resonance reconstruction module is configured to reconstruct the motion-corrected tracer motion map from the measured k-space data. The measured k-space data is undersampled. The measured k-space data is T1-weighted. The measured k-space data is dynamic contrast-enhanced k-space data.

[0055] In another aspect, the present invention provides a computer program product having machine-executable instructions for execution by a processor controlling a medical system. The machine-executable instructions include a magnetic resonance reconstruction module configured to reconstruct a motion-corrected tracer motion map from measured k-space data. The measured k-space data is undersampled. The measured k-space data is T1-weighted. The measured k-space data is dynamic contrast-enhanced k-space data. Execution of the machine-executable instructions causes the processor to receive the measured k-space data. Execution of the machine-executable instructions further causes the processor to reconstruct a motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module.

[0056] It is understood that one or more of the above-described embodiments of the present invention can be combined, unless the combined embodiments are mutually exclusive.

[0057] As will be appreciated by one of skill 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 generally herein as a "circuit," "module," or "system." Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer-readable medium(s) having computer-executable code embodied therein.

[0058] 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, "computer-readable storage medium" encompasses any tangible storage medium capable of storing instructions executable by a processor 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. In some embodiments, a computer-readable storage medium may also store data accessible by a processor 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 processor register files. Examples of optical disks include compact discs (CDs) and digital versatile discs (DVDs), such as CD-ROM, CD-RW, CD-R, DVD-ROM, DVD-RW, or DVD-R discs. The term computer-readable storage medium also refers to various types of storage media that can be accessed by a computer device over a network or communications link. For example, data can be obtained via a modem, over the Internet, or over a local area network. Computer-executable code embodied on 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 of the foregoing.

[0059] A computer-readable signal medium may include, for example, a propagated data signal with computer-executable code embodied 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.

[0060] "Computer memory" or "memory" is one example of a computer-readable storage medium. Computer memory is any memory that is directly accessible by a processor. "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 be computer memory, or vice versa.

[0061] As used herein, a "processor" encompasses an electronic component that can execute a program or machine-executable instructions or computer-executable code. References to a computing device having a "processor" should be interpreted as including two or more processors or processing cores. A processor may be, for example, a multi-core processor. A processor may also refer to a collection of processors within a single computer system or distributed among multiple computer systems. The term computing device should also be interpreted as referring to a collection or network of computing devices, each containing one or more processors. Computer-executable code may be executed by multiple processors, which may be within the same computing device or distributed across multiple computing devices.

[0062] Computer-executable code may include machine-executable instructions or programs that cause a processor to perform an aspect 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 programming languages, including object-oriented programming languages ​​such as Java, Smalltalk®, or C++, and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages, and compiled into machine-executable instructions. In some cases, the computer-executable code is in the form of a high-level language or in a pre-compiled form and can be used in combination with an interpreter that generates the machine-executable instructions on the fly.

[0063] The computer executable code may run entirely on the user's computer, partly on the user's computer, as a standalone software package, partly on the user's computer and partly on a remote computer, or entirely on a remote computer or server. In the latter scenario, 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 the connection may be to an external computer (e.g., via the Internet using an Internet Service Provider).

[0064] Aspects of the present invention will be described with reference to flowcharts 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 portion of a block, of the flowcharts, diagrams, and / or block diagrams, where applicable, can be embodied by computer program instructions in the form of computer-executable code. Furthermore, it will be understood that combinations of blocks in different flowcharts, diagrams, and / or block diagrams can be combined, if not mutually exclusive. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to form a machine, such that the instructions, executed via the processor of the computer or other programmable data processing apparatus, form means for performing the functions / acts specified in the flowcharts and / or block diagrams.

[0065] These 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 containing instructions that implement the functions / acts specified in the flowcharts and / or block diagrams.

[0066] The computer program instructions may also be loaded into a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be executed on the computer, other programmable apparatus, or other device, such that the instructions executing on the computer or other programmable apparatus generate a computer-implemented process that provides processing for performing the functions / operations specified in the flowcharts and / or block diagrams.

[0067] 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" is sometimes referred to as a "human interface device." A user interface can provide information or data to an operator and / or receive information or data from an operator. A user interface can enable a computer to receive input from an operator and provide output from the computer to a user. In other words, a user interface can enable an operator to control or manipulate a computer, and the interface can enable a computer to show 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.

[0068] As used herein, a "hardware interface" encompasses an interface that allows a processor of a computer system to interact with and / or control external computing devices and / or devices. A hardware interface may allow a processor to send control signals or instructions to external computing devices and / or devices. A hardware interface may also allow a processor 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.

[0069] As used herein, a "display" or "display device" encompasses an output device or user interface adapted to display images or data. A display can output 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, electronic paper, 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.

[0070] 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 apparatus during a magnetic resonance imaging scan. K-space data is an example of medical image data. A magnetic resonance imaging (MRI) image or MR image is defined herein as a reconstructed two-dimensional or three-dimensional visualization of anatomical data contained within the magnetic resonance imaging data. This visualization can be performed using a computer.

[0071] The following preferred embodiments of the present invention are described, by way of example only, with reference to the following drawings, in which: [Brief explanation of the drawings]

[0072] [Figure 1] Figure 1 shows a healthcare system. [Figure 2] FIG. 2 illustrates a method of operating the medical system of FIG. [Figure 3] FIG. 3 shows another example of a medical system. [Figure 4] FIG. 4 illustrates a method of operating the medical system of FIG. [Figure 5] FIG. 5 shows another example of a medical system. [Figure 6] FIG. 6 illustrates a method of operating the medical system of FIG. [Figure 7] FIG. 7 shows several methods for calculating the tracer motion map. [Figure 8] Figure 8 compares several different algorithms for computing tracer kinetic maps. [Figure 9] Figure 9 further compares several different algorithms for computing tracer kinetic maps. [Figure 10] Figure 10 further compares several different algorithms for computing tracer kinetic maps. DETAILED DESCRIPTION OF THE INVENTION

[0073] Like numbered elements in the figures are equivalent elements or perform the same function. An element previously described will not necessarily be described in a later figure if the function is equivalent.

[0074] FIG. 1 illustrates one embodiment of a medical system 100. The medical system 100 is shown to include a computer 102. The computer 102 has a processor 104. The processor 104 is intended to represent one or more processing cores distributed across one or more computers. For example, the computer 102 may actually represent one or more computers connected via a network. The processor 104 is shown connected to a hardware interface 106. The hardware interface 106 may, for example, enable the processor 104 to communicate with and / or control other components of the medical system 100. The processor 104 is shown to be further connected to an optional user interface 108. The processor 104 is also connected to a memory 110.

[0075] Memory 110 may be any combination of memory accessible to processor 104. This memory may include memory such as main memory, cache memory, and non-volatile memory such as flash RAM, a hard drive, or other storage device. In some examples, memory 110 may be considered a non-transitory computer-readable medium.

[0076] The memory 110 is further shown as including machine-executable instructions 120, which enable the processor 104 to control other components of the medical system 100 as well as perform basic data analysis and image processing techniques. The memory 110 is further shown as including a magnetic resonance reconstruction module 122, which is also part of the machine-executable instructions 120. The magnetic resonance reconstruction module may be executable code that enables the processor 104 to acquire measured k-space data and reconstruct a tracer motion map. The memory 110 is further shown as including measured k-space data 124. The memory 110 is further shown as including a motion-corrected tracer motion map 126 reconstructed by inputting the measured k-space data 124 into the magnetic resonance reconstruction module 122.

[0077] Figure 2 shows a flow chart illustrating a method of operating the medical system 100 of Figure 1. First, measured k-space data 124 is received in step 200. Next, in step 202, a motion-corrected tracer motion map 126 is reconstructed by inputting the measured k-space data 124 into the magnetic resonance reconstruction module 122.

[0078] FIG. 3 illustrates another embodiment of a medical system 300. The medical system of FIG. 3 is similar to the medical system 100 illustrated in FIG. 1. Instead of directly inputting the measured k-space data 124 into the magnetic resonance reconstruction module 122, the medical system 300 of FIG. 3 is modified such that the processor 104 first corrects / calculates motion-corrected k-space data 302 from the measured k-space data 124 using machine-executable instructions 120. The k-space data 302 is then input into the magnetic resonance reconstruction module 122, which outputs a motion-corrected tracer motion map 126. For example, the measured k-space data 124 may include self-navigation k-space data, or there may be an external system signal used to measure the subject's motion phase. Either of these can be used to calculate the motion-corrected k-space data 302 from the measured k-space data 124. In particular, rigid-body transformations between acquired portions of the measured k-space data 124 can be corrected as phase changes in the measured k-space data 124 to calculate the motion-corrected k-space data 302.

[0079] Figure 4 shows a flow chart illustrating a method of operating the medical system 300 of Figure 3. Initially, step 200 is performed as shown in Figure 2. The method then proceeds to step 400, in which motion-corrected k-space data 302 is calculated using the measured k-space data 124. After step 400 is performed, the method proceeds to step 202 as shown in Figure 2.

[0080] Figure 5 illustrates another embodiment of a medical system 500. This medical system 500 is similar to the medical system 100 illustrated in Figure 1, except that the medical system 500 further includes a magnetic resonance imaging system 502. Features of the medical system 300 illustrated in Figure 3 can also be incorporated into the medical system 500 illustrated in Figure 5.

[0081] The magnetic resonance imaging system 502 includes a magnet 504. The magnet 504 is a superconducting cylindrical magnet with a bore 506 extending therethrough. Various types of magnets can be used, including both split cylindrical magnets and so-called open magnets. Split cylindrical magnets are similar to standard cylindrical magnets, except that the cryostat is divided 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 above the other, with a space large enough to accommodate a subject between them, similar to that of a Helmholtz coil. Open magnets are popular because they provide less restraint for the subject. Inside the cryostat of the cylindrical magnet is a collection of superconducting coils.

[0082] Within the bore 506 of the cylindrical magnet 504 is an imaging zone 508 where the magnetic field is strong and uniform enough to perform magnetic resonance imaging. A field of view 509 is shown within the imaging zone 508. The magnetic resonance data acquired is typically acquired relative to the field of view 509. A subject 518 is shown as being supported by a subject support 520 such that at least a portion of the subject 518 is located within the imaging zone 508 and field of view 509.

[0083] Also present within the magnet bore 506 is a set of gradient coils 510, which are used for preliminary magnetic resonance data acquisition to spatially encode magnetic spins within the imaging zone 508 of the magnet 504. The gradient coils 510 are connected to a gradient coil power supply 512. The gradient coils 510 are intended to be representative. Typically, the gradient coils 510 include three separate sets of coils for spatial encoding in three orthogonal spatial directions. The gradient coil power supply supplies current to the gradient coils 510. The current supplied to the gradient coils 510 can be controlled as a function of time, gradient, or pulsed.

[0084] Adjacent to the imaging zone 508 is a radio frequency coil 514 for manipulating the orientation of magnetic spins within the imaging zone 508 and for receiving radio frequency radiation from spins within the imaging zone 508. The radio frequency antenna may include multiple coil elements. The radio frequency antenna may also be referred to as a channel or antenna. The radio frequency coil 514 is connected to a radio frequency transceiver 516. The radio frequency coil 514 and the radio frequency transceiver 516 may be replaced by separate transmit and receive coils and separate transmitters and receivers. The radio frequency coil 514 and the radio frequency transceiver 516 are understood to be representative. The radio frequency coil 514 is also intended to represent a dedicated transmit antenna and a dedicated receive antenna. Similarly, the transceiver 516 may also represent separate transmitters and receivers. The radio frequency coil 514 may also have multiple receive / transmit elements, and the radio frequency transceiver 516 may have multiple receive / transmit channels. For example, if a parallel imaging technique such as SENSE is implemented, the radio frequency coil 514 will have multiple coil elements.

[0085] The transceiver 516 and gradient controller 512 are shown connected to the hardware interface 106 of the computer system 102. The memory 110 is further shown as containing pulse sequence commands 530. The pulse sequence commands 530 are commands or data that can be converted into commands that control the magnetic resonance imaging system 502 to acquire k-space data 124.

[0086] The memory 110 is further shown as including pulse sequence commands 530. The pulse sequence commands 530 may be used by the processor 104 to control the magnetic resonance imaging system 502 to acquire the measured k-space data 124.

[0087] Figure 6 shows a flowchart illustrating a method of operating the medical system 500 of Figure 5. First, in step 600, the processor 104 controls the medical system 502 using pulse sequence commands 530 to acquire measured k-space data 124. After step 600 is performed, the method proceeds to steps 200 and 202 as shown in Figure 2.

[0088] As a specific example, first-pass perfusion cardiac magnetic resonance imaging (FPP-CMR) enables the evaluation of coronary heart disease. However, conventional FPP-CMR suffers from drawbacks such as low spatial resolution, insufficient cardiac coverage, and requires long breath-holding procedures. Currently, perfusion abnormalities are typically identified visually by highly trained physicians. In recent years, quantitative analysis of FPP-CMR has emerged as a more reliable and operator-independent approach to identifying perfusion disorders. Typically, quantitative FPP-CMR first reconstructs individual dynamic images, which are then converted to contrast agent concentrations. Finally, quantitative myocardial perfusion maps are generated using tracer kinetic modeling. Here, we propose a model-based FPP-CMR reconstruction approach that combines image reconstruction and tracer kinetic modeling to more effectively exploit redundancies in FPP-CMR data. We demonstrate that such a synergistic approach enables a very high undersampling rate in each time frame, thus enabling much higher spatial resolution and wider coverage than conventional methods. Furthermore, the proposed method, in combination with respiratory motion correction and kt undersampling, can improve the quantification of myocardial perfusion while significantly increasing patient comfort.

[0089] Coronary artery disease (CAD) is a leading cause of death worldwide. This disease is usually caused by atherosclerosis (myocardial ischemia), which reduces blood flow to the heart. Positron emission tomography (PET) is the clinical standard for noninvasive myocardial perfusion quantification in ischemic patients. Nevertheless, first-pass perfusion cardiac magnetic resonance imaging (FPP-CMR) is rapidly evolving into an essential tool for detecting myocardial perfusion disorders. Compared with PET, this technique offers advantages such as higher spatial resolution, no radiation exposure, wider availability, and lower scan costs. However, FPP-CMR requires ultrafast acquisition (to capture the first pass of the contrast bolus), electrocardiogram (ECG) gating, and a breath-hold technique to reduce cardiac and respiratory motion, leading to a trade-off between spatial resolution (~2.5 mm) and cardiac coverage (~3 slices). Diagnostic accuracy is also impaired by false-positive defects due to respiratory-induced motion artifacts (patients are often unable to hold their breath) and dark rim artifacts. Furthermore, perfusion abnormalities are often identified visually, but this has prognostic value that depends on the operator's level of training and experience.

[0090] The lack of reproducible and accurate results is a major factor limiting the widespread clinical adoption of FPP-CMR. Quantitative FPP-CMR methods typically require first reconstructing individual dynamic contrast-enhanced images, which are then converted to contrast agent concentrations, and finally generating tracer kinetic (TK) parameter maps using TK modeling. These approaches can be referred to as "indirect" approaches. Direct model-based parametric reconstruction methods have been used in several applications in PET and dynamic contrast-enhanced MR imaging to directly obtain TK parameter maps from the acquired data.

[0091] This approach demonstrated superior quantitative performance compared to traditional indirect quantification methods. Furthermore, the direct model-based reconstruction approach reduces the dimensionality of the problem. That is, the image reconstruction problem is reduced to finding 2–4 TK parameter maps per pixel instead of 60 time points. Therefore, this approach provides accurate TK parameter maps while enabling very high acceleration factors by exploiting the redundancy of spatial information between time points. To date, compressed sensing (CS) and parallel imaging reconstruction approaches have been used to accelerate FPP-CMR acquisition by up to 8 times and achieve higher spatial resolution. In this study, the DIRect QuanTitative (DIREQT) FPP-CMR reconstruction framework is proposed to directly estimate quantitative myocardial perfusion maps from undersampled data. The proposed framework was evaluated on a numerical FPP-CMR phantom and a patient with suspected CAD.

[0092] The terms DIREQT and DIREQT-TV refer to two different configurations of the magnetic resonance reconstruction module 122 formulated as an optimization problem.

[0093] FIG. 7 shows two different methods for calculating a motion-corrected tracer motion map 126 from measured k-space data 124. In a conventional method, first, in step 700, the measured k-space data 124 is acquired. Next, multiple magnetic resonance signal and intensity images are reconstructed (702) from the measured k-space data 124. In the next step, contrast agent concentrations 704 are calculated. Finally, from these images 704, an indirect reconstruction 706 of the tracer motion map 126 is reconstructed. In this example, multiple images are reconstructed in step 702. Therefore, it would not be possible to acquire undersampled k-space data. An alternative is a direct model-based reconstruction 710 using the magnetic resonance reconstruction module 122. In this method, the motion-corrected tracer motion map 126 is calculated directly from the measured k-space data 124. This can be done in several different ways. In one method, a trained convolutional neural network can be used. In another approach, an optimization problem can be formulated and then solved to directly solve the motion-corrected tracer motion map 126 from the measured k-space data 124 .

[0094] As mentioned above, FIG. 7 shows the steps required in the traditional indirect method (700, 702, 704, 706) and the DIREQT forward model (710) that converts TK parameters into (multi-coil under-sampled) FPP-CMR measurements (k-space data 124).

[0095] The proposed DIREQT method estimates TK parameter maps directly from measured FPP-CMR data. This is achieved by inverting a forward model (shown by arrow 708 in Figure 7) that includes the process described below. The TK parameters map to contrast agent concentration. The Patlak model is used to estimate the contrast agent concentration over time, C(r,t):

number

[0096] Figure 7 shows a flowchart illustrating the indirect method and the proposed DIREQT reconstruction for obtaining TK parameters from multi-coil (undersampled) data d. The indirect reconstruction consists of three steps (blue arrows): First, an FPP-CMR signal intensity image s is estimated from the acquired (k,t) spatial data d. Next, the contrast agent concentration over time C is estimated from s. Finally, a TK parameter map is estimated from C. In DIREQT reconstruction, the TK parameters are estimated directly from the (k,t) spatial data d by solving an inverse problem using an iterative reconstruction method (long red arrow). The forward model used to convert the TK parameter map to the (k,t) spatial data d is shown by the small red arrow. Here, r∈(x,y) are spatial coordinates in the image domain, and C AIF is the arterial input function, and K Trans and v p are the TK parameters, which represent the contrast transfer coefficient and the fractional plasma volume, respectively. The parameter K Trans is involved in vascular permeability and blood flow.

[0097] Contrast concentration versus signal intensity. Contrast concentration C(r,t) modifies T1 according to the following formula:

number

number

[0098] The undersampled (k,t) spatial data d(k,t) is:

number

number

number

[0099] The spatial sparsity constraint on the TK parameter map is given in equation (6):

number

[0100] To solve this indirect problem, individual dynamic contrast-enhanced images can be reconstructed from the undersampled (k,t) spatial data by solving the following optimization problem:

number

[0101] The above direct formulation can be modified for motion compensation by modifying equations (4) and either equation (6) or equation (7). The undersampled (k,t) spatial data d(k,t) is related to s(r,t) as follows to replace equation (1):

number

[0102] The regularization of M can be based on the stored energy function of a hyperelastic material such as Saint-Venant-Kirchhoff or Ogden material, which leads to the following minimization problem to replace equation (6) above:

number

[0103] R, a hyperelasticity-based regularizer hyper allows for large, smooth deformations while maintaining elastic behavior. Other models can also be used, such as curvature-based regularization, affine transformations, and free-form deformations (FFDs) parameterized using cubic B-spline models. This problem can also be solved using an alternating minimization scheme.

[0104] Similarly, equation (7) becomes:

number

[0105] Equations (10) and (11) are regularization terms

number

[0106] The fully sampled FPP-CMR data were acquired using the MRXCAT numerical phantom and the following parameters: field of view (FOV): 320 × 320 × 80 mm 3 ;Spatial resolution: 2×2mm 2 slice thickness: 5 mm; TS / TR / TE: 150.0 / 2.0 / 1.0 ms; flip angle: 15°; contrast agent dose: 0.075 mmol / kg; contrast agent relaxation rate: 5.6 L / mmol·s; 6 receiver coils; 32 time frames; and population average C AIF The data were generated using the ; A radial kt sampling method was used to undersample the acquisition by 10, 20, 30, and 40 times. Gaussian noise was added to each dataset to obtain a contrast-to-noise ratio (CNR) of 40. For each undersampling ratio, six noise realizations were performed. DIREQT and indirect reconstructions were obtained from the undersampled datasets.

[0107] An in vivo experiment was also performed. The remaining FPP-CMR complete sampling acquisition was performed in one patient with suspected CAD using a dual-bolus technique with 0.0075 + 0.075 mmol / kg gadobutrol (Gadovist; Bayer, Germany) and a 3T scanner (Achieva; Philips Healthcare). A saturation-recovery turbo field echo (TFE) ECG triggering sequence was used with the following parameters: FOV: 320 × 320 mm. 2 Resolution: 2.8 x 2.8 mm 2 A single short-axis slice was acquired with free breathing using a slice thickness of 10 mm; TS / TR / TE: 120.0 / 1.96 / 0.93 ms; flip angle: 15°; acquisition window: 224.3 ms; total acquisition time: 1 min 20 s; and contrast relaxation rate: 5.0 L / mmol s. The same radial sampling method used in the simulation was used to generate 20x, 30x, and 40x undersampled datasets. CAIF was detected using a large region of interest drawn in the left ventricle, and pre-contrast T1(r,0) was extracted from the T1 mapping sequence. Furthermore, signal intensity was normalized to the pre-contrast signal.

[0108] To perform motion correction, free-breathing FPP-CMR acquisitions were first reconstructed using the vendor's default reconstruction. Using dynamic images, frame-by-frame translational motion was estimated by aligning each frame to a moving average (±7 frames) of the previous frames. Translational motion correction was then performed directly in k-space by applying a linear phase shift. Finally, these motion-corrected datasets were reconstructed using the indirect and DIREQT methods. The value of adding a spatial sparsity constraint to the TK parameter maps in the form of spatial total variation (TV) regularization was also examined (see Equation (7)). The regularization parameters were chosen empirically for all methods.

[0109] The TK parameter maps obtained with DIREQT and the indirect method were quantitatively evaluated against the reference (fully sampled) TK parameter maps using normalized mean square error (NMSE) and correlation coefficient (CC). Reconstructions were performed using MATLAB (MathWorks, USA) on an Intel i7-86508 @ 1.9 GHz laptop with 32 GB memory.

[0110] Figure 8 compares several different ways of calculating the motion-corrected tracer motion map 126. In this example, there are two different quantities that are calculated as part of the map 126: K Trans 800 and v p 802. Calculations are made for fully sampled measurements 804, 10x undersampling 806, 20x undersampling 808, 30x undersampling 810, and 40x undersampling 812. Each of these is performed for three different algorithms. This is done for the indirect method 830 shown by steps 700, 702, 704, and 706 in FIG. 7, and then using the DIREQT 832 and DIREQT-VV algorithms 834. From FIG. 8, it can be seen that both DIREQT-TV 834 and DIREQT 832 perform well with the value K even at high rates of undersampling. Trans 800 and v pIt can be seen that the DIREQT method performs an excellent job of computing 802. Figure 8 shows DIREQT reconstructions with and without TV regularization obtained from simulated undersampled data, along with fully sampled reference and indirect reconstructions. For the indirect method, the image quality of the TK map at 10x acceleration is comparable to the reference image. At higher acceleration rates, the quality of the TK parameter map rapidly degrades, and spurious perfusion defects become visible. In comparison, the overall image quality of the TK parameter maps obtained with DIREQT is superior to the indirect method at all levels of undersampling. However, at high acceleration rates, the DIREQT problem becomes ill-posed, leading to noise amplification. In such cases, regularization methods can be used to stabilize the solution. Figure 8 shows that TV regularization helps reduce noise amplification at high accelerations and also improves the convergence rate.

[0111] Figure 9 shows the reference image and K Trans 800 and v p The normalized mean squared error 900 and correlation coefficient 902 between the TK maps of 802 are shown for the indirect algorithm 830, the DIREQT algorithm 832 and the DIREQT-TV 834 algorithm. Figure 9 shows the quantitative results of the reconstruction of the TK parameters. The highest CC and lowest NMSE values ​​are achieved with the proposed DIREQT method, indicating a better match with the reference image.

[0112] Figure 10 shows the K values ​​of the numerical phantom obtained from undersampled data at 10x, 20x, 30x, and 40x using the indirect method, the proposed DIREQT, and DIREQT with TV regularization (DIREQT-TV). Trans and v p The reconstruction is shown. A reference image is shown for comparison. The proposed DIREQT produces high-quality TK maps even at very high undersampling rates.

[0113] FIG. 10 illustrates the difference between the Indirect (830) and DIREQT (832) algorithms using K values ​​obtained from fully sampled measurements 804, 20x undersampled measurements 808, 30x undersampled measurements 810, and 40x undersampled measurements 812 of patient data. Trans Finally, Figure 10 shows the TK parameter maps estimated from fully sampled and undersampled patient data using DIREQT. Note that the FPP-CMR data were acquired without breath-holding to improve patient comfort and minimize respiratory motion artifacts, which can significantly affect the quantification results. The proposed method also produced good results at high acceleration rates. The total reconstruction times for the indirect and DIREQT methods were ~290 and ~185 seconds, respectively.

[0114] The Patlak model was chosen in this study because it provides results comparable to other TK models commonly used in FPP-CMR, such as the Fermi and two-compartment models, and has the advantages of being linearizable and computationally simple. However, comparisons between different TK models, including the blood-tissue exchange (BTEX) model, will be the subject of future studies. Furthermore, other regularization methods can be employed to further increase the robustness of the proposed method against noise and further accelerate it. In future studies, the DIREQT method will be evaluated in a large cohort of patients suspected of CAD using prospective undersampled acquisitions. These studies will also aim to achieve much higher spatial resolution and coverage, and therefore higher diagnostic accuracy.

[0115] Various techniques can be used for kt sampling and to provide dynamic acquisition techniques. Dynamic (parallel) images of the heart have a high degree of spatiotemporal correlation and redundancy due to the static background and relatively small dynamic regions (or contrast changes).

[0116] To exploit spatiotemporal correlations and redundancies across a dynamic FPP-CMR series and thus achieve a high degree of incoherence, dynamic undersampling patterns, i.e., different k-space undersampling patterns at each time point t, can be used. These kt sampling trajectories acquire data in a way that minimizes signal overlap. Therefore, any aliasing artifacts generated add incoherently. This contrasts with the more standard approach of acquiring fully sampled partial Fourier or parallel imaging accelerated FPP-CMR datasets separately for each time frame. DIREQT also works with this standard approach, but can reach much higher accelerations using the kt sampling method.

[0117] This can be achieved, for example, using a Cartesian or non-Cartesian trajectory with a spiral or radial order and golden angle or small golden angle increments, or using a non-repeating Poisson-disk sampling trajectory, where the sampling can have a higher density in the center of k-space, although this is not required.

[0118] DIREQT also does not require training data or profiles, such as the k-tSENSE and k-tPCA approaches. However, DIREQT can be used in combination with these types of techniques. DIREQT can also be combined with other FPP-CMR acquisition techniques, including simultaneous multislice imaging.

[0119] In summary, the recovery of TK parameter maps is facilitated by the fact that the FPP-CMR dynamic frames are spatially and temporally correlated. The use of kt trajectories in DIREQT allows for a significant reduction in the amount of data required to obtain high-quality TK parameter maps, and further allows for improved spatial and temporal resolution.

[0120] Respiratory motion and cardiac contraction degrade FFP-CMR image quality. These rigid and non-rigid deformations limit the quantification accuracy of FFP-CMR. Therefore, to obtain accurate quantitative maps, first-pass data must be motion-corrected to minimize resulting artifacts. The proposed DIREQUT method, combined with motion correction techniques, can provide accurate quantitative maps from highly accelerated, free-breathing, and / or continuously acquired data.

[0121] Several approaches can be used to minimize respiratory and cardiac motion. For example, the most common methods use ECG triggering to freeze cardiac motion and breath-holding to reduce respiratory motion. As another example, self-navigation techniques can be used to extract motion information directly from the data or from assisted acquisitions. This motion information can be used retrospectively to correct acquired data. For example, translational motion correction can be performed directly in k-space by applying a linear phase shift. As another example, data binning can be performed to separate the data into different respiratory motion states and / or cardiac phases. Furthermore, affine or non-rigid motion can be iteratively estimated and corrected. Therefore, the framework can be formulated to jointly estimate motion and motion-corrected quantitative maps directly from FPP-CMR data. Furthermore, the problem can also be formulated so that the arterial input function can be estimated together with the TK parameter maps and motion. A population AIF can be used as an initial estimate.

[0122] Deep learning can also be used to reconstruct motion-corrected tracer motion maps. DIREQT reconstruction of 3D volumes can potentially require long computational times. A potential solution to reduce reconstruction time is to use deep learning methods to directly estimate TK parameter maps from undersampled FPP-CMR data (DIREQT-NET). One of the main advantages of deep learning-based reconstruction methods is their computational efficiency, which enables real-time applications.

[0123] One approach to solving equation (5) using deep learning is to directly learn a nonlinear mapping between the undersampled k-space data d or the aliased zero-filled undersampled reconstruction and the fully sampled TK parameter map using a deep neural network (e.g., a convolutional neural network, CNN). Thus, the training step consists of pairing the undersampled k-space (or image) with the desired ground truth TK parameter map. The reconstruction can then be trained in an end-to-end manner, where the TK parameter map is reconstructed from the undersampled data using the network and compared to the ground truth.

[0124] The trained CNN can then be used to generate artifact-free TK parameter maps from the undersampled FPP-CMR data. For very high acceleration rates (when regularization can be used to stabilize the DIREQT reconstruction), prior information or regularization can be implicitly learned from the undersampled data (or images). Therefore, they do not need to be specified during training. As another example, a deep residual learning method can be used, in which the network learns a residual parameter map (between the collapsed TK map and the ground truth TK map), which has a sparser and simpler representation than the parameter map. Thus, the network is trained to learn a mapping between the undersampled k-space data (or images) and the TK parameter map and output an estimate of the residual map. When k-space is the input data, the neural network can consist of a fully connected layer followed by a CNN. The main function of the fully connected layer is to learn a nonlinear mapping between k-space and the image domain. As another example, an unfolded recurrent network can be used, which ensures that the reconstruction matches the k-space data.

[0125] Several loss functions can be used to train deep neural networks. A common choice is the mean squared error between the TK parameter map estimates and the ground truth (or residuals). One can also include a forward physics model loss function (equation (5)) between the input data and the model-generated data.

[0126] If ground truth is available, the network can be trained in an end-to-end manner. However, it may be impossible to obtain fully sampled 2D high-resolution or 3D whole-heart FPP-CMR data. Therefore, if ground truth images are not available, unsupervised deep learning methods can be used to jointly solve the CNN weights and the reconstruction training set parameter maps.

[0127] The weights of a CNN may be referred to as weight coefficients of a CNN or other type of neural network.

[0128] Deep learning typically requires a large number of datasets for training, but such datasets are often unavailable in FPP-CMR. Nevertheless, it is possible to train the network using data augmentation techniques, which can be used to increase the number of datasets and prevent overfitting.

[0129] Deep neural networks other than CNNs, such as recurrent neural networks, (cycle) generative adversarial networks, Bayesian neural networks, and ADMM-Net, can be used for DIREQT-NET.

[0130] The use of a Bayesian neural network can be beneficial as it can additionally provide an uncertainty map that can help assess the accuracy of the motion-corrected tracer motion map.

[0131] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description is to be considered illustrative or exemplary and not restrictive. The invention is not limited to the disclosed embodiments.

[0132] 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 "comprise" does 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 certain 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 can be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, together with or supplied as part of other hardware, but also in other forms, such as via the Internet or other wired or wireless communication systems. Reference signs in the claims should not be construed as limiting the scope. [Explanation of symbols]

[0133] 100 Healthcare Systems 102 Computer 104 processors 106 Hardware Interface 108 User Interface 110 memory 120 machine executable instructions 122 Magnetic Resonance Reconstruction Module 124 measured k-space data 126 Motion-corrected tracer motion map 200 Receive measured k-space data 202 Reconstruct a motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module. 300 Healthcare Systems 302 Motion-corrected k-space data 400 Calculate tracer motion maps by inputting motion-corrected k-space data into the magnetic resonance reconstruction module 500 Healthcare Systems 502 Magnetic Resonance Imaging System 504 Magnet 506 Magnet Bore 508 Imaging Zone 509 Field of view 510 Gradient magnetic field coil 512 Gradient magnetic field coil power supply 514 Radio Frequency Coil 516 Transceiver 518 subjects 520 Subject Support 530 Pulse Sequence Commands Use 600 pulse sequence commands to control the magnetic resonance imaging system and acquire the measured k-space data. 700 multi-coil (k,t) spatial data 702 MR signal strength s 704 Contrast agent concentration 706 indirect reconfiguration 708 Arrow 710 Direct Model-Based Reconstruction 800 K Trans 802v p 804 Fully Sampling 806 10x undersampling 808 20x undersampling 810 30x undersampling 812 40x undersampling 830 Indirect 832 DIREQT 834 DIREQT-TV 900 Normalized Mean Squared Error 902 Correlation Coefficient

Claims

1. a memory storing machine-executable instructions and a magnetic resonance reconstruction module for reconstructing a motion-corrected tracer motion map from measured k-space data, the measured k-space data being undersampled, T1-weighted, dynamic contrast-enhanced k-space data; A processor that controls the medical system wherein execution of the machine-executable instructions causes the processor to: receiving the measured k-space data; reconstructing the motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module, wherein the magnetic resonance reconstruction module reconstructs the motion-corrected tracer motion map as a direct model-based reconstruction from the measured k-space data; In the healthcare system, the magnetic resonance reconstruction module solves the motion-compensated tracer motion map as an optimization problem, the optimization problem having a motion compensation regularization term; The medical system, wherein the optimization problem is a single optimization problem that solves the motion-corrected tracer motion map directly from the measured k-space data.

2. The medical system of claim 1 , wherein the motion compensation regularization term is formulated from a deformation map of the motion-compensated tracer motion map, the deformation map having time and space dependence.

3. The motion compensation regularization term may further include: a stored energy function that depends on the deformation map; a hyperelastic material model dependent on said deformation map; a curvature-based regularization term that depends on the deformation map; a free-form deformation model using a cubic B-spline model dependent on the deformation map; and An affine transformation model that depends on the deformation map The medical system of claim 2 , wherein the medical system is formulated as any one of:

4. The medical system of claim 1 , wherein the motion compensation regularization term is a non-rigid motion compensation regularization term.

5. The medical system of claim 1 , wherein the motion compensation regularization term is a rigid body motion compensation regularization term.

6. 6. The medical system of claim 1, wherein the optimization problem is formulated as minimizing the motion compensation regularization term plus the norm of the difference between the measured k-space data and a k-space model that maps the motion-corrected tracer motion map to the undersampled k-space data.

7. The optimization problem is [0016] 7. The medical system of claim 1, wherein r is spatial position, t is time, TK(r) is a tracer motion map term, R(M(r,t)) is the motion compensation regularization term, d(k,t) is the measured k-space data, M(r,t) is a deformation map, f(TK(r),M(r,t)) is a forward model of the k-space data for given TK(r) and M(r,t), and norm is a mathematical norm.

8. The medical system of claim 1 , wherein the optimization problem is formulated as minimizing the norm of the difference between measured k-space data and a k-space model that maps the motion-corrected tracer motion map to the undersampled k-space data.

9. The optimization problem is [Equation 17] where r is spatial position, t is time, TK(r) is a tracer motion map term, d(k,t) is the measured k-space data, M(r,t) is a deformation map, f(TK(r),M(r,t)) is a forward model of the k-space data for given TK(r) and M(r,t), and norm is a mathematical norm.

10. 2. The medical system of claim 1, wherein the magnetic resonance reconstruction module is a neural network, the neural network being trained to output the motion-corrected tracer motion map in response to inputting k-space data, and the neural network being trained using any one of the following: a motion-corrected tracer motion map paired with simulated motion-corrected k-space data, an unsupervised deep learning method for jointly solving weighting coefficients and a reconstructed training set parameter map, and a combination thereof.

11. 11. The medical system of claim 10, wherein the neural network is any one of a convolutional neural network, a deep neural network, a recurrent neural network, a paired generative adversarial neural network, a paired cycle generative adversarial neural network, a Bayesian neural network, and an ADMM-Net.

12. 12. The medical system of claim 1, wherein the magnetic resonance reconstruction module solves the motion-corrected tracer motion map as an optimization problem or is a trained convolutional neural network.

13. 13. The medical system of claim 1, wherein the motion-corrected tracer motion map is a mapping of any one of relative extracellular volume, intravascular plasma volume, plasma flow, permeability surface area product, tissue extracellular extravascular space, inflow mass transfer rate of contrast agent, myocardial blood flow, and combinations thereof.

14. The medical system of claim 1 , wherein the measured k-space data is first-pass perfused cardiac k-space data or abdominal dynamic contrast-enhanced k-space data.

15. 15. The medical system of claim 1, wherein the measured k-space data is undersampled by at least 5 times, the measured k-space data is undersampled by at least 10 times, the measured k-space data is undersampled by at least 20 times, the measured k-space data is undersampled by at least 30 times, the measured k-space data is undersampled by at least 40 times, the measured k-space data is undersampled by at least 50 times, the measured k-space data is undersampled by at least 60 times, the measured k-space data is undersampled by at least 70 times, or the measured k-space data is undersampled by at least 80 times.

16. 16. The medical system of claim 1, further comprising a magnetic resonance imaging system that acquires the measured k-space data from an imaging zone, the memory further comprising pulse sequence commands that cause the measured k-space data to be acquired according to a first-pass perfusion cardiac magnetic resonance imaging protocol or an abdominal dynamic contrast-enhanced magnetic resonance imaging protocol, and execution of the machine-executable instructions further causes the processor to control the magnetic resonance imaging system with the pulse sequence commands to acquire the measured k-space data.

17. The medical system of claim 16 , wherein the pulse sequence commands cause the measured k-space data to be acquired using a self-navigating k-space sampling pattern.

18. 1. A method of operating a medical system, comprising: receiving measured k-space data; and reconstructing a motion-corrected tracer motion map by inputting the measured k-space data into a magnetic resonance reconstruction module.

1. A method of operating a medical system, comprising: the magnetic resonance reconstruction module reconstructing a motion-corrected tracer motion map from measured k-space data, the measured k-space data being undersampled, the measured k-space data being T1-weighted, and the measured k-space data being dynamic contrast-enhanced k-space data; and the magnetic resonance reconstruction module reconstructing the motion-corrected tracer motion map as a direct model-based reconstruction from the measured k-space data, the magnetic resonance reconstruction module solves the motion-compensated tracer motion map as an optimization problem, the optimization problem having a motion compensation regularization term; A method of operating a medical system, wherein the optimization problem is a single optimization problem that solves the motion-corrected tracer motion map directly from the measured k-space data.

19. 1. A computer program comprising machine-executable instructions for execution by a processor controlling a medical system, the machine-executable instructions comprising a magnetic resonance reconstruction module for reconstructing a motion-corrected tracer motion map from measured k-space data, the magnetic resonance reconstruction module reconstructing the motion-corrected tracer motion map as a direct model-based reconstruction from the measured k-space data, the measured k-space data being undersampled, the measured k-space data being T1-weighted, and the measured k-space data being dynamic contrast-enhanced k-space data, wherein execution of the machine-executable instructions causes the processor to: receiving the measured k-space data; and reconstructing the motion-corrected tracer motion map by inputting the measured k-space data into the magnetic resonance reconstruction module; In a computer program, the magnetic resonance reconstruction module solves the motion-compensated tracer motion map as an optimization problem, the optimization problem having a motion compensation regularization term; The computer program, wherein the optimization problem is a single optimization problem that solves the motion-corrected tracer motion map directly from the measured k-space data.

Images