Determining readout direction and phase coding direction for parallel magnetic resonance imaging

Abstract

Disclosed herein is a method comprising: receiving (200) an initial pulse sequence command (122); receiving (202) calibration data (124); receiving (204) a volume of interest (126); receiving (206) a slice thickness (128) of the at least one slice stack; and receiving (208) a stack-related stack orientation (130) of the at least one slice stack. The method further includes performing, for a slice, the following: determining (210) a field of view (132) for the slice using the volume of interest, the slice thickness, and the stack orientation of the stack of slices; determining (212) a readout direction (136) for the slice and a phase encoding direction for the slice using the calibration data to optimize image metrics within the slice; and constructing (214) a slice-specific pulse sequence command (140) by modifying the initial pulse sequence command using the field of view, the readout direction, and the phase encoding direction. A system for performing the method is disclosed.

Description

Technical Field

[0001] This invention relates to magnetic resonance imaging, and more particularly to parallel magnetic resonance imaging protocols. Background Technology

[0002] As part of the process for generating images within a patient's body, a magnetic resonance imaging (MRI) system, or scanner, uses a large static magnetic field to align the nuclear spins of atoms. This large static magnetic field is called the B0 field or main magnetic field. By controlling the gradient magnetic field and radio frequency pulses, nuclear spins can be manipulated to generate radio frequency signals, which can then be sampled or measured as k-space data. The k-space data can then be reconstructed into an MRI image that visualizes the internal anatomy of the object. The so-called pulse sequence describes the time-dependent control of the radio frequency signal, the time-dependent magnetic field gradient, and the radio frequency pulses, as well as the sampling of the k-space data.

[0003] The time spent acquiring k-space data to reconstruct MRI images can be sufficient to cause discomfort or movement of the subject during the examination. Therefore, it is beneficial to expedite MRI examinations as quickly as possible. One method to accelerate acquisition is the use of parallel imaging techniques. In parallel imaging, multiple antenna elements or coils that acquire radio signals from spatially correlated regions are used to acquire k-space data. Before acquiring k-space data to obtain clinical MRI images, a calibration procedure is performed, in which a set of coil sensitivity maps is measured and used to determine how much the k-space data measured by a particular coil contributes to the overall MRI image during final reconstruction.

[0004] Mooiweer et al.'s paper, "Combining a Reduced Field of Excitation With SENSE-Based Parallel Imaging for Maximum Imaging Efficiency" ("Magnetic Resonance in Medicine", 78, pp. 88-96, 2017), discloses the generation of G-factor plots to predict the performance of SENSE and rSENSE accelerated scans. Summary of the Invention

[0005] The present invention provides medical systems, computer programs, and methods in the independent claims. Embodiments are given in the dependent claims.

[0006] In one aspect, the present invention provides a medical system including a memory storing machine-executable instructions. The medical system also includes a computing system. Execution of the machine-executable instructions causes the computing system to receive an initial pulse sequence command configured to control a magnetic resonance imaging system to acquire slice-specific k-space data according to a parallel imaging magnetic resonance imaging protocol. Execution of the machine-executable instructions also causes the computing system to receive calibration data. Execution of the machine-executable instructions further causes the computing system to receive a volume of interest. Execution of the machine-executable instructions further causes the computing system to receive the slice thickness of at least one slice stack located within the volume of interest. Execution of the machine-executable instructions further causes the computing system to receive the stack-related stacking orientation of the at least one slice stack.

[0007] Execution of the machine-executable instructions causes the computing system to perform the following operations for slices in the at least one slice stack. This includes determining a field of view for the slice using the volume of interest, the slice thickness, and the slice-related stacking orientation. This also includes using the calibration data to determine a readout orientation and a phase-encoding orientation for the slice to optimize image metrics within the slice. Furthermore, this includes constructing a slice-specific pulse sequence command by modifying the initial pulse sequence command using the field of view, the readout orientation, and the phase-encoding orientation.

[0008] In another aspect, the present invention provides a computer program comprising machine-executable instructions for execution by a computing system. Execution of the machine-executable instructions causes the computing system to receive an initial pulse sequence command configured to control a magnetic resonance imaging system to acquire k-space data according to a parallel imaging magnetic resonance imaging protocol. Execution of the machine-executable instructions also causes the computing system to receive calibration data. Execution of the machine-executable instructions further causes the computing system to receive a volume of interest. Execution of the machine-executable instructions further causes the computing system to receive the slice thickness of at least one slice stack located within the volume of interest. Execution of the machine-executable instructions further causes the computing system to receive the stack-related stacking orientation of the at least one slice stack.

[0009] Execution of the machine-executable instructions also causes the computing system to perform the following operations for slices in the at least one slice stack. This includes determining a field of view for the slice using the volume of interest, the slice thickness, and the stack-related orientation. This also includes using the calibration data to determine a readout orientation and a phase-encoding orientation for the slice to optimize image metrics within the slice. This further includes constructing a slice-specific pulse sequence command by modifying the initial pulse sequence command using the field of view, the readout orientation, and the phase-encoding orientation.

[0010] On the other hand, the present invention provides a magnetic resonance imaging method. The method includes receiving an initial pulse sequence command configured to control a magnetic resonance imaging system to acquire k-space data according to a parallel imaging magnetic resonance imaging protocol. The method also includes receiving calibration data. The method further includes receiving a volume of interest. The method also includes receiving the slice thickness of at least one slice stack located within the volume of interest. The method further includes receiving the stack-related stacking orientation of the at least one slice stack.

[0011] The method further includes performing the following operations for slices in the at least one slice stack: This includes determining a field of view for the slice using the volume of interest, the slice thickness, and the stacking orientation of the slice stack. This also includes using the calibration data to determine a readout orientation and a phase encoding orientation for the slice to optimize image metrics within the slice. Furthermore, this includes constructing a slice-specific pulse sequence command by modifying the initial pulse sequence command using the field of view, the readout orientation, and the phase encoding orientation. Attached Figure Description

[0012] In the following preferred embodiments, the invention will be described by way of example only and with reference to the accompanying drawings, in which: Figure 1 An example of a medical system is illustrated.

[0013] Figure 2 The illustration shows the use of Figure 1 A flowchart of the methods used in medical systems.

[0014] Figure 3 The illustration shows another example of a medical system.

[0015] Figure 4 The illustration shows the use of Figure 3 A flowchart of the methods used in medical systems.

[0016] Figure 5 The diagram illustrates the neural network architecture.

[0017] Figure 6 The diagram illustrates the convolutional input stage of a neural network.

[0018] Figure 7 Another example of the input states of a neural network is illustrated.

[0019] Figure 8 The illustration shows another example of the input state of a neural network modeled after the U-net neural network.

[0020] Figure 9 The illustration shows another example of the input state of a neural network modeled after the F-net neural network. Detailed Implementation

[0021] In these figures, elements with the same number are either equivalent elements or perform the same function. If the functions are equivalent, elements that have been discussed previously will not necessarily be discussed again in later figures.

[0022] In this example, the medical system may include a memory storing machine-executable instructions. The medical system also includes a computing system. Execution of the machine-executable instructions causes the computing system to receive an initial pulse sequence command configured to control the magnetic resonance imaging system to acquire k-space data according to a parallel imaging magnetic resonance imaging protocol. The initial pulse sequence command may be, for example, a template or example pulse sequence command typically used to configure the magnetic resonance imaging system for acquiring k-space data for a specific parallel imaging magnetic resonance imaging protocol. The initial pulse sequence command may, for example, be designed to be modified for a specific acquisition.

[0023] The execution of machine-executable instructions also enables the computing system to receive calibration data. Calibration data can be received, for example, by acquiring measurements on a specific object using a particular magnetic resonance imaging (MRI) system. In other examples, calibration data can be cached or stored in memory, such as MRI data and / or non-MRI data. In some examples, calibration data can also be averaged data.

[0024] Calibration data can take various forms, for example. In some examples, calibration data can be survey images. In other examples, calibration data can be a collection of images used to provide coil sensitivity maps for parallel imaging protocols. In still other examples, calibration data can be data previously acquired from the object or from different objects, or even data averaged over many objects (e.g., anatomical atlases).

[0025] The execution of machine-executable instructions also enables the computing system to receive the volume of interest. When the calibration data is an image or includes an image, the location of the volume of interest can be indicated by or by referring to the calibration data.

[0026] In the example, execution of machine-executable instructions causes the computing system to receive the slice thickness of at least one slice stack located within the volume of interest. Typically, when acquiring magnetic resonance images, the images can be acquired as a three-dimensional dataset or a stack of two-dimensional slices. In this case, k-space data is acquired for individual slices, such that the three-dimensional dataset is acquired as a stack of slices. When performing such a three-dimensional imaging procedure, it is possible to acquire more than one slice stack. Typically, the image has different thicknesses and in-plane resolutions for a given stack. By acquiring multiple stacks, isotropic three-dimensional images or datasets can be constructed. Execution of the machine-executable instructions also causes the computing system to receive the stack-related orientation of at least one slice stack. The stack-related orientation can be, for example, a vector perpendicular to the planes of the slices that make up the particular slice stack. If multiple slice stacks exist, there may be a vector for each received slice stack.

[0027] Execution of machine-executable instructions enables a computing system to perform the following steps for a slice or slices in at least one slice stack. For example, these steps might be performed for each slice in at least one slice stack. These steps include determining the field of view for a slice using the volume of interest, slice thickness, and stack-related stack orientation. The slice can be compared to the volume of interest. The intersection of a particular slice within the volume of interest will be the field of view. These steps also include using calibration data to determine the readout orientation and phase-encoding orientation for the slice to optimize image metrics within the slice. The properties of the coils used in a parallel imaging system may change depending on the readout orientation and phase-encoding orientation. Therefore, the readout orientation and phase-encoding orientation can be selected slice-by-slice to optimize image metrics.

[0028] Image metrics can be, for example, image resolution or signal-to-noise ratio (SNR). In some cases, image metrics can be different quantities, for example, proportional to or related to the SNR. These steps also include constructing slice-specific pulse sequence commands by modifying the initial pulse sequence command using the field of view, readout direction, and phase encoding direction. In this step, the initial pulse sequence command may then be adjusted for each individual slice using the field of view, readout direction, and phase encoding direction. This enables the acquisition of k-space data for these individual slices using a magnetic resonance imaging system. Thus, the example can provide an improved means of acquiring multi-slice magnetic resonance images using parallel imaging magnetic resonance imaging techniques. Although multiple slices are acquired, the readout direction and phase encoding direction are selected for each slice by optimizing the image metrics. This, for example, achieves all slices with a constant SNR, minimum resolution, or optimized resolution. In particular, this facilitates the construction of three-dimensional datasets using individual slices.

[0029] In the example, the pulse sequence command can use a given or fixed timing, such as echo train length (ETL), effective TE, etc. (Note: Strictly speaking, the term ETL is only applicable to single-shot TSE / TFE imaging to maintain constant image contrast or image weighting. The equivalent of the FFE sequence is the number of phase encoding steps, which defines the acquisition time for each slice).

[0030] The phase coding direction and scan acceleration factor are determined by optimizing the highest image resolution or other image metrics achievable for a given ETL.

[0031] For a given phase-encoded orientation, the field of view can be defined as the bounding box at the intersection of the imaging plane (slice) and the target. In this step, using a support for the target, this support can be determined from the sum of squares or from a volume coil image captured by a reference scan from SENSE (or other parallel imaging techniques). Therefore, another example of using calibration data is determining the desired FOV size for a given slice location. This example requires an image, but the image is independent of spatially varying coil sensitivity. Instead, the image is a single-channel image showing only the distribution of tissue and air.

[0032] Next, in some examples, the maximum acceleration factor r along this phase encoding direction can be determined. This can be accomplished using several methods, but it can be based on a free parameter that is considered an acceptable threshold, i.e., one that can be used to control the noise level of the reconstructed image and is most likely application-specific.

[0033] The achievable image resolution can now be calculated as... Choose a phase encoding direction that minimizes this value.

[0034] Grid search can be a practical alternative that solves the phase encoding direction optimization problem using inverse problem algorithms (e.g., by changing the phase encoding direction in 30° steps). Several methods for calculating the maximum speedup factor are described below.

[0035] In another example, the image metric is image resolution. This example can be useful because it provides a means of acquiring slices from a slice stack that have an image resolution above a predetermined threshold or that may have the highest image resolution.

[0036] In one example, the image metric is the signal-to-noise ratio (SNR) of the slices, and this SNR is determined to be above a predetermined minimum. This can be useful, for example, when constructing a 3D dataset using slices from at least one slice stack.

[0037] In another example, the so-called g-factor metric is optimized. This can be considered an indirect optimization for the signal-to-noise ratio.

[0038] In another embodiment, the image resolution is optimized using the maximum acceleration factor determined along the phase encoding direction.

[0039] In another example, the image metric is the signal-to-noise ratio (SNR). Execution of machine-executable instructions causes the computational system to optimize the SNR for a fixed image resolution. This can be beneficial because the image resolution can be fixed, thus ensuring that the images are consistent with each other and have an optimized SNR.

[0040] In MRI, there are always trade-offs between resolution, SNR, and acquisition time. Depending on the application, one aspect may be more favored than another.

[0041] For example, in an application with a rapidly moving organ, the length of a fixed snapshot would freeze the motion, which would limit the number of contours (ETL) that can be measured. Given a certain field of view (FOV), the achievable resolution can be optimized by setting the correct phase encoding and readout directions.

[0042] In another example, one might want to achieve images with the same SNR. Considering only the relative SNR would be sufficient, as the achievable signal and receiver noise would remain constant. For different slice orientations, the resolution can be optimized (via an acceleration factor) while keeping the number of contours constant. Of course, standard textbook teachings regarding voxel size, slice thickness, bandwidth, number of measurements, etc., are taken into account here.

[0043] In yet another example, the resolution is kept constant for each slice and orientation. SNR optimization then selects the phase encoding direction that minimizes the acceleration factor (including considering the g-factor). Speed ​​optimization keeps the maximum acceleration (allowed given the g-factor limit) constant, thus resulting in variable ETL acceleration for the total acquisition.

[0044] In another example, the optimization direction treats the g-factor graph as a proxy for the relative SNR, with several options, such as the average g-factor on the image, the maximum g-factor in any voxel, and ranking statistics (such as minimizing a given percentile).

[0045] In another example, the image metrics used for the trial phase coding direction or trial readout direction are repeatedly calculated to optimize the image metrics and determine the readout direction and phase coding direction for the slice. In this example, various search or grid techniques can be used to select the optimal or acceptable phase coding direction or readout direction. This can be a convenient or efficient way to determine the numerical values ​​of the phase coding direction or readout direction.

[0046] In another example, the calibration data is a set of coil sensitivity maps. For example, the calibration data could be a set of images, and each image from a multi-channel magnetic resonance imaging coil could be acquired by various antenna elements. Execution of the machine-executable instructions also causes the computing system to calculate the g-factor using the coil sensitivity maps for the experimental phase encoding direction or experimental readout direction. Execution of the machine-executable instructions also causes the computing system to determine the maximum acceleration factor for the experimental phase encoding direction or experimental readout direction using the g-factors for different phase encoding directions. Execution of the machine-executable instructions also causes the computing system to calculate image metrics using the maximum acceleration factor. This embodiment can provide a numerically efficient way to determine the phase encoding direction or readout direction for an individual slice.

[0047] In this example, one could maximize the acceleration factor so that the maximum value of the SENSE g-factor plot is below the acceptance threshold. In some cases, this can be computationally expensive. Therefore, for practical purposes, this computation can be simplified, for example, by considering only a fixed set of acceleration factors (1, 1.2, 1.4, ..., 3.8, 4) and calculating the g-factor only on a subset of voxels in the imaging plane.

[0048] In another embodiment, a subset of voxels within the slice is used to compute the g-factor. This can make numerical computation faster and more efficient.

[0049] In another example, the calibration data is a set of coil sensitivity maps. Execution of the machine-executable instructions also causes the computing system to calculate the maximum speedup factor by calculating the singular value decomposition of the coil sensitivity maps along the phase encoding direction. Execution of the machine-executable instructions also causes the computing system to calculate the maximum speedup factor by determining the effective ranking by recording the exponents of the singular values ​​reaching the total signal power in the singular value decomposition. Execution of the machine-executable instructions also causes the computing system to calculate the maximum speedup factor by setting the maximum speedup factor using the effective ranking. Execution of the machine-executable instructions also causes the computing system to use the maximum speedup factor to calculate image metrics.

[0050] In this example, based on the (low-resolution) coil sensitivity plot (CSM = synergistic effect / qbc), one can select a line along the acceleration direction and may consider a matrix for each line. Where Np and Nc are the number of pixels and the number of channels, respectively. For example, decompose matrix X: And by recording the singular value s of the total signal power reaching a certain amount. i The exponent determines the effective rank. The lower the exponent, the less acceleration the coil array has. Since these methods rely only on the data from the SENSE reference scan and do not use information about sequence timing, the calculation of the maximum speedup factor can be done independently and before the actual scan begins.

[0051] In another example, the calibration data is a set of coil sensitivity maps. The memory also stores a neural network configured to output a readout direction and a phase-encoding direction for the slice in response to receiving the set of coil sensitivity maps and the field of view of the slice as input. Execution of machine-executable instructions further causes the computing system to receive the readout direction and the phase-encoding direction for the slice in response to inputting the calibration data and the field of view of the slice into the neural network. This embodiment can be advantageous because it provides a very fast means of selecting the readout direction and the phase-encoding direction.

[0052] One architecture that can be used for neural networks is the visual transformer neural network architecture.

[0053] Another usable general neural network architecture is one that receives the coil sensitivity map. This architecture processes the coil sensitivity map using a series of convolutional stages that pass the input through flattening stages and to several fully connected layers. Additional information, such as details about the scanning geometry or metadata describing the object, can be directly passed to the fully connected layers.

[0054] The neural network can be trained by collecting coil sensitivity maps and determining the readout direction and phase encoding direction using one of the numerical methods described above. The results of the numerical calculations can be stored and then used as training data.

[0055] While in most of the examples detailed in this application the calibration data is a set of CSMs, in another example the calibration data is a lookup table that stores readout orientation and / or phase encoding orientation for a slice based on one or more properties of the slice's field of view. This example provides the motivation and possible implementation of an example of calibration data that is not an image. Execution of machine-executable instructions also enables the computational system to determine the readout orientation and / or phase encoding orientation for a slice by comparing one or more properties of the slice's field of view with the lookup table. The field of view can be referenced, for example, with respect to anatomical properties or landmarks of a particular individual with respect to a particular MRI coil. In various individuals, the orientation of the readout orientation or phase encoding orientation may not vary very much with the object. In this case, instead of performing complex numerical calculations or using neural networks to store the results of numerical calculations and then averaging these results over many different objects, a lookup table can be provided as an alternative. For example, this can have the advantage of being extremely fast and eliminating the need for numerical calculations.

[0056] The lookup table can provide a practical alternative optimized for patient-specific PE, and the acceleration factor is a calculated lookup table containing the average of a large number of values ​​optimized for patient specificity.

[0057] This lookup table method will not use reformulated CSM as input, but will instead use more abstract, patient-agnostic data (such as coils used, anatomical structures, contrast, slice orientation, etc.).

[0058] It is possible to determine the appropriate set of input parameters through cluster analysis of the output data. Again, instead of using lookup tables, machine learning techniques can be used to implement the mapping of these abstract categories to output values ​​(e.g., random forests).

[0059] In another example, the slice-specific pulse sequence command is configured to maintain constant MRI weighting. For example, various temporal sequences within the readout or pulse interval, as well as the number of pulses, can determine the type of weighting, such as T2, T1, or T2* weighting for a particular image. The benefit of doing this is that various slices will have the same contrast and can be more easily used to construct a 3D dataset.

[0060] The main changes to the initial sequence involve variations in the gradient orientation used for imaging various stacks / orientations. The readout intensity and duration, as well as the number, intensity, and increment of phase encoding steps, can be updated accordingly. For magnetization preparation sequences (inversion, T2 preparation, etc.), the time between the preparation pulse and the acquisition of the k-space center, as well as the contour order, determine the contrast. If the sequence timing is modified to optimize a certain metric, different ETLs, sliced ​​TRs, etc., may be used to scan different stacks / orientations. Preferably, the effective echo time (acquiring the k-space center) or the contour order (which part of the k-space to acquire after preparation) should not be modified. If the (sliced) TR is modified, this must preferably not affect the steady state (e.g., ensuring sufficient full relaxation).

[0061] In another example, the medical system also includes a magnetic resonance imaging (MRI) system. Execution of the machine-executable instructions further enables the computational system to repeatedly control the MRI system using slice-specific pulse sequence commands to acquire slice-specific k-space data. Slice-specific k-space data is k-space data used to reconstruct an image of a particular slice. Execution of the machine-executable instructions also enables the computational system to use the slice-specific k-space data to reconstruct a slice-specific two-dimensional image.

[0062] In another example, execution of machine-executable instructions also enables the computing system to assemble slice-specific two-dimensional images, possibly for each slice in a stack of at least one slice, into a three-dimensional magnetic resonance image. This can be beneficial because it can provide a more uniform three-dimensional magnetic resonance image.

[0063] In another example, the memory stores preliminary pulse sequence commands configured for acquiring preliminary k-space data. Execution of the machine-executable instructions also enables the computing system to use the preliminary pulse sequence commands to control the magnetic resonance imaging system to acquire preliminary k-space data. Execution of the machine-executable instructions also enables the computing system to reconstruct calibration data based on the preliminary k-space data. This can be advantageous, for example, as it provides a means of acquiring calibration space data. The acquired preliminary k-space data may be a portion of a calibration sequence used to acquire coil sensitivity maps for a parallel imaging magnetic resonance imaging protocol.

[0064] Figure 1An example of a medical system 100 is illustrated. The medical system 100 is shown to include a computer 102. Computer 102 is intended to represent one or more computers at one or more locations. Computer 102 is shown to include a computing system 104. Computing system 104 is intended to represent one or more computing systems or computing cores. Computing system 104 is shown to communicate with an optional hardware interface 106. If other components are present or integrated into the medical system 100, the hardware interface 106 may be used, for example, to communicate with and / or control said other components. Computing system 104 is also shown to communicate with an optional user interface 108, which allows a user or operator to control and operate the functions of the medical system 100.

[0065] The computing system 104 is also shown communicating with memory 110. Memory 110 is intended to represent various types of memory that can be accessed by the computing system 104. Memory 110 may, for example, be a non-transient storage medium. Memory may include volatile and non-volatile memory, storage device devices, and components. In some embodiments, memory may be based on or may rely on cloud-based data stored in a logical pool spanning different commodity storage servers located in an enterprise's own physical facility or a data center managed by a third-party cloud provider.

[0066] Memory 110 is shown to contain or store machine-executable instructions 120. The machine-executable instructions 120 enable the computing system 104 to perform various tasks, such as controlling other components, performing numerical calculations, and manipulating data and / or images. Memory 110 is also shown to contain an initial pulse sequence command 122 configured to control the magnetic resonance imaging system to acquire k-space data according to a parallel imaging magnetic resonance imaging protocol. The initial pulse sequence command 122 may be, for example, a template. Memory 110 is also shown to contain calibration data 124. In different examples, the calibration data 124 may take different forms. In one example, calibration 124 may be a coil sensitivity map used to calibrate the magnetic resonance imaging system during a parallel imaging magnetic resonance imaging protocol. Calibration data 124 may also be an investigation scan or an image reconstructed from an image used to calibrate a multi-element magnetic resonance imaging coil by providing a coil sensitivity map.

[0067] Memory 110 is also shown to include a volume of interest 126. When calibration data 124 is image data or a coil sensitivity map, the volume of interest 126 can be specified based on its location within the calibration data 124. Memory 110 is also shown to include slice thickness 128 and stack-related orientation 130 for slice stacks intersecting with the volume of interest 126. Memory 110 is also shown to include a field of view 132 for slices obtained by finding the intersection of a specific slice with the volume of interest 126 or by finding the intersection volume between a specific slice and the volume of interest 126. Memory 110 is also shown to include an optional numerical module 134. In some examples, the optional numerical module 134 may be, for example, a module for performing numerical calculations, or the optional numerical module 134 may be a neural network. If the numerical module performs numerical calculations, it can perform optimization of image metrics 139. The memory 110 is also shown to contain a readout direction 136 and a phase encoding direction 138 obtained for a slice having a field of view 132, a slice thickness 128, and a stack-related orientation 130. This is performed using calibration data 124. The memory 110 is also shown to contain a slice-specific pulse sequence command 140 obtained by fitting or modifying the initial pulse sequence command 122 using the field of view 132, the readout direction 136, and the phase encoding direction 138.

[0068] Figure 2 The illustrated operation is shown. Figure 1 A flowchart of a method for a medical system 100. In step 200, an initial pulse sequence command 122 is received. The initial pulse sequence command 122 is configured to control the magnetic resonance imaging system to acquire k-space data according to a parallel imaging magnetic resonance imaging protocol. In step 202, calibration data 124 is received. In step 204, a volume of interest 126 is received. In step 206, the slice thickness 128 of at least one slice stack located within the volume of interest is received. In step 208, a stack-related orientation 130 of at least one slice stack is received. The stack-related orientation 130 may be different for each slice in the at least one slice stack. Steps 210, 212, and 214 may be repeated for each slice in the at least one slice stack. In step 210, the field of view of slice 132 is determined using the volume of interest 126, the slice thickness 128, and the stack-related orientation 130. In step 212, the calibration data 124 is used to determine the readout direction 136 and the phase encoding direction 138 for the slice to optimize image metrics 139 within the slice. In some embodiments, step 210 is fed into step 212. In step 214, a slice-specific pulse sequence command 140 is constructed by modifying the initial pulse sequence command 122 using the field of view 132, the readout direction 136, and the phase encoding direction 138.

[0069] Figure 3 Another example of a medical system 300 is illustrated. Medical system 300 and... Figure 1 The medical system 300 is similar to the medical system 100 depicted in the figure, except that the medical system 300 additionally includes a magnetic resonance imaging system 302.

[0070] The magnetic resonance imaging system 302 includes a magnet 304. The magnet 304 is a superconducting cylindrical magnet with a hole 306 passing through it. Different types of magnets are also possible; for example, split cylindrical magnets and so-called open magnets can also be used. A split cylindrical magnet is similar to a standard cylindrical magnet, except that the cryostat has been divided into two sections to allow access to the equiplanar plane of the magnet; such a magnet can be used, for example, in conjunction with charged particle beam therapy. An open magnet has two magnet sections, one above the other, with a sufficiently large space between them to receive the object: the arrangement of the two sections is similar to that of Helmholtz coils. Open magnets are popular because the object is less restricted. An assembly of superconducting coils is located inside the cryostat of the cylindrical magnet.

[0071] An imaging region 308 exists within the aperture 306 of a cylindrical magnet 304, in which the magnetic field is sufficiently strong and uniform to perform magnetic resonance imaging. A volume of interest 309 is shown within the imaging region 308. Within the imaging region 308, a stack of slices 322 is located within the volume of interest 309. The stack of slices 322 is entirely perpendicular to a stack-associated stack orientation 324. k-space data is acquired for a field of view indicated by one of the slices 322. An object 318 is shown supported by an object support 320, such that at least a portion of the object 318 is within the imaging region 308 and the volume of interest 309.

[0072] Within the aperture 306 of the magnet, a set of magnetic field gradient coils 310 is also present. These coils are used to acquire preliminary k-space data for spatial encoding of the magnetic spins within the imaging region 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. Typically, the magnetic field gradient coils 310 comprise three separate sets of coils used for spatial encoding in three orthogonal spatial directions. The magnetic field gradient power supply supplies current to the magnetic field gradient coils. The current supplied to the magnetic field gradient coils 310 is controlled as a function of time and can be either ramped or pulsed.

[0073] Adjacent to the imaging region 308 is an RF coil 314, which is used to manipulate the orientation of magnetic spins within the imaging region 308 and to receive radio transmissions from spins also located within the imaging region 308. The RF antenna may comprise multiple coil elements. The RF antenna may also be referred to as a channel or antenna. The RF coil 314 is connected to an RF transceiver 316. The RF coil 314 and the RF transceiver 316 may be replaced by separate transmit and receive coils, or by separate transmitters and receivers. It should be understood that the RF coil 314 and the RF transceiver 316 are representative. The RF coil 314 is also intended to represent a dedicated transmit antenna and a dedicated receive antenna. Similarly, the transceiver 316 may also represent a separate transmitter and receiver. The RF coil 314 may also have multiple receive / transmit elements, and the RF transceiver 316 may have multiple receive / transmit channels to perform parallel imaging.

[0074] Transceiver 316 and gradient controller 312 are shown as hardware interface 106 connected to computer system 102.

[0075] The memory 110 is also shown to contain a preliminary pulse sequence command 330. The memory 110 is also shown to contain preliminary k-space data 322 acquired by controlling the magnetic resonance imaging system 302 using the preliminary pulse sequence command 330.

[0076] The memory 110 is also shown to contain slice-specific k-space data acquired by controlling the magnetic resonance imaging system using slice-specific pulse sequence commands 140. The memory 110 is also shown to contain a slice-specific two-dimensional image 336 reconstructed from the slice-specific k-space data 334. The slice-specific two-dimensional image 336 can be reconstructed for each slice in the slice 322. The memory 110 is also shown to contain a three-dimensional magnetic resonance image 338, which is reconstructed by assembling the slice-specific two-dimensional image 336 for each possible slice in the stack of slices 322.

[0077] Figure 4 The illustrated operation is shown. Figure 3 A flowchart of a method for a medical system 300. In step 400, a preliminary pulse sequence command 330 is used to control the magnetic resonance imaging system 302 to acquire preliminary k-space data 332. In step 402, calibration data 124 is reconstructed or calculated based on the preliminary k-space data 332. Figure 2As shown, steps 200, 202, 204, 206, 208, 210, 212, and 214 are executed. Steps 404 and 406 can be executed for each slice in the stack of slices 322. In step 404, the magnetic resonance imaging system 302 is controlled using slice-specific pulse sequence command 140 to acquire slice-specific k-space data 334. In step 406, a slice-specific two-dimensional image 336 is reconstructed based on the slice-specific k-space data 334. In step 408, the slice-specific two-dimensional images 336 from the individual slices in the stack of slices 322 are assembled into a three-dimensional magnetic resonance image 338.

[0078] The following Figures 5 to 9 The illustration depicts an implementation of a neural network that can be used to determine the readout direction and / or phase encoding direction for slices. Training data for the artificial intelligence (AI) component (neural network) can be generated directly because the inputs and outputs of the AI ​​component are the same as those of slow iterative (numerical) optimization; that is, for any input data, the target output data can be computed by performing conventional optimization. Since this is a slow computation, a database containing input / target output data pairs should be pre-computed before training the AI ​​component begins.

[0079] Figure 5 The diagram illustrates the architecture of a neural network 500 that can be used to determine readout direction 136 and / or phase encoding direction 138. The input 502 to the neural network 500 is a coil sensitivity map (CSM) and various scalar values ​​504. The scalar values ​​504 can, for example, be used as input for the field of view of slice 132. The coil sensitivity map is processed by multiple convolutional layers 506, which are then passed through a flattening layer 508. The result of the flattening layer 508 and the scalar values ​​504 are passed to one or more fully connected layers 510. The fully connected layers 510 then provide an output 512, which can include the readout direction 136 and / or phase encoding direction 138 for a particular slice.

[0080] There is a series of components that can be used to implement the neural network 500.

[0081] Since the CSM portion of the input data is a multi-channel 2D image, it is beneficial to use convolutional boxes 506 to process this data. To reduce the number of variables, the convolutional boxes can be interleaved with the pooling layers.

[0082] The output of the convolutional stage is flattened 508 and combined with the scalar input (acceleration / number of PE steps) 504, and passed through one or more fully connected layers 510. The output of the fully connected layer 510 is either the two output parameters directly, or the density of the two parameters, which is reduced to the final output value by a weighted average. The latter case has the advantage of being able to define the training loss on the density rather than on the two scalar outputs, which gives stronger control over the gradient of the trainable parameters.

[0083] One practical problem is that the number of parameters in a fully connected layer is constant. This may require the reformulated CSM to also have a constant size. This can be achieved by zero-padding the CSM of the actual imaging slice to a standard number of pixels that cover all FOVs associated with a given anatomical structure.

[0084] Furthermore, the CSM can preferably have a constant size in the channel dimension. The most straightforward way to achieve this is to train a network for each available MR coil. This also makes sense from an application perspective, as different coils are often associated with different clinical problems and the anatomical structures being imaged. Training a separate network for each receiver coil should significantly make the learning task easier.

[0085] However, another way to achieve a constant number of channels is to convert the physical channel's CSM to the coil's eigenmode and then select a fixed number of the strongest eigenmodes.

[0086] Figure 6 The diagram illustrates the convolution stage 506. The convolution stage has an input 502, in which a coil sensitivity map is received and then passed through a convolution sequence 600, which is a combination of convolution and nonlinear operations. The outputs of these convolution sequences 600 are then passed to a flattening operation 508.

[0087] Figure 7 The diagram illustrates the target Figure 6 An alternative to the convolution input stage 506 shown. In Figure 7 In this example, there is an encoder input stage 506'. Again, it includes a convolutional sequence 600'; however, in this example, it is a sequence containing convolution, pooling, and non-linear operations. It can be, for example, equivalent to the descent half of the U-net neural network.

[0088] Figure 8An alternative approach for the neural network is illustrated. In this case, a reverse multi-scale network 800 is shown. Input 502 is again fed into a convolution sequence 600, which includes convolution operations and nonlinear operators. The results of input 502 and convolution sequence 600 are passed to a functional box sequence 802. Within box 802, there is a pooling layer 804 that receives input 502 or input from a previous box. A second pooling layer 806 receives the output of convolution sequence 600 or the result of another convolution sequence 808 from a previous box. The results of the pooling layers from boxes 804 and 806 are fed into the input for another convolution sequence 808, which includes convolution operators and nonlinear operators for each layer. Box 802 can be repeated multiple times. The output of the final convolution sequence 808 is then input into a flattening layer 508.

[0089] Here, multiple volumes with reduced spatial resolution are created based on the input volume. These volumes are processed by a convolutional network. The output of the final convolutional stage with higher spatial resolution is downsampled and combined with the input at subsequent spatial scales. This architecture is similar to a multi-scale neural network, except that the convolutional layers are joined in the opposite direction. In a multi-scale neural network, the final output has the same resolution as the input. Here, the final output has the same resolution as the lowest spatial scale.

[0090] Figure 9 Another example of the input is shown in the diagram. Figure 9 The example in is similar to Figure 8 The difference in the examples is that, Figure 9 The example follows the architecture of the reverse F-net 900. Pooling layer 804 is again fed into the input of convolutional sequence 808, but the second pooling layer 806 is fed into the output of convolutional sequence 808 or the last layer. Here, feature concatenation occurs at the output of the convolutional network. This is similar to F-net, except that, as before, the direction of feature concatenation is reversed.

[0091] Several variations of the above-described general architecture exist. For example, the convolutional box can include a batch normalization layer, or the convolutional box can operate in residual mode, i.e., there are skip connections around the convolutional box. Another variation is the choice of pooling operations (maximum, mean, ...), scaling factors, or nonlinear operations (ReLU, tanh, ...).

[0092] It should be understood that one or more embodiments of the foregoing examples or embodiments of the present invention may be combined, as long as the combined embodiments are not mutually exclusive.

[0093] Those skilled in the art will recognize that aspects of the present invention can be implemented as apparatus, method, or computer program product. Therefore, aspects of the present invention can take the form of a completely hardware embodiment, a completely software embodiment (including firmware, resident software, microcode, etc.), or a combination of software and hardware aspects, all collectively referred to herein as "circuit," "module," or "system." Furthermore, aspects of the present invention can take the form of a computer program product implemented on one or more computer-readable media having computer-executable code implemented thereon.

[0094] Any combination of one or more computer-readable media can be used. A computer-readable medium can 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 or computing system of a computing device. A computer-readable storage medium may be referred to as a computer-readable non-transient 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 can also store data accessible by a computing system of a computing device. The term "computer-readable storage medium" also refers to various types of recording media accessible by a computer device via a network or communication link. For example, data can be retrieved on a modem, on the Internet, or on a local area network. Any suitable medium can be used to transmit computer-executable code implemented on a computer-readable medium, including but not limited to: wireless, wired, fiber optic cable, RF, etc., or any suitable combination thereof.

[0095] Computer-readable signal media may include, for example, propagated data signals in baseband or as a portion of a carrier wave, in which computer-executable code is implemented. Such propagated signals 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 may be any computer-readable medium that is not a computer-readable storage medium and is capable of delivering, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0096] "Computer memory" or "memory" is an example of a computer-readable storage medium. Computer memory is any memory that can be directly accessed by a computing system. "Computer storage device" or "storage device" is another example of a computer-readable storage medium. A computer storage device is any non-volatile computer-readable storage medium. In some embodiments, a computer storage device may also be computer memory, or vice versa.

[0097] As used herein, the term "computing system" encompasses electronic components capable of running programs or machine-executable instructions or computer-executable code. References to computing systems, including examples of "computing systems," should be interpreted as potentially including more than one computing system or processing core. A computing system can, for example, be a multi-core processor. A computing system can also refer to a collection of computing systems, either within a single computer system or distributed across multiple computer systems. The term "computing system" should also be interpreted as potentially referring to a collection or network of multiple computing devices, each of which includes a processor or computing system. Machine-executable code or instructions can be run by multiple computing systems or processors that may be within the same computing device or even distributed across multiple computing devices.

[0098] Machine-executable instructions or computer-executable code may include instructions or programs that instruct a processor or other computing system to perform an aspect of the invention. Computer-executable code for performing operations toward the aspects of the invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​(e.g., Java, Smalltalk, C++, etc.) and conventional programming languages ​​(e.g., the "C" programming language or similar programming languages), and compiled into machine-executable instructions. In some instances, the computer-executable code may be in the form of a high-level language or in a pre-compiled form, and may be used in conjunction with an interpreter that generates the machine-executable instructions at runtime. In other instances, the machine-executable instructions or computer-executable code may be in the form of programming a programmable gate array.

[0099] Computer executable code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer can 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 it can be connected to an external computer (e.g., via the Internet provided by an Internet service provider).

[0100] Aspects of the invention have been described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block or portion of the flowchart, illustration, and / or block diagram can be implemented by computer program instructions in the form of computer-executable code, where appropriate. It should also be understood that blocks in different flowcharts, illustrations, and / or block diagrams can be combined without mutual exclusion. These computer program instructions can be provided to a computing system of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which run via the computing system of the computer or other programmable data processing apparatus, create units for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0101] These machine-executable instructions or computer program instructions may also be stored in a computer-readable medium that can instruct a computer, other programmable data processing apparatus or other device to function in a particular manner, such that the instructions stored in the computer-readable medium produce an article of manufacture, the article of manufacture including instructions that implement functions / actions specified in flowcharts and / or one or more block diagrams.

[0102] Machine-executable instructions or computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device, thereby creating a computer-implemented process, such that the instructions running on the computer or other programmable apparatus provide for performing the functions / actions specified in the flowchart and / or one or more block diagram boxes.

[0103] 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" can also be referred to as a "human-machine interface device." A user interface can provide information or data to and / or receive information or data from an operator. A user interface enables input from an operator to be received by the computer and can provide output from the computer to the user. In other words, a user interface allows an operator to control or manipulate a computer, and the interface allows the computer to indicate the effects of the operator's control or manipulation. Displaying data or information on a monitor 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, game controller, webcam, head-mounted device, foot pedal, wired gloves, remote control, and accelerometer are all examples of user interface components that enable the reception of information or data from an operator.

[0104] As used herein, "hardware interface" encompasses the interfaces that enable 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 enables a computing system to exchange data with external computing devices and / or devices. Examples of hardware interfaces include, but are not limited to: Universal Serial Bus (USB), IEEE 1394 port, parallel port, IEEE 1284 port, serial port, RS-232 port, IEEE-488 port, Bluetooth connectivity, wireless LAN connectivity, TCP / IP connectivity, Ethernet connectivity, control voltage interfaces, MIDI interfaces, analog input interfaces, and digital input interfaces.

[0105] As used herein, the term "display" or "display device" encompasses an output device or user interface suitable for displaying images or data. A display can output visual, auditory, and / or tactile data. Examples of displays include, but are not limited to, computer monitors, television screens, touchscreens, tactile electronic displays, etc.

[0106] k-space data are defined in this paper as measurements of radio frequency signals emitted via atomic spins, recorded by the antenna of a magnetic resonance imaging (MRI) device during a magnetic resonance imaging (MRI) scan. Magnetic resonance data is an example of tomographic medical image data.

[0107] Magnetic resonance imaging (MRI) images, or MR images, are defined in this paper as two-dimensional or three-dimensional visualizations reconstructed from anatomical data contained within magnetic resonance imaging data. Such visualizations can be performed using a computer.

[0108] Although the invention has been illustrated and described in detail in the accompanying drawings and the foregoing description, such illustrations and descriptions should be considered illustrative or exemplary, and not restrictive; the invention is not limited to the disclosed embodiments.

[0109] Those skilled in the art, through studying the accompanying drawings, disclosure, and claims, will be able to understand and implement other variations of the disclosed embodiments when practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude multiple. A single processor or other unit may implement the functions of several items recited in the claims. Although certain measures are recited in dissimilar dependent claims, this does not imply that combinations of these measures cannot be advantageously used. Computer programs may be stored / distributed on suitable media, such as optical storage media or solid-state media supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. No reference numerals in the claims should be construed as limiting the scope.

Claims

1. A medical system (100, 300), comprising: The computing system (104) is configured as follows: Receive (200) an initial pulse sequence command (122), the initial pulse sequence command being configured to control the magnetic resonance imaging system (302) to acquire slice-specific k-space data (334) in accordance with a parallel imaging magnetic resonance imaging protocol. Receive (202) calibration data (124); Receive (204) Volume of Interest (126); Receive (206) at least one slice (322) within the volume of interest, stacked slice thickness (128); as well as Receive (208) the stack-related stacking orientation (130) of the at least one slice stack. The computing system is further configured to perform the following operations on slices in the at least one slice stack: The field of view (132) for the slice is determined (210) by using the volume of interest, the slice thickness and the stacking orientation. The readout direction (136) and phase encoding direction (138) for the slice are determined (212) using the calibration data to optimize image metrics within the slice (139); and (214) Slice-specific pulse sequence commands (140) are constructed by modifying the initial pulse sequence command using the field of view, the readout direction and the phase encoding direction.

2. The medical system according to claim 1, wherein, The image metric is image resolution.

3. The medical system according to claim 1, wherein, The image metric is signal-to-noise ratio (SNR), wherein the execution of machine-executable instructions causes the computing system to optimize the SNR for a fixed image resolution.

4. The medical system according to any one of claims 1 to 3, wherein, The image metric is repeatedly calculated for either the experimental phase encoding direction or the experimental readout direction to optimize the image metric and determine the readout direction and the phase encoding direction for the slice.

5. The medical system according to claim 4, wherein, The calibration data is a set of coil sensitivity maps, wherein the calculation system is configured as follows: The g-factor for the test phase encoding direction or the test readout direction is calculated using the coil sensitivity map. The maximum acceleration factor for the experimental phase encoding direction or the experimental readout direction is determined by using the g-factor for different phase encoding directions; and The image metric is calculated using the maximum acceleration factor.

6. The medical system according to claim 5, wherein, The g-factor is calculated using a subset of voxels within the slice.

7. The medical system according to claim 4, wherein, The calibration data is a set of coil sensitivity maps, wherein the calculation system is configured as follows: Calculate the singular value decomposition (SVD) of the coil sensitivity map along the phase encoding direction; The effective rank is determined by re-encoding the exponent of the singular value that reaches the total signal power in the singular value decomposition; The maximum acceleration factor is set by using the effective rank; and The image metric is calculated using the maximum acceleration factor.

8. The medical system according to claim 1, wherein, The calibration data is a set of coil sensitivity maps, wherein the computing system is configured to receive the readout direction for the slice and the phase encoding direction for the slice in response to inputting the calibration data and the field of view of the slice into a neural network, wherein the neural network is configured to output the readout direction for the slice and the phase encoding direction for the slice in response to receiving the set of coil sensitivity maps and the field of view of the slice as input.

9. The medical system according to claim 1, wherein, The calibration data is a lookup table that stores the readout direction and / or the phase encoding direction for the slice as a function of one or more attributes of the field of view of the slice, wherein the computing system is configured to determine the readout direction and / or the phase encoding direction for the slice by comparing the one or more attributes of the field of view of the slice with the lookup table.

10. The medical system according to any one of the preceding claims, wherein, The slice-specific pulse sequence command is configured to maintain constant magnetic resonance imaging weighting.

11. The medical system according to any one of the preceding claims, wherein, The medical system also includes a magnetic resonance imaging system, wherein the computing system is configured as follows: The magnetic resonance imaging system is repeatedly controlled (404) using the slice-specific pulse sequence command to acquire the slice-specific k-space data; and (406) Slice-specific two-dimensional images are reconstructed using the slice-specific k-space data.

12. The medical system according to claim 11, wherein, The computing system is configured to assemble (408) slice-specific two-dimensional images of slices in the at least one slice stack into a three-dimensional magnetic resonance image.

13. The medical system according to claim 11 or 12, wherein, The computing system is configured as follows: The magnetic resonance imaging system is controlled (400) using a preliminary pulse sequence command to acquire preliminary k-space data; and The calibration data (402) is reconstructed based on the preliminary k-space data.

14. A magnetic resonance imaging method, comprising: Receive (200) an initial pulse sequence command (122), the initial pulse sequence command being configured to control the magnetic resonance imaging system (302) to acquire slice-specific k-space data (334) in accordance with a parallel imaging magnetic resonance imaging protocol. Receive (202) calibration data (124); Receive (204) Volume of Interest (126); Receive (206) the slice thickness (128) of at least one slice stack located within the volume of interest. as well as Receive (208) the stack-related stacking orientation (130) of the at least one slice stack. The method further includes performing the following operations on slices in the at least one slice stack: The field of view (132) for the slice is determined (210) by using the volume of interest, the slice thickness, and the stacking orientation of the slice stack. The image metrics within the slice are optimized by using the calibration data to determine (212) the readout direction (136) and the phase encoding direction for the slice; and (214) Slice-specific pulse sequence commands (140) are constructed by modifying the initial pulse sequence command using the field of view, the readout direction and the phase encoding direction.

15. A computer program comprising machine-executable instructions, wherein, Execution of the machine-executable instructions causes the computing system to perform the method according to claim 14.

Images