Computer-implemented method for providing a control sequence to be used, method for acquiring measurement data, provision and / or recording system, computer program and electronically readable data carrier

Abstract

The invention relates to a computer-implemented method for providing a control sequence (38) to be used, which is selective for a target area, for producing a target excitation state for a data acquisition process of a test object with a magnetic resonance device (44), wherein the control sequence (38) comprises high-frequency pulses to be output via transmission channels of a high-frequency coil arrangement (49), wherein the method comprises the following steps: - prior to the investigation encompassing the measurement process, a set of basic sequences (37) for spaced support points (33, 35) covering a parameter interval (34, 36) to be used, of at least one requirement parameter describing the target excitation state, and provision of the basic sequences (37) with the associated support points (33, 35) at the magnetic resonance device (44), - Providing a measurement process value of the request parameter at the magnetic resonance device (44), - Determining a control sequence (38) to be used for the measurement process, wherein, if the measurement process value for the request parameter deviates from the support points (33, 35), the control sequence (38) is determined by interpolation and / or extrapolation using a derivation algorithm from the basic sequences (37).

Description

[0001] The invention relates to a computer-implemented method for providing a control sequence, selective for a target area, for generating a target excitation state for acquiring measurement data from a test object using a magnetic resonance device, wherein the control sequence comprises high-frequency pulses to be output via transmission channels of a high-frequency coil arrangement. The invention further relates to a method for acquiring measurement data with a magnetic resonance device, a provision and / or acquisition system, a computer program, and an electronically readable data carrier.

[0002] In magnetic resonance imaging (MRI), particularly in medical applications, it has been suggested that higher magnetic field strengths for the background magnetic field (B0 field) can be used, as this results in higher image quality of the measurement data. Ultra-high-field magnetic resonance imaging (UHF MRI) is defined as using magnetic field strengths of 7 Tesla or more. However, the high background magnetic field also results in a short Larmor wavelength, which can lead to a spatially highly inhomogeneous high-frequency excitation field (B1 field). This is due, for example, to the relatively good tissue conductivity for high frequencies and reflections at interfaces between different biological tissues or between biological tissue and air. Furthermore, even small deviations in the homogeneity of the background magnetic field lead to significantly different Larmor frequencies.

[0003] This can lead to a situation where a desired target excitation state, for example a specific, homogeneous flip angle of the spins of the object under investigation, is not achieved.

[0004] In this regard, the prior art proposed the use of multiple transmit channels in high-frequency coil arrangements and the parallel transmit (pTx) technique. In this approach, several high-frequency coils (transmit coils) are driven via their respective transmit channels with their own high-frequency pulses of different pulse shapes. The resulting high-frequency excitation field (B1 field) is the interference of the fields generated by the individual high-frequency coils. Its spatial distribution can therefore be controlled.

[0005] A distinction can be made between so-called static pTx and so-called dynamic pTx. In static parallel transmission, all transmission channels are driven with the same high-frequency pulse shape, for example, a square wave or a sinc wave, which is scaled with transmission channel-specific magnitudes and phases. This creates a (static) excitation field that is spatially more homogeneous than an uncalibrated excitation field. In dynamic parallel transmission, time-varying, individually inhomogeneous B1 fields are generated, which are simultaneously applied with equally time-varying B0 gradient fields that describe a transmission k-space trajectory. This allows for very homogeneous flip angle distributions at the end of the dynamic parallel transmission.

[0006] In parallel transmission, it is also known to use multiple high-frequency pulses per transmission channel, which are particularly sequential in time, in order to homogenize the spatial distribution of the flip angles across several high-frequency pulses. This approach is also referred to as multi-pulse pTx.

[0007] To achieve desired target excitation states, optimization methods are typically used to determine the optimization parameters that describe the pulse shapes and timing of the dynamic and / or multi-pulse ptX control sequences. However, optimizing dynamic and multi-pulse control sequences for parallel transmission is extremely complex, computationally intensive, and dependent on the anatomical structure and the position or orientation of the acquisition area on the object under investigation, for example, a patient's body part. Therefore, individual, optimal high-frequency pulses and, if necessary, gradient pulses for a specific measurement procedure can only be calculated once field distribution maps of the object under investigation have been acquired, i.e., once the object is ready for examination.Lengthy optimization processes for the pTx pulses lead to very long and clinically unmanageable delays in the examination.

[0008] Therefore, various approaches have already been proposed in the prior art to enable faster readiness for measurement processes on test objects in magnetic resonance facilities by calculating basic sequences in advance and keeping them available at the magnetic resonance facility. These sequences can be used as control sequences or from which the control sequence to be used is derived.

[0009] The concept of universal pulses has shown that a large proportion of inhomogeneities can be compensated for by pre-calibrated pTx pulses, i.e., pulses optimized before the actual examination, if the universal pulses are determined using field distribution maps of several subjects. In other words, with the universal pulse concept, optimization is performed for a cohort of reference subjects, and the once-optimized pulses are then used for all subjects without further calibration. This approach is described in more detail, for example, in WO 2017 / 060142 A1 or in the article by Vincent Gras et al., “Universal Pulses: A New Concept for Calibration-Free Parallel Transmission,” Magnetic Resonance in Medicine 77 (2017), pages 635 to 643.

[0010] To further improve this approach, methods have been proposed that allow for the rapid implementation of further improvements to pre-calculated basic sequences, for example, using universal pulses, immediately prior to an examination. The concept of standardized universal pulses (SUPs) is explained in an article by Caroline Le Ster et al., “Standardized universal pulse: A fast RF calibration approach to improve flip angle accuracy in parallel transmission,” Magnetic Resonance in Medicine 87 (2022), pages 2839–2850. This article proposes the use of a standardized database in which each B1+ map has been normalized to a reference transmitted RF field distribution. When acquiring a new subject, a rapid acquisition of three B1+ slices is performed to adjust the standardized universal pulses through a linear transformation.

[0011] Another possibility is known as FOCUS and is described, for example, in an article by Jürgen Herrler et al., “Fast online-customized (FOCUS) parallel transmission pulses: A combination of universal pulses and individual optimization”, Magnetic Resonance in Medicine 85 (2021), pages 3140-3153. Reference is also made to EP 3 809 151 A1. This method uses universally optimized transmit k-space trajectories, which can be described by a small number of parameters, high-frequency pulse shapes, and associated optimization-relevant parameters, such as those for energy regularization and / or subpulse durations. After acquiring field distribution maps, the pulses, especially the pulse shapes, of the basic sequences are then further individually optimized.

[0012] In a further development of this approach, EP 3 901 648 A1 proposes a method and a device for controlling a magnetic resonance imaging (MRI) system. Pulse data is selected or calculated in a pulse design unit for generating pulse data to control the MRI system. Starting from a set of B0 and B1 maps of various patients, pulses are designed by interpolation on a grid and / or based on a neural network. A lexicon of selective pulses can be created on the grid, allowing optimal pulses for a measurement to be determined based on an examination scheme. The examination scheme includes, for example, information about the contrasts to be measured and the sequence and shape of subregions of a region of interest to be acquired.

[0013] It has also been proposed, in a more general sense, to assign pre-calculated universal pulses to different clusters. Based on parameters describing the current measurement process, a pulse sequence particularly suitable for the measurement process can be selected from such cluster-specific pulse sequences. These parameters can be related to the object under investigation. However, it has also been suggested, particularly within the framework of the aforementioned FOCUS method, to use field distribution maps, especially B1 maps, as the basis for selecting a cluster-specific pulse sequence. For example, the B1 maps for the current measurement process can be evaluated by a trained cluster determination function, particularly one comprising a neural network, to determine a cluster-specific pulse sequence.

[0014] In an article by Raphael Tomi-Tricot et al., “SmartPulse, a machine learning approach for calibration-free dynamic RF shimming: Preliminary study in a clinical environment,” Magnetic Resonance in Medicine 82 (2019), pages 2016 to 2031, a calibration-free pulse design method is proposed in which a database of field distribution maps was divided into clusters based on the mutual affinity of their associated “kT-points” pulses. A machine learning classifier was trained, based on further field distribution maps, to select the best common “kT-points” pulse from the three available clusters.

[0015] The pre-calculation must in each case be determined for specific target excitation states, and thus for specific sets of target excitation parameters describing those states. This task remains manageable, especially if the excitation is not spatially selective or at least not variably selective. In practice, however, many cases arise where target areas, such as layers, have widely varying orientations, positions, and / or extents, and the target excitation should be restricted to these areas while maintaining the highest possible homogeneity.

[0016] In classical magnetic resonance imaging at lower background magnetic field strengths, it is common practice to use so-called sinc pulses as high-frequency pulses. The frequency spectrum of such high-frequency pulses is rectangular, corresponding to a rectangular function. If a gradient field is applied simultaneously with such a high-frequency pulse, such that the magnetic field strength increases linearly along the layer direction, the Larmor frequencies of the nuclear spins also increase in the layer direction. Only those nuclear spins whose Larmor frequency lies within the frequency band of the high-frequency pulse are excited. In this way, specific layers are selectively excited.

[0017] To determine layer-selective excitations as quickly as possible (and especially without pre-calculation), even for very high background magnetic field strengths, either static parallel transmission can be used for each layer, or simple dynamic pTx pulses, particularly so-called spoke pulses, can be employed. These consist of several subpulses with a sinc waveform. This is also known as a composite pulse. Pulse sequences with such spoke pulses, consisting of multiple subpulses, are optimized so that an optimized pulse is calculated for each layer to be created. To calculate this, the individual gradient blips ("spokes") required between the individual subpulses and the optimal amplitudes and phases of each subpulse are determined for each layer.For such composite pulses / spoke pulses, optimization is relatively quick, as only amplitude and phase factors for scaling the individual subpulses and the moment of the intervening gradient blips need to be optimized. Furthermore, this only needs to be done for the voxels within a single layer, for example, a few thousand voxels. However, such approaches have the disadvantage that they can only compensate for field inhomogeneities within the excited layer to a limited extent. This is because, in the case of static parallel transmission, only the central transmit k-space point is used, and in the case of spoke pulses, only two to three k-space points can be selectively targeted in-plane. Using more subpulses would lead to a longer overall output duration of the drive sequence, thus increasing the measurement time. For certain types of magnetic resonance sequences, especially gradient echo sequences, however, short excitation times are desirable.Short measurement times are necessary.

[0018] An alternative approach to determining better drive sequences with improved radio frequency pulses (and especially gradient pulses) would be optimization for the entire three-dimensional volume starting from significantly more complex radio frequency pulses and, if necessary, gradient pulses (or at least allowing such more complex pulses). To define the target excitation state as a target magnetization, a resolution of 0.2 mm along the layer direction can be assumed, for example, and the target area, such as the layer to be excited, is set to one or a specific target flip angle. For each point of this target magnetization, an additional dimension is added with the values ​​of the transmit-channel-specific B1 maps and the B0 map (field distribution maps). This results in a very high voxel count, which can range, for example, between 100,000 and 1,000,000.For this target excitation state, more complex high-frequency pulses, possibly in conjunction with gradient pulses, can now be optimized in an optimization process. The optimization can be performed under constraints due to the performance and / or power handling capacity of the magnetic resonance instrument, for example, by limiting the maximum high-frequency pulse voltages, the rise rate of the B0 field gradients, and / or the pulse power or pulse energy. Boundary conditions or terms of the objective function can also be formulated more generally based on SAR limitations for the object under investigation. Such an optimization process for a clearly defined special case, i.e., a specific measurement procedure with specific target excitation parameters, can take approximately one hour or even longer.This is clearly not practical for examining test subjects, especially patients, as they would have to spend the entire, extremely long optimization period in the magnetic resonance device in order for the field distribution maps to remain applicable.

[0019] While it would be theoretically possible to use the approaches described above for pre-calculated basic sequences, such as Universalpulse, FOCUS, or SmartPulse, for all conceivable target stimulation parameters or variable requirement parameters selected from them, this would result in pre-calculation times of several months, and in some cases even several years, due to the high variability, particularly regarding the target range.

[0020] The invention is therefore based on the objective of providing a means for the rapid provision of control sequences to be used in an investigation, which are suitable for target excitation states defined by requirement parameters, in particular with regard to the target area, and at the same time sufficiently homogenize the excitation field for clinical application and preferably have a short overall output duration.

[0021] To solve this problem, the invention provides a computer-implemented provisioning method, a capture method, a provisioning and / or recording system, a computer program, and an electronically readable data carrier according to the independent claims. Advantageous embodiments are described in the dependent claims.

[0022] A method of the type mentioned at the outset comprises the following steps according to the invention: - prior to the investigation encompassing the measurement process, a set of basic sequences for spaced support points covering a parameter interval to be used, at least one requirement parameter describing the target excitation state, is pre-calculated, and the basic sequences with the assigned support points are provided at the magnetic resonance device. - Providing a measurement process value of the requested parameter at the magnetic resonance device, - Determining a control sequence to be used for the measurement process, whereby, if the measurement process value for the requirement parameter deviates from the support points, the control sequence is determined from the basic sequences by interpolation and / or extrapolation using a derivation algorithm.

[0023] In particular, it can be provided that the basic sequences and the drive sequence also include at least one gradient pulse for at least one gradient channel of a gradient coil arrangement of the magnetic resonance device. This pulse can also be subject to precalculation and / or interpolation and / or extrapolation. Generally, the invention can use complex pulse shapes for the high-frequency pulses and optionally also for the gradient pulses, optionally also transmit k-space trajectories, which allow the overall output duration of the drive sequence to be kept short, for example, shorter than 1 ms. It should be noted in general that the present invention relates to parallel transmission (pTx), in particular, for the realization of a parallel transmission process, where several transmit channels with respective high-frequency pulses are provided.The method described here is particularly preferred for dynamic and / or multi-pulse pTx, where transmit k-space trajectories are used and, in particular, incorporated into the pre-calculation, especially the optimization process, for the pre-calculated drive sequences. The deployment method (as well as the acquisition method) is preferably used in ultra-high-field magnetic resonance imaging, i.e., at base magnetic field strengths of at least 7 Tesla.

[0024] A fundamental idea of ​​the present invention is to deliberately avoid covering the entire conceivable variability of the requirement parameters during the pre-calculation, but instead to cover only (a few) support points for the then significantly faster pre-calculation, whereby interpolation and / or extrapolation is performed for measurement process values ​​of the requirement parameter that do not correspond to the support points, in order to nevertheless determine a control sequence to be used that provides the target excitation state, in particular with regard to spatial selectivity, in high quality.

[0025] In other words, it is proposed to optimize the parameter space of the request parameters only for specific support points. This takes a longer time, but significantly less than the months or years mentioned. Then, if, during a specific investigation, a measurement process value is requested for at least one request parameter that lies between or outside the support points, a suitable control sequence is calculated by interpolation or extrapolation using at least the (existing) basic sequences for the immediately adjacent support points.

[0026] This also results in significant differences compared to previously known approaches. The SmartPulse method, for example, only selects from pre-calculated universal pulses based on specific properties / parameters (localizer data, height, weight, BMI, etc.) without further interpolation. Consequently, an optimal control sequence can never be selected, as not all parameter combinations are included in the universal pulses. With standardized universal pulses, complex factors for each transmission channel are calculated based on a comparison of the B1 cards of the subject under investigation with reference B1 cards, which served as the basis for the standardized universal pulses. No further parameters beyond the B1 cards are considered. Therefore, this method also fails to produce control sequences better tailored to specific requirements.Furthermore, the SmartPulse method and the method of standardized universal pulses have so far only been developed for non-selective excitations, so that requirements parameters relating to the target area are not addressed.

[0027] It should be noted at this point that if a measurement process value matches a support point, corresponding basic sequences can of course be used directly as control sequences, especially if measurement process values ​​for other requirement parameters do not differ and interpolation or extrapolation is nevertheless required in this regard.

[0028] In a specific, preferred embodiment of the invention, the target area can be a layer, wherein at least one of the at least one requirement parameter from the group comprising a layer thickness, a layer orientation, and a layer position is selected. Requirement parameters can therefore relate in particular to target areas, preferably layers. In other words, the present invention can provide improved spatially selective excitation. For example, for the layer thickness, support points can be defined at intervals of one millimeter and / or for a parameter interval of 1 to 10 mm, and / or for the layer orientation, the support points can comprise an axial, a sagittal, and a coronal orientation. A corresponding subdivision is also conceivable for layer positions. For example, the pre-calculation for the layer thickness can include support points at 1, 2, 3, 4, and 5 mm.If a suitable control sequence for a layer thickness of, for example, 2.5 mm is to be determined, this can be done quickly by means of an interpolation that uses at least the basic sequences for 2 mm and 3 mm.

[0029] However, requirement parameters can also relate to other aspects of the target excitation state or the measurement process in general. For example, it may be provided that at least one of the at least one requirement parameter is selected from the group comprising a target flip angle and an energy parameter describing a desired energy regularization. For example, ten support points for (particularly frequently used) target flip angles can be selected, whereby further target flip angles can be covered by interpolation or extrapolation. The procedure according to the invention can also be extended to SAR treatment by having at least one requirement parameter relate to the energy regularization, which concerns the energy input, optionally with local resolution, into the object under investigation by reaching the target excitation state.In this regard, for example, with the known method of standardized universal pulses, it was previously known to simply lengthen the high-frequency pulses and / or gradient pulses by a factor and reduce the voltages (pulse shapes) by the same factor when SAR limits were violated. Within the scope of the present invention, a significantly improved treatment of SAR problems, tailored to the actual circumstances, is possible.

[0030] The pre-calculated base sequences are already available before imaging of the subject takes place. For example, they are provided during the manufacturing process or as part of a routine update of the magnetic resonance imaging (MRI) system.

[0031] The basic sequences can preferably be determined in an optimization process and / or based on reference field distribution maps, comprising at least one B0 map and at least one B1 map, of a cohort of reference investigation objects. This means that, for example, procedures already proposed for universal pulses and the like, including a comprehensive optimization, can be used to determine the basic sequences in the pre-computation process. In this preferably comprehensive optimization process, complex high-frequency pulses and, if necessary, gradient pulses are determined for a multitude of optimization parameters. In particular, pulse shapes and / or the time sequence can also be the subject of optimization in the optimization process.The pre-calculations for determining the predicted control sequences in optimization processes are typically extremely complex and result in highly intricate pulse shapes that are not easy to describe mathematically, for which a large number of variables to be optimized are used. For example, the pulses of the predicted control sequences (and also of the resulting control sequences to be used) may not be temporally and / or spatially coherent.

[0032] It may be possible to determine and provide several sets of base sequences, each assigned to a cluster within the cohort. For example, the cohort of reference test objects, or the reference field distribution maps, can be divided into classes of comparable cases, so-called clusters. The optimization process is then performed separately for each cluster to obtain optimal base sequences for the clusters as a whole. Regarding interpolation or extrapolation, it may be possible to use only the base sequences whose clusters result from the current measurement process. This can be deduced, for example, from field distribution maps recorded on the test object during the measurement process. This will be discussed in more detail below.

[0033] In principle, it is conceivable within the scope of the present invention to use classical interpolation and / or extrapolation methods in the derivation algorithm. For this purpose, for example, a model can be defined and / or Bloch simulations can be performed. However, the use of such classical interpolation and / or extrapolation methods, which can, for example, utilize fits, is rather time-consuming. Moreover, the interpolation or extrapolation problem is generally non-convex and, in particular, non-linear.

[0034] Therefore, a particularly preferred embodiment of the present invention provides that the derivation algorithm includes the use of a trained subfunction to perform the interpolation and / or extrapolation. Artificial intelligence, particularly in the form of a trained neural network, can thus be used to perform the interpolation and / or extrapolation. Artificial intelligence is also characterized by particularly fast computation times, so that the use of a trained subfunction is superior in this respect to "classical" methods that do not rely on artificial intelligence.

[0035] In general, a trained function replicates cognitive functions that people associate with other human brains. Through training based on training data (machine learning), the trained function is able to adapt to new circumstances and detect and extrapolate patterns.

[0036] Generally speaking, the parameters of a trained function can be adjusted through training. Specifically, supervised learning, semi-supervised learning, unsupervised learning, reinforcement learning, and / or active learning can be used. Furthermore, representational learning (also known as feature learning) can be employed. The parameters of the trained function can be adjusted iteratively through multiple training steps.

[0037] A trained function can, for example, comprise a neural network, a support vector machine (SVM), a decision tree, and / or a Bayesian network, and / or the trained function can be based on k-means clustering, Q-learning, genetic algorithms, and / or assignment rules. Specifically, a neural network can be a deep neural network, a convolutional neural network (CNN), or a deep CNN. Furthermore, the neural network can be an adversarial network, a deep adversarial network, and / or a generative adversarial network (GAN).

[0038] In the present case, it can be particularly advantageous to include a CNN, especially a U-Net, in the subfunction. This has yielded particularly positive results.

[0039] Specifically, it can be provided that the subfunction is made available as the result of a training process in which the subfunction is trained based on fundamental truths obtained in a pre-computation process also used for the base sequences. This can involve, firstly, pre-computing, and in particular simulating, additional values ​​of the request parameter located between support points in order to determine fundamental truths for training datasets. However, it is also conceivable, and especially additionally, to assume support point base sequences are unknown, thus using them as fundamental truths, and to allow them to be interpolated / extrapolated from the other base sequences.

[0040] The training process can be performed separately for at least one reference set of field distribution maps. The reference set for a cluster can contain representative field distribution maps. If the training is based on a specific reference set of field distribution maps, variability with respect to these maps is eliminated, allowing optimization processes for determining base sequences and / or fundamental truths (of values ​​to be interpolated / extrapolated of the at least one requirement parameter) to be completed more quickly. The training then takes place on these base sequences. For example, representative field distribution maps for a cluster can be used to determine additional fundamental truths for training, even for clusters for which the base sequences are already available.

[0041] Preferably, field distribution maps acquired from the object under investigation, comprising a B0 map and at least one B1 map, can be provided, and the derivation algorithm can use the field distribution maps as input data in addition to the measurement process value. The required field distribution maps can be acquired with the magnetic resonance device itself, particularly with a recording duration of several tens of seconds, especially 30 to 60 seconds. The field distribution maps allow the properties of the object under investigation to be taken into account. B1 maps each describe the spatial B1 field distribution for a specific transmitting coil of the transmitting coil arrangement, i.e., for a specific transmitting channel of the high-frequency coil arrangement. In other words, the B1 maps describe the spatial sensitivity of the corresponding transmitting coil.B0 maps describe, with spatial resolution, the deviations of the fundamental magnetic field (B0 field) from the desired homogeneous field pattern, and thus the local deviation of the Larmor frequency from the desired (nominal) Larmor frequency.

[0042] Within the scope of the present invention, field distribution maps can be used for several different purposes. For example, it can be provided that, within the derivation algorithm, the field distribution maps are used as a starting point for assigning them to a cluster and for selecting the basic sequences assigned to the cluster. It should be noted that artificial intelligence can also be used for such an assignment to a cluster, for example in the form of a trained cluster determination function.

[0043] It can also be provided, for example, that the trained subfunction uses the field distribution maps as input data. This is particularly useful if the investigations associated with new measurement process values ​​also result in more highly variable field distribution maps, possibly even within a cluster. Changes in the B0 and / or B1 field distribution that occur due to the different value of the requirement parameter, especially with regard to the target range, can thus also be taken into account, for example, in extensions with regard to large slabs.

[0044] In a particularly advantageous embodiment of the present invention, the derivation algorithm may include a step following the interpolation and / or extrapolation for optimizing the derived control sequence to be used in order to optimally achieve the target excitation state, taking into account the field distribution maps in at least one optimization process. Approaches such as the FOCUS method can be used here, where the interpolated and / or extrapolated control sequence ultimately serves as the starting point for further optimization based on the current field distribution maps. In this case, the interpolation and / or extrapolation provides, so to speak, starting values ​​for achieving further improvement through an optimization process.For example, knowing the excellent starting point for the optimization process, a Bloch simulation can be used to compare the results with the target excitation state. Based on the results, an optimization algorithm, such as a gradient descent method, can then be used to improve the results by adjusting a specific set of optimization parameters. Objective functions can relate to the degree to which the target excitation state is achieved and / or the energy input into the device under investigation, and these aspects can also be covered by boundary conditions. Further boundary conditions can relate to the performance and / or resilience of the magnetic resonance instrument or its corresponding components.

[0045] While the basic sequences were indeed determined in an optimization process based on corresponding field distribution maps, deviations from the optimum can still occur due to interpolation and / or extrapolation. Therefore, an additional, "classical" optimization process, performed after the interpolation / extrapolation, can improve the quality. This also increases the reliability and robustness of the control sequence to be used.

[0046] In an inventive method for acquiring measurement data of a test object using a magnetic resonance device, a control sequence to be used is provided by means of a computer-implemented provisioning method according to the invention, and during the acquisition process, the high-frequency pulses of the control sequence to be used are output via the high-frequency coil arrangement to establish the target excitation state. If the control sequence to be used also includes gradient pulses, the gradient pulses are output via gradient coils of a gradient coil arrangement of the magnetic resonance device. All aspects relating to the computer-implemented provisioning method can be applied analogously to the acquisition method according to the invention, so that the aforementioned advantages can also be obtained with it.In particular, the acquisition method according to the invention is suitable for clinical use, i.e. for medical imaging, due to its rapid provision of the control sequence to be used, in order to obtain high-quality measurement data and thus especially magnetic resonance image datasets, particularly when control sequences with a short total output duration are required, for example in gradient echo sequences.

[0047] As already mentioned, it may be advantageous to record the field distribution maps with the magnetic resonance device before providing the control sequence to be used. Corresponding, rapid magnetic resonance sequences, for example, exemplified by those executable in less than one minute, have already been proposed in the prior art. Alternatively, it would be conceivable, for example, to use measuring probes or similar devices to measure the field distribution maps.

[0048] The invention also relates to a data provision and / or acquisition system with a computing unit. The computing unit can be configured to perform both a data provision method and an acquisition method according to the invention. In particular, in the case of an acquisition system, it can also include a magnetic resonance device comprising a high-frequency coil arrangement and a control unit, which can form part of the computing unit. In particular, the magnetic resonance device can also include a gradient coil arrangement for outputting gradient pulses and / or a base magnetic field strength of at least 7 Tesla, i.e., be configured for ultra-high-field magnetic resonance imaging.

[0049] Particularly in the case of a deployment system, this system, specifically the computing unit, can also include an optimization unit, for example, in a data center. The optimization unit can be configured to pre-calculate the basic sequences. The basic sequences can then be provided to the magnetic resonance imaging (MRI) device, for example, during or after its manufacture and / or during an update. The control unit of the MRI device can, in addition to a storage medium for, among other things, the basic sequences, include an interface for receiving the measurement process value for at least one request parameter and a derivation unit for executing the derivation algorithm.For the acquisition process, the control unit can further comprise a sequence unit that generally controls the recording operation of the magnetic resonance device and is thus specifically designed to control the acquisition process and, if applicable, the recording of the field distribution maps. Additional functional units can also be provided for implementing further developments of the methods according to the invention. All aspects relating to the methods according to the invention can be applied analogously to the provision and / or acquisition system according to the invention, so that the aforementioned advantages can also be obtained with this system.

[0050] A computer program according to the invention can be directly loaded into a storage medium of a computing device and comprises program elements which, when the computer program is executed, cause the computing device to perform the steps of a method according to the invention. The computer program can be stored on an electronically readable data carrier according to the invention, which therefore includes control information stored thereon, comprising at least one computer program according to the invention and designed such that, when the data carrier is used in a computing device, the latter is configured to perform a method according to the invention.When the acquisition method is implemented as a computer program, a control unit of the computing device is instructed to acquire measurement data and / or field distribution maps, and to control further components of a magnetic resonance device, in particular the high-frequency coil arrangement and the gradient coil arrangement, accordingly.

[0051] Further advantages and details of the present invention will become apparent from the exemplary embodiments described below and from the drawings. These show: Fig. 1 an embodiment of an artificial neural network, Fig. 2 an example implementation of a Convolutional Neural Network, Fig. 3 a flowchart of an exemplary embodiment of a detection method according to the invention, Fig. 4. A schematic diagram for providing basic functions for support points and their use for interpolation and / or extrapolation, and Fig. 5 a schematic diagram of a provision and reception system according to the invention.

[0052] Fig. Figure 1 shows an embodiment of an artificial neural network 1. English terms for the artificial neural network 1 are “artificial neural network”, “neural network”, “artificial neural net” or “neural net”.

[0053] Artificial neural network 1 comprises nodes 6 to 18 and edges 19 to 21, where each edge 19 to 21 is a directed connection from a first node 6 to 18 to a second node 6 to 18. Generally, the first node 6 to 18 and the second node 6 to 18 are distinct nodes 6 to 18; however, it is also conceivable that the first node 6 to 18 and the second node 6 to 18 are identical. For example, in Fig. 1. Edge 19 is a directed connection from node 6 to node 9, and edge 21 is a directed connection from node 16 to node 18. An edge 19 to 21 from a first node 6 to 18 to a second node 6 to 18 is called an incoming edge for the second node 6 to 18 and an outgoing edge for the first node 6 to 18.

[0054] In this embodiment, the nodes 6 to 18 of the artificial neural network 1 can be arranged in layers 2 to 5, wherein the layers can have an intrinsic order introduced by the edges 19 to 21 between the nodes 6 to 18. In particular, edges 19 to 21 can only be provided between adjacent layers of nodes 6 to 18. In the illustrated embodiment, there is an input layer 2 that contains only nodes 6, 7, and 8, each without an incoming edge. The output layer 5 comprises only nodes 17 and 18, each without outgoing edges, with hidden layers 3 and 4 further situated between the input layer 2 and the output layer 5. In the general case, the number of hidden layers 3 and 4 can be chosen arbitrarily.The number of nodes 6, 7, 8 in input layer 2 usually corresponds to the number of input values ​​into neural network 1, and the number of nodes 17, 18 in output layer 5 usually corresponds to the number of output values ​​of neural network 1.

[0055] In particular, a (real) number can be assigned to nodes 6 to 18 of neural network 1. Here, x denotes... (n) Let i be the value of the i-th node 6 to 18 of the n-th layer 2 to 5. The values ​​of nodes 6, 7, 8 of input layer 2 are equivalent to the input values ​​of neural network 1, while the values ​​of nodes 17, 18 of output layer 5 are equivalent to the output values ​​of neural network 1. Furthermore, each edge 19, 20, 21 can be assigned a weight in the form of a real number. In particular, the weight is a real number in the interval [-1, 1] or in the interval [0, 1, ]. Here, w denotes (m,n)i,j is the weight of the edge between the i-th nodes 6 to 18 of the m-th layer 2 to 5 and the j-th nodes 6 to 18 of the n-th layer 2 to 5. Furthermore, the abbreviation wi,j(n) for the weight wi,j(n,n+1) defined.

[0056] To calculate the output values ​​of neural network 1, the input values ​​are propagated through neural network 1. In particular, the values ​​of nodes 6 to 18 of the (n+1)th layer 2 to 5 can be calculated based on the values ​​of nodes 6 to 18 of the nth layer 2 to 5 by xj(n+1)=f(∑ixi(n)⋅wi,j(n)).

[0057] Here, f is a transfer function, which can also be called an activation function. Well-known transfer functions include step functions, sigmoid functions (for example, the logistic function, the generalized logistic function, the hyperbolic tangent, the arctangent, the error function, the smoothstep function), and rectifier functions. The transfer function is primarily used for normalization purposes.

[0058] Specifically, the values ​​are propagated layer by layer through neural network 1, with values ​​of input layer 2 being given by the input data of neural network 1. Values ​​of the first hidden layer 3 can be calculated based on the values ​​of input layer 2 of neural network 1, values ​​of the second hidden layer 4 can be calculated based on the values ​​in the first hidden layer 3, and so on.

[0059] To determine the values wi,j(n) To be able to define the parameters for edges 19 to 21, neural network 1 must be trained using training data. Specifically, training data includes training input data and training output data, which are referred to below as t. i The training step involves applying neural network 1 to the training input data to determine the calculated output data. Specifically, the training output data and the calculated output data comprise a number of values, where this number is determined by the number of nodes 17 and 18 in output layer 5.

[0060] In particular, a comparison between the calculated output data and the training output data is used to recursively adjust the weights within neural network 1 (backpropagation algorithm). Specifically, the weights can be adjusted accordingly. w′i,j(n)=wi,j(n)−γ⋅δj(n)⋅xi(n) to be changed, where γ is a learning rate and the numbers δj(n) can be calculated recursively as δj(n)=(∑kδk(n+1)⋅wi,k(n+1))⋅f′(∑ixi(n)⋅wi,j(n)) based on δj(n+1), if the (n+1)th layer is not the starting layer 5, and δj(n)=(xk(n+1)−tj(n+1))⋅f′(∑ixi(n)⋅wi,j(n)) if the (n+1)th layer is the output layer 5, where f' is the first derivative of the activation function and yj(n+1) the comparison training value for the j-th node 17, 18 of the initial layer 5 is.

[0061] The following will be discussed with regard to Fig. Section 2 also provides an example of a Convolutional Neural Network (CNN). It is important to note that the term "layer" is used slightly differently there than in classical neural networks. For a classical neural network, the term "layer" refers only to the set of nodes that form a layer, i.e., a specific generation of nodes. For a Convolutional Neural Network, the term "layer" is often used to describe an object that actively modifies data; in other words, a set of nodes of the same generation and either the set of incoming or outgoing edges.

[0062] Fig. Figure 2 shows an embodiment of a convolutional neural network 22. In the illustrated embodiment, the convolutional neural network 22 comprises an input layer 23, a convolutional layer 24, a pooling layer 25, a fully connected layer 26, and an output layer 27. In alternative embodiments, the convolutional neural network 22 can contain multiple convolutional layers 24, multiple pooling layers 25, and multiple fully connected layers 26, just like other types of layers. The order of the layers can be chosen arbitrarily, with fully connected layers 26 typically forming the last layers before the output layer 27.

[0063] In particular, within a Convolutional Neural Network 22, the nodes 28 to 32 of one of the layers 23 to 27 can be understood as arranged in a d-dimensional matrix or as a d-dimensional image. Specifically, in the two-dimensional case, the value of a node 28 to 32 with the indices i, j in the nth layer 23 to 27 can be represented as x (n) [i,j] It should be noted that the arrangement of nodes 28 to 31 of a layer 23 to 27 has no effect whatsoever on the calculations within the Convolutional Neural Network 22 as such, since these effects are determined solely by the structure and the weights of the edges.

[0064] A convolution layer 24 is particularly distinguished by the fact that the structure and weights of the incoming edges form a convolution operation based on a specific number of kernels. In particular, the structure and weights of the incoming edges can be chosen such that the values xk(n) node 29 of the folding layer 24 as a folding xk(n)=Kk*x(n−1) based on the values ​​x (n-1) The node 28 of the preceding layer 23 is determined, where the folding * in the two-dimensional case can be defined as xk(n)[i,j]=(Kk*x(n−1))[i,j]=∑i′∑j′Kk[i′,j′]⋅x(n−1)[i−i′, j−j′].

[0065] In this, the k-th kernel is K ka d-dimensional matrix, in this embodiment a two-dimensional matrix, which is usually small compared to the number of nodes 28 to 32, for example a 3x3 matrix or a 5x5 matrix. In particular, this implies that the weights of the incoming edges are not independent, but are chosen such that they generate the convolution equation above. In the example for a kernel forming a 3x3 matrix, there are only nine independent weights (where each entry of the kernel matrix corresponds to an independent weight), regardless of the number of nodes 28 to 32 in the corresponding layer 23 to 27. In particular, for a convolution layer 24, the number of nodes 29 in the convolution layer 24 is equivalent to the number of nodes 28 in the preceding layer 23 multiplied by the number of convolution kernels.

[0066] If the nodes 28 of the preceding layer 23 are arranged as a d-dimensional matrix, the use of multiple kernels can be understood as adding another dimension, also called the depth dimension, so that the nodes 29 of the convolution layer 24 are arranged as a (d+1)-dimensional matrix. If the nodes 28 of the preceding layer 23 are already arranged as a (d+1)-dimensional matrix with one depth dimension, the use of multiple convolution kernels can be understood as an expansion along the depth dimension, so that the nodes 29 of the convolution layer 24 are likewise arranged as a (d+1)-dimensional matrix, the size of the (d+1)-dimensional matrix in the depth dimension being larger than in the preceding layer 23 by the factor formed by the number of kernels.

[0067] The advantage of using convolutional layers 24 is that the spatially local correlation of the input data can be exploited by creating a local connection pattern between nodes of neighboring layers, in particular by ensuring that each node has connections only to a small area of ​​the nodes of the preceding layer.

[0068] In the illustrated embodiment, the input layer 23 comprises thirty-six nodes 28 arranged as a two-dimensional 6x6 matrix. The convolution layer 24 comprises seventy-two nodes 29 arranged as two two-dimensional 6x6 matrices, each of which is the result of convolution of the values ​​of the input layer 23 with a convolution kernel. Similarly, the nodes 29 of the convolution layer 24 can be understood as arranged in a three-dimensional 6x6x2 matrix, the latter dimension being the depth dimension.

[0069] A pooling layer 25 is characterized by the fact that the structure and weights of the incoming edges, as well as the activation function of their nodes 30, define a pooling operation based on a nonlinear pooling function f. For example, in the two-dimensional case, the values ​​x (n) Node 30 of pooling layer 25 based on the values ​​x (n+1) the node 29 of the preceding layer 24 as x(n)[i,j]=f(x(n−1)[id1, jd2], …, x(n−1)[id1+d1−1, jd2+d2−1]) The number of nodes 29, 30 can be calculated by using a pooling layer 25. This is achieved by replacing a number of d1 × d2 adjacent nodes 29 in the preceding layer 24 with a single node 30, which is calculated as a function of the values ​​of the aforementioned number of adjacent nodes 29. Specifically, the pooling function f can be a maximum function, an averaging function, or the L2 norm. In particular, for a pooling layer 25, the weights of the incoming edges can be fixed and not modified by training.

[0070] The advantage of using a pooling layer 25 is that the number of nodes 29, 30 and the number of parameters are reduced. This leads to a reduction in the required computational effort within the Convolutional Neural Network 22 and thus to control overfitting.

[0071] In the illustrated embodiment, pooling layer 25 is a max-pooling layer in which four adjacent nodes are replaced by a single node whose value is the maximum of the values ​​of the four adjacent nodes. Max-pooling is applied to each d-dimensional matrix of the preceding layer; in this embodiment, max-pooling is applied to each of the two two-dimensional matrices, thus reducing the number of nodes from seventy-two to eighteen.

[0072] A fully connected layer 26 is characterized by the presence of a plurality, in particular all, edges between the nodes 30 of the preceding layer 25 and the nodes 31 of the fully connected layer 26, the weight of each edge being individually adjustable. In this embodiment, the nodes 30 of the preceding layer 25 and the fully connected layer 26 are shown both as two-dimensional matrices and as non-connected nodes (represented as a row of nodes, the number of which has been reduced for clarity). In this embodiment, the number of nodes 31 in the fully connected layer 26 is equal to the number of nodes 30 in the preceding layer 25. In alternative embodiments, the number of nodes 30 and 31 can be different.

[0073] Furthermore, in this embodiment, the values ​​of the nodes 32 of the output layer 27 are determined by applying the softmax function to the values ​​of the nodes 31 of the preceding layer 26. By applying the softmax function, the sum of the values ​​of all nodes 32 of the output layer 27 is one, and all values ​​of all nodes 32 of the output layer are real numbers between 0 and 1. If the convolutional neural network 22 is used to classify input data, the values ​​of the output layer 27 can, in particular, be interpreted as the probability that the input data falls into one of the different classes.

[0074] A Convolutional Neural Network 22 can also have a ReLU layer, where ReLU is an acronym for "rectified linear units". Specifically, the number of nodes and the structure of the nodes within a ReLU layer are equivalent to the number of nodes and the structures of the nodes in the preceding layer. The value of each node in the ReLU layer can be calculated, in particular, by applying a rectifier function to the value of the corresponding node in the preceding layer. Examples of rectifier functions are f(x) = max(0,x), the hyperbolic tangent, or the sigmoid function.

[0075] Convolutional Neural Networks 22 can be trained, in particular, based on the backpropagation algorithm. To avoid overfitting, regularization methods can be used, such as dropout of individual nodes 28 to 32, stochastic pooling, use of artificial data, weight decay based on the L1 or L2 norm, or maximum norm restrictions.

[0076] Fig. Figure 3 shows a flowchart of an embodiment of a detection method according to the invention, which includes a provision method according to the invention.

[0077] The provisioning method serves to provide a control sequence for generating a target excitation state for the acquisition of measurement data from a test object using a magnetic resonance device (MRI). This acquisition process is then carried out as part of the MRI procedure. The MRI has a main magnet that generates a fundamental magnetic field (B0 field) with a field strength of 7 Tesla or more. A high-frequency coil array of the MRI has several high-frequency coils that can be controlled via corresponding transmission channels to excite the nuclear spins of the test object to generate the target excitation state. Furthermore, a gradient coil array is provided, comprising several gradient coils (in this case, three), each assigned to one of the Cartesian spatial directions (X, Y, Z) of the MRI.The gradient coils can be controlled via corresponding gradient channels to output gradient pulses.

[0078] To establish the target excitation state, dynamic parallel transmission is used, so that the control sequence comprises, on the one hand, gradient pulses, particularly for implementing a transmit k-space trajectory, and on the other hand, high-frequency pulses for the transmit channels. These each have highly complex pulse shapes. Multiple high-frequency pulses per transmit channel can also be used (multi-pulse pTx).

[0079] The target excitation state here involves spatially selective excitation, meaning that the generation of a homogeneous flip angle distribution should only occur in a target region, in this case, a single layer; other nuclear spins of the subject should not be excited. The layer is defined by several parameters, namely the layer thickness, the orientation of the layer, and the layer position, for example, defined by the position of the layer's central point. Other parameters relating to the target excitation state can also be used, such as the desired flip angle and at least one energy parameter related to energy regularization. This energy parameter should limit energy inputs that are relevant to the SAR (Specific Action Rate) of the subject, in this case, a patient.

[0080] To find the most suitable control sequence for such measurement processes and target excitation states, a preliminary preparation step (S1) is required. In this step, basic sequences are pre-calculated through optimization processes. These basic sequences then serve as the basis for determining the control sequence for specific applications. However, these basic sequences are not determined for all possible values ​​of the required parameters, but only for a limited number of reference points. For example, reference points at 1 mm, 2 mm, 3 mm, 4 mm, and 5 mm can be used to cover a parameter range of 1 to 5 mm, or even larger, such as 0.5 to 5.5 mm with possible extrapolation.Regarding layer orientation, three support points can be used: the primarily used coronal orientation, the sagittal orientation, and the axial orientation. For layer position, similar support points are used as for layer thickness, for example, on a grid. For instance, 27 support points are conceivable here. As for the flip angle, 10 support points can be used, for example.

[0081] This is schematically illustrated for two requirement parameters in Fig. Figure 4 illustrates, for example, the orientation of the layer with the three support points 33 in parameter interval 34 and the layer thickness with the five support points 35 in parameter interval 36. Through optimization processes, which can be carried out as is generally known in the prior art, the support points are now determined in each case. Fig. 4. The basic sequences 37, which are also indicated, are pre-calculated. The parameter space for the requirement parameters is thus covered selectively, for example in the form of a grid.

[0082] The optimization processes presented here are based on field distribution maps, i.e., B0 maps and B1 maps, for a cohort of objects under investigation, from which the optimization is performed. The objects under investigation, or rather their field distribution maps, are divided into clusters, as is generally known, so that sets of basic sequences 37 (for the various support points 33, 35) are pre-calculated for each cluster. Each cluster can also be assigned a representative set of B0 maps and B1 maps. The optimization process then aims to achieve the target excitation state as precisely as possible for all field distribution maps of the cluster. For this purpose, magnetizations of the optimization parameters can be determined, for example, by Bloch simulations for test sets and compared with the magnetization defined by the target excitation state, on which the test set can then be adjusted.Boundary conditions can also relate to SAR, the performance and the load-bearing capacity of the magnetic resonance device.

[0083] In one step S2 (see above). Fig. 3) The pre-calculated basic sequences 37 are provided at the magnetic resonance device. For example, they are stored in a storage medium of a control unit of the magnetic resonance device after its production or during an update.

[0084] During normal operation of the magnetic resonance imaging (MRI) system, measurement parameters are provided for measurement processes, for example, based on user input and / or from an information system, such as a radiology information system, and / or based on the selection of an acquisition program on the MRI system. In step S3, it is therefore checked whether a measurement process with associated measurement process values ​​of the request parameters is pending. If this is the case, a derivation algorithm is performed in step S4 to determine the control sequence to be used.

[0085] When clusters are used, a current cluster can first be determined by evaluating current field distribution maps for the current object of investigation, acquired with the magnetic resonance instrument, for example by a trained cluster determination function, as is generally known in the prior art. Only the basic functions 37 assigned to this current cluster are used subsequently.

[0086] The derivation algorithm has two substeps, namely step S4a and step S4b. In step S4a, a trained subfunction comprising a CNN 22 is executed to derive, where necessary, the control sequence 38 to be used from the basic sequences 37 by interpolation, as shown in Fig. Figure 4 is also shown schematically. The requirement parameters for orientation and layer thickness lie between any two of the support points 33, 35, so that interpolation is necessary for at least these two requirement parameters. If the measurement process value of a requirement parameter corresponds to one of the support points 33, 35, then of course no interpolation is necessary in this respect. In this case, at least the basic sequences 37 of the immediately adjacent support points 33, 35 are included in the interpolation, as indicated by the arrows 39 in Figure 4. Fig. 4 is indicated.

[0087] The trained subfunction is provided after a training process has been performed. Training datasets can be provided for this process in different ways: firstly, through further optimization processes for values ​​of the requirement parameters lying between support points 33 and 35; and secondly, by considering the basic sequences 37 for certain support points 33 and 35 as fundamental truths to be derived from neighboring basic sequences 37. For further optimization processes, only the representative set of B0 map and B1 field maps is used as the reference set for each cluster in this case, in order to simplify matters.

[0088] It is advantageous to include the field distribution maps as input data during the training process of the subfunction. In subsequent use, the trained subfunction also uses the field distribution maps of the current object under investigation as input data. This allows deviations, particularly within clusters, to be taken into account.

[0089] In substep S4b, the control sequence derived in substep S4a is further optimized using the field distribution maps of the current object under investigation, thus better adapting it to the specific recording situation. Techniques such as those used in the FOCUS method can be applied here, or, particularly for a reduced number of optimization parameters, standard optimization algorithms can be performed to achieve the most accurate possible realization of the target magnetization. For example, an objective function can be used that minimizes the deviation of a magnetization obtained with a test set of optimization parameters in a Bloch simulation from the target magnetization according to the target excitation state.Boundary conditions can relate to the energy input into the object under investigation (SAR considerations) as well as the performance and / or load capacity of the magnetic resonance instrument. New test sets can be determined, for example, using a gradient descent method.

[0090] In step S5, the optimized control sequence is then used to perform the measurement process with the magnetic resonance device in order to acquire measurement data from which, for example, a magnetic resonance image dataset can be determined. For this purpose, the high-frequency pulses are output via the high-frequency coil arrangement and the gradient pulses via the gradient coil arrangement.

[0091] Fig.Figure 5 shows a schematic diagram of a provision and recording system 40 according to the invention. This system comprises, firstly, a computing device 41 designed to carry out the method according to the invention, which in this case consists of two main components: an optimization device 42, which can, for example, be part of a data center and has high computing power, and the control device 43 of a magnetic resonance device 44, which also forms part of the provision and recording system 40.

[0092] The optimization unit 42 is designed to pre-calculate the basic sequences 37, see step S1, which are then provided to the control unit 43 via a corresponding interface, see step S2. In the control unit 43, they can be stored in a storage medium 45.

[0093] The magnetic resonance device 44 comprises, as is generally known, a main magnet unit 46, which includes the superconducting main magnet (not shown in detail here) for generating the fundamental magnetic field with a fundamental magnetic field strength of at least 7 Tesla. The main magnet unit 46 defines a cylindrical patient receptacle 47 into which a patient can be moved using a patient table (not shown in detail here). Surrounding the patient receptacle 47 are, in this case, a gradient coil arrangement 48 and a high-frequency coil arrangement 49, wherein the high-frequency coil arrangement 49, or a further high-frequency coil arrangement, can also be implemented as a local coil arrangement.

[0094] The gradient coil arrangement 48 comprises gradient coils that can be controlled via corresponding gradient channels, while the high-frequency coil arrangement 49 comprises transmit coils that can be controlled via corresponding transmit channels. The control unit 43 controls the operation of the magnetic resonance device 44.

[0095] The recording operation of the magnetic resonance device 44 is controlled in particular by the sequence unit 50 of the control unit 43, which can therefore also control the recording of the field distribution maps and the measurement data in step S5. In a derivation unit 51, the derivation algorithm can be carried out according to step S4.

[0096] Although the invention has been illustrated and described in detail by the preferred embodiment, the invention is not limited by the disclosed examples and other variations can be derived by the person skilled in the art without leaving the scope of protection of the invention.

[0097] Regardless of the grammatical gender of a particular term, persons with male, female or other gender identities are included. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] WO 2017 / 060142 A1

[0009] EP 3 809 151 A1

[0011] EP 3 901 648 A1

[0012] Zitierte Nicht-Patentliteratur

[0000] Vincent Gras et al., „Universal Pulses: A New Concept for Calibration-Free Parallel Transmission“, Magnetic Resonance in Medicine 77 (2017), Seiten 635 bis 643

[0009] Caroline Le Ster et al., „Standardized universal pulse: A fast RF calibration approach to improve flip angle accuracy in parallel transmission“, Magnetic Resonance in Medicine 87 (2022), Seiten 2839-2850

[0010] Jürgen Herrler et al., „Fast online-customized (FOCUS) parallel transmission pulses: A combination of universal pulses and individual optimization“, Magnetic Resonance in Medicine 85 (2021), Seiten 3140-3153

[0011] Raphael Tomi-Tricot et al., „SmartPulse, a machine learning approach for calibration-free dynamic RF shimming: Preliminary study in a clinical environment“, Magnetic Resonance in Medicine 82 (2019), Seiten 2016 bis 2031

[0014]

Claims

[1] Computer-implemented method for providing a control sequence (38) to be used, which is selective for a target area, for establishing a target excitation state for a process of acquiring measurement data of an examination object with a magnetic resonance device (44), wherein the control sequence (38) comprises radio-frequency pulses to be output via transmission channels of a radio-frequency coil arrangement (49), the method comprising the following steps: - before the examination comprising the measurement process, pre-calculating a set of basic sequences (37) for spaced-apart support points (33, 35) of at least one request parameter, which describes the target excitation state, covering a parameter interval (34, 36) to be used, and providing the basic sequences (37) with the associated support points (33, 35) at the magnetic resonance device (44), - providing a measurement process value of the request parameter at the magnetic resonance device (44), - Determining a control sequence (38) to be used for the measuring process, wherein, if the measuring process value for the request parameter deviates from the support points (33, 35), the control sequence (38) is determined by interpolation and / or extrapolation using a derivation algorithm from the basic sequences (37). [2] Method according to claim 1, characterized by that the target area is a layer, wherein at least one of the at least one requirement parameter is selected from the group comprising a layer thickness, a layer orientation and a layer position. [3] Method according to claim 1 or 2, characterized by that at least one of the at least one requirement parameter is from the group comprising a target flip angle and an energy parameter describing a desired energy regularization. [4] Method according to one of the preceding claims, characterized by that the base sequences (37) are determined in an optimization process and / or based on reference field distribution maps, comprising at least one B0 map and at least one B1 map, of a cohort of reference examination objects. [5] Method according to claim 4, characterized by that several sets of base sequences (37), each assigned to a cluster of the cohort, are determined and provided. [6] Method according to one of the preceding claims, characterized by that the derivation algorithm includes the use of a trained subfunction to perform the interpolation and / or extrapolation. [7] Method according to claim 6, characterized by that the subfunction comprises a CNN (22), in particular a U-Net. [8] Method according to claim 6 or 7, characterized bythat the subfunction is provided as a result of a training process in which the subfunction is trained on the basis of ground truths obtained in a pre-computation process also used for the base sequences (37). [9] Method according to claim 8, characterized by that the training process is carried out separately for at least one reference set of reference field distribution maps. [10] Method according to one of the preceding claims, characterized by that field distribution maps recorded on the object under investigation, comprising a B0 map and at least one B1 map, are provided and the derivation algorithm uses the field distribution maps as input data in addition to the measurement process value. [11] Method according to claim 10, characterized bythat the derivation algorithm has a step (S4b) following the interpolation and / or extrapolation for optimizing the control sequence (38) derived there to be used for optimally achieving the target excitation state, taking into account the field distribution maps in at least one optimization process. [12] Method for acquiring measurement data of an examination object with a magnetic resonance device (44), wherein a control sequence (38) to be used is provided by means of a method according to one of the preceding claims and, in the acquisition process for establishing the target excitation state, the radio-frequency pulses of the control sequence (38) to be used are output by means of the radio-frequency coil arrangement (49). [13] Provision and / or recording system (40) with a computing device (41) which is designed to carry out a method according to one of the preceding claims. [14] Computer program, comprising program means which, when the computer program is executed on a computing device (41), cause the computing device (41) to carry out the steps of a method according to one of claims 1 to 12. [15] An electronically readable data carrier on which a computer program according to claim 14 is stored.

Images