Computer-implemented method for operating a magnetic resonance device, magnetic resonance device, computer program and electronically readable data carrier

Abstract

The invention relates to a computer-implemented method for operating a magnetic resonance device (20) during a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections (1) of a magnetic resonance sequence, wherein each sequence section (1) comprises a preparation module (8) and a readout module (10), and each readout module (10) comprises several readout submodules (2) in which radio frequency pulses (12, 13) precede readout periods for recording magnetic resonance data, wherein in each readout module (10) at least one navigator submodule (11, 11a, 11b) is used for recording navigator data of the sequence section (1) along a k-space trajectory, wherein for at least one pair of navigator submodules (11, 11a, 11b) the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process.- for each sequence segment (1), correlation information is determined by comparing the navigator data set of the sequence segment (1) with at least one navigator data set of another sequence segment (1) and the same k-space trajectory, and - the correlation information is evaluated for the selection of sequence segments (1) whose magnetic resonance data are discarded, and / or for assigning a weight to the magnetic resonance data of at least a part of the sequence segments (1) before a reconstruction of a magnetic resonance image.

Description

The invention relates to a computer-implemented method for operating a magnetic resonance imaging (MRI) device during a recording process according to a recording protocol, which in at least one repetition comprises several sequence segments of a MRI sequence, wherein each sequence segment comprises a preparation module and a readout module, and each readout module comprises several readout submodules in which radio frequency pulses precede readout periods for recording MRI data. The invention further relates to a MRI device, a computer program, and an electronically readable data carrier. Magnetic resonance imaging (MRI) is now a widely used diagnostic and monitoring tool in medical applications. Due to the longer acquisition times of typical MRI data acquisition protocols, motion is a crucial factor in improving image quality. Even with correction, movement during the acquisition process can lead to a loss of image quality and thus compromise the usability of the MRI data and any reconstructed MRI images / image datasets. When acquiring a patient, motion encompasses both cyclical movements, such as breathing and heartbeat, within the living subject, as well as other intentional and unintentional movements of the imaging area, triggered both externally and internally. The avoidance or consideration of motion proves particularly relevant in TSE sequences (Turbo Spin Echo sequences, also known as RARE sequences or FSE sequences). In a TSE sequence, a high-frequency excitation pulse in a preparation module is followed by a so-called echo train in a readout module. Within this readout module, several readout submodules follow refocusing high-frequency pulses, each with its own readout period, all benefiting from the same excitation high-frequency pulse. This results in acquisition times of several minutes with multiple echo trains. Nevertheless, the TSE sequence is used as the "workhorse" of medical imaging due to its advantages.Movement is similarly relevant in other, similarly structured sequence types, for example in diffusion imaging with gradient echo (GRE) readout, in which a preparation module with diffusion gradients is followed by a readout module in which a readout period follows an excitation radio frequency pulse in the respective readout submodules of a readout module. To work with less magnetic resonance data and thus shorten acquisition time, trained reconstruction algorithms have been proposed that directly use the (especially undersampled) magnetic resonance data in k-space as input and output magnetic resonance images in image space, i.e., spatial space. An example of such functions are those that use unrolled neural network architectures. An overview of deep learning approaches in magnetic resonance reconstruction can be found, for example, in an article by Florian Knoll, “Deep-Learning Methods for Parallel Magnetic Resonance Imaging Reconstruction: A Survey of the Current Approaches, Trends, and Issues,” IEEE Signal Process Mag. 37 (2020), pages 128 to 140. Such trained reconstruction algorithms allow for a significant reduction in acquisition time with high acceleration factors and yet better image quality compared to conventional reconstruction methods such as GRAPPA (Generalized autocalibrating partially parallel acquisitions, see the article of the same name by Mark A. Griswald in Magn. Reson. Med. 47 (2002), pages 1202-1210). However, due to the noise-suppressing effect of the trained function, especially the included neural network, image artifacts caused by motion are reconstructed more visibly than with conventional reconstruction methods, where such artifacts are often lost in the noise. Fundamentally, however, these reconstruction approaches, such as GRAPPA, are also affected by motion effects. To reduce motion artifacts, various approaches have already been proposed in the prior art. One well-known example is SAMER (Scout accelerated motion estimation and reduction), see D. Polak et al., MRM 87 (2022), pages 163-178. In this approach, a reference dataset is acquired at the beginning of the acquisition process using a short reference scan. In each sequence segment, particularly echo train, several low-resolution navigator echoes are then acquired. Subsequently, an optimization procedure within the image reconstruction process fits the navigator echoes to the reference dataset in order to derive motion correction parameters. These parameters are then used to appropriately correct the magnetic resonance data of the respective readout submodules. Currently, this approach cannot be combined with deep learning reconstructions, i.e., the use of trained reconstruction functions, and therefore offers only limited acceleration potential.Furthermore, it cannot compensate for pulsating movement effects (pulsation effects). In the subsequently published German patent application DE 10 2024 203 342.1, it was proposed for a method mentioned above to use at least one navigator submodule in each readout module for acquiring navigator data of the sequence segment along a k-space trajectory that runs along the readout direction through the k-space center. For each sequence segment, correlation information is determined by comparing the navigator data set of the sequence segment with at least one navigator data set of another sequence segment with the same k-space trajectory. This information is then evaluated to select sequence segments whose magnetic resonance data are to be discarded and / or to assign a weight to the magnetic resonance data of at least some of the sequence segments before reconstructing a magnetic resonance image. In particular, the k=0 line is acquired in each sequence segment using a Cartesian acquisition scheme.Because the trained reconstruction algorithm (especially an unrolled neural network) can obtain a mask of the recorded k-space lines, individual echo trains or sequence segments can easily be omitted if they are affected by strong movement. However, it turns out that movements along the phase-encoding direction, in particular, lead to strong artifacts. Since the navigator data set recorded along the readout direction represents a projection onto the readout direction, movements essentially along the phase-encoding direction are only inadequately captured by this method, or in extreme cases, not captured at all. The publication Z. Hu et al.: Motion-compensated 3D turbo spin-echo for more robust MR intracranial vessel wall imaging. In: Magn Reson Med., 2021, August, Vol. 86 (2): p. 637-647 discloses a motion compensation method that combines volumetric navigators with self-gating. The publication US 10162037 B2 discloses a method for correcting magnetic resonance data using different navigator sequences. The invention is based on the objective of providing a method for obtaining magnetic resonance data improved with regard to motion artifacts by improving the detection of both pulsating and other movements. This problem is solved according to the invention by a computer-implemented method, a magnetic resonance device, a computer program, and an electronically readable data carrier according to the dependent claims. Advantageous embodiments are described in the sub-claims. In a method of the type mentioned at the outset, the invention provides that: - in each readout module at least one navigator submodule is used to record navigator data of the sequence segment along a k-space trajectory, wherein for at least one pair of navigator submodules the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process; - for each sequence segment, correlation information is determined by comparing the navigator data set of the sequence segment with at least one navigator data set of another sequence segment and the same k-space trajectory; and - the correlation information is evaluated for the selection of sequence segments whose magnetic resonance data are discarded and / or for assigning a weight to the magnetic resonance data of at least some of the sequence segments before a reconstruction of a magnetic resonance image. Two fundamental approaches are conceivable here. While it is possible, on the one hand, to weight the magnetic resonance data of the sequence segments as a whole, for example, depending on the magnitude of the movement described by the deviation in the comparison, another advantageous variant provides that if at least one deviation condition evaluating the correlation information is fulfilled for one of the sequence segments, the magnetic resonance data of at least that sequence segment is discarded or, in a subsequent reconstruction of a magnetic resonance image, weighted less heavily than the magnetic resonance data of sequence segments that do not fulfill the deviation condition themselves. In embodiments in which the k-space trajectory is changed with each new sequence segment (as will be explained later), the discarding and / or weighting can also apply to adjacent and / or intermediate sequence segments in which the respectiveThis applies if a different k-space trajectory was used. For example, if, in the case of five layers, the k-space trajectory is used in the phase-encoding direction in the first, third, and fifth layers, and the k-space trajectory is used in the readout direction in the second and fourth layers, and the deviation condition is met for the third and fifth layers but not otherwise, then the discard and / or weighting can also be performed for the layers acquired with motion between the layers, in this example, the fourth layer. Weighting may be applied with an interpolated strength. A weaker weighting or even discarding can also be applied to adjacent layers with a different k-space trajectory, here the second layer. However, since it is not known exactly when the motion begins or ends, the overall effect of this measure may be weakened.Naturally, a combination of weights can also be applied if the deviation condition for the other k-space trajectory is met in the intermediate or adjacent trajectory. These explanations can also be applied, with appropriate limitations, if the first and second space trajectories are switched after each repetition. A partial correction may also be conceivable in combination with refraction and / or weighting scenarios, which will be discussed in more detail later. The deviation condition, or more generally, the presence of a deviation described by the correlation information, indicates—due to the difference between the navigator dataset of an examined sequence segment and at least one navigator dataset of another sequence segment with the same acquisition trajectory—that the examined sequence segment could be affected by movement. In this context, movement in relation to a patient as the subject of the study can include periodic body movements such as breathing and heartbeat, but also other movements, whether intentional or unintentional, especially macroscopic ones. With regard to heartbeat, the term pulsation is often used in this context.The proposed method therefore concerns the improvement of the image quality of magnetic resonance images with regard to any movements in the recording area by identifying and discarding magnetic resonance data affected by movement, or at least weighting them less for reconstruction in order to avoid motion artifacts. Discarding data means that the magnetic resonance (MRI) data from the sequence segment is no longer included in subsequent reconstructions. This means that at least one MRI image is reconstructed from the remaining MRI data from the acquisition process, so that the number of motion-related artifacts can be significantly reduced due to the removal of motion-related MRI data. This can also be achieved with a lower weighting of such MRI data, whereby a certain sampling density can still be maintained, but the motion-related MRI data has less influence on the MRI image. For example, the previously mentioned amplification effect in trained reconstruction functions can be specifically counteracted.One advantage of the procedure described here when used after the recording process is that it can be easily implemented before reconstruction, since only a simple step to identify motion-related sequence sections and discard or set the weighting is required before the actual image reconstruction, but the reconstruction itself does not need to be modified. It can be particularly advantageous to use a reconstruction function, especially a trained one, that compensates for missing magnetic resonance data, and in particular a reconstruction function based on an unrolled neural network architecture, which takes magnetic resonance data in k-space as input data and provides at least one magnetic resonance image in image space (position space) as output data. The combination of identifying and discarding motion-laden magnetic resonance (MRI) data with trained reconstruction functions, particularly k-space-to-image-space-based unrolled neural networks (e.g., known as "Deep Resolve Boost"), is especially advantageous because these trained reconstruction functions are resilient to slight changes in the sampling pattern. This means that even if MRI data is missing due to discarding, high-quality reconstruction is still possible. In other words, the network architecture of the trained reconstruction function does not necessarily require regularly sampled MRI data as input. Therefore, with the method described here, motion-laden sequence segments, especially echo trains, can be identified and discarded before reconstruction. However, in a further advantageous development, it is also possible to incorporate weighting into the input data of the reconstruction function. For example, it can preferably be provided that the trained reconstruction function uses, in addition to the magnetic resonance data as input data, a sampling mask that describes the distribution of the sampled sample points in the sampled k-space, wherein: - for magnetic resonance data to be discarded, the corresponding sample points are marked as unsampled, and / or - for magnetic resonance data to be weighted, the weight is entered into the sampling mask at the corresponding sample points, wherein the reconstruction function is designed to use the weight during the reconstruction and / or to check the consistency between the reconstruction result and the magnetic resonance data. In this context, a binary sampling mask can be used, in which sampled points in k-space are assigned a one, and unsampled points a zero. For discarded magnetic resonance data, the corresponding portions of k-space in the mask would simply be marked with zero. However, with weighting, values ​​between zero and one can be introduced, which is particularly advantageous. This can, of course, also be applied to sampling masks that work with other base values ​​and / or even already use weighting. The sampling mask thus determines how much weight the magnetic resonance data should receive during reconstruction or data consistency checks. It can be particularly useful to apply weighting only or primarily during consistency checks. In this case, the magnetic resonance data contributes fully to the reconstruction, thus maintaining the sampling basis, but is given less weight during consistency checks, so that motion-based artifacts can still be significantly reduced. Of course, a combination with other reconstruction techniques, such as GRAPPA, is also conceivable. Here, discarded magnetic resonance data in k-space, for example, missing k-space lines, must be additionally reconstructed, which may require the calculation of additional reconstruction kernels, such as GRAPPA kernels. Implementing a weighting mechanism is also conceivable. In general, it can be said that the weighting, particularly of the less weighted magnetic resonance data, is chosen depending on the magnitude of the deviation of the navigator data of one sequence segment from that of at least one other sequence segment, especially the degree of violation of the deviation condition (possibly with respect to a specific direction), and / or depending on their position in the sampled k-space. In exemplary embodiments, the magnitude of the deviation can be described directly by the correlation information, for example, a correlation value.With regard to the first aspect of the position in k-space, it should be noted that it is also generally conceivable to choose different weightings within a sequence segment. For example, a higher weighting could be assigned to magnetic resonance data relating to sampling points located closer to the k-space center than to magnetic resonance data relating to sampling points located further away from the k-space center. The weighting could, for instance, be applied k-space line-wise (or k-space trajectory segment-wise). It can therefore be said that, advantageously, central regions of the sampled k-space should contribute more than the edge of the sampled k-space, since the main part of the signal is located in the central regions. Additionally or alternatively, the weighting can depend on the severity of the deviation, particularly the violation of the deviation condition, which describes the magnitude of the relative detected motion. In the sampling mask example discussed above, which uses values ​​from zero to one, the weighting can be close to zero for strong motion and close to one for only slight motion. For example, for sequence segments that satisfy the deviation condition, a weighting that decreases linearly or otherwise functionally with the severity of the deviation condition violation can be chosen.When switching the k-space trajectory between sequence segments and extending the weighting to adjacent and / or intervening sequence segments of other k-space trajectories, the weighting can be interpolated or extrapolated, for example, linearly. For sequence segments that do not meet the deviation condition but for which interpolation or extrapolation is to be performed, the correlation information or a weight of one can still be considered. If the correlation information allows for a corresponding breakdown, the deviation condition and / or the weighting can also be evaluated or selected with respect to at least one breakdown direction.For example, in scenarios where movements in the phase-encoding direction can result in stronger artifacts, these movements can be weighted more heavily, or, if at least a partial correction is possible in one decoding direction, weighting can be applied to the remaining decoding direction(s). However, if no deviation condition is to be used, it is also conceivable that the magnitude of the deviation is determined for each sequence segment to determine the weighting, with the weighting decreasing, particularly linearly, with the magnitude of the deviation. In other words, all sequence segments can be sorted according to the magnitude of the deviation, particularly the movement intensity, and a weighting decreasing according to movement intensity, particularly linearly, can be assigned to all sequence segments. The explanations regarding adjacent and intervening sequence segments during k-spatial trajectory changes apply accordingly. Here, too, the assignment can optionally be made with reference to a breakdown direction, particularly a phase-encoding direction. A further general advantage of the method according to the invention is that even short-term influences on the timescale of a single sequence segment, for example echo train, can be identified and taken into account for a single layer. In particular, it is therefore not necessary to discard or give less weight to an entire repetition, i.e., a TR period, if a motion effect only extends over a few hundred milliseconds of a sequence segment. This method is particularly advantageous for turbo spin echo (TSE) sequences. The magnetic resonance sequence can therefore be configured as a turbo spin echo sequence, with the radio frequency pulses of the readout submodules serving as refocusing pulses. In TSE sequences, an excitation radio frequency pulse is output in the preparation module, which is then used in the echo train within the respective TSE submodules for multiple readout periods. A refocusing radio frequency pulse is output before each readout period. Specifically, it is proposed to identify sequence segments containing motion based on an additional navigator echo per sequence segment. Thus, navigator data is recorded in at least one corresponding navigator submodule, so that at least one navigator data record exists for each navigator submodule and therefore for each sequence segment. Each navigator data record is recorded using a specific k-space trajectory. It can be correlated with other navigator data records that used the same k-space trajectory, i.e., compared to determine whether there is a deviation caused by motion. For example, the deviation condition can check whether there is a significant deviation that warrants separate treatment. To allow for good comparability of the navigator data sets, a convenient further development of the invention provides that the navigator data for all sequence segments are recorded along at least one identical k-space trajectory and / or that the at least one k-space trajectory includes the k-space center. Recording a k-space trajectory, in particular a k-space line, at the k-space center has the advantage that most of the signal is present there, resulting in less noise and thus further improving the comparability. Naturally, to improve comparability, the same high-frequency pulse is also advantageously used in all navigator submodules with the same k-space trajectory. While the procedure can generally be implemented without changing the sequence section timing, especially the echo train timing, the movement along at least one phase encoding direction can now also be easily detected. It should be noted that, for the sake of simplicity, most of the implementation examples discussed below refer to layer-by-layer acquisition and the single phase-encoding direction used there. Typically, a readout direction, a phase-encoding direction, and a layer selection direction are employed. Three-dimensional scanning often uses two phase-encoding directions, which, ideally, can be covered by at least some of the navigator submodules. It is expediently provided that in each navigator submodule where the k-space trajectory, which at least partially follows the phase-encoding direction, is used due to a corresponding k-space trajectory gradient pulse, a prephasing pulse and / or a rephasing pulse associated with the k-space trajectory gradient pulse are output in the navigator submodule. The additional prephasing pulse can be output either before a high-frequency pulse of the navigator submodule or (in this case, with a different sign) immediately before the k-space trajectory gradient pulse. The rephasing pulse is output after the navigator data has been read out. This ensures that the gradient moment of the k-space trajectory gradient pulse is balanced. This is necessary to maintain the CPMG conditions for the other echo / readout submodules. In exemplary embodiments, the navigator data set is acquired at least partially perpendicular to the k-space lines of the magnetic resonance data in the readout direction. However, when using k-space trajectories that run (purely) in the phase-encoding direction, which may be easy to implement, movement along the readout direction is no longer easily detectable—conversely to the problem described above. Even for "mixed directions" (for example, k-space trajectories at a 45° angle to both the readout and phase-encoding directions), a perpendicular direction always results, which is more difficult to detect. Preferred embodiments therefore provide for the use of at least two different k-space trajectories, for example, a k-space line in the phase-encoding direction and a k-space line in the readout direction, preferably each passing through the k-space center. Various approaches are conceivable in this regard. It can therefore be generally stated that preferably a first k-space trajectory, particularly one running in the phase-encoding direction, and a second k-space trajectory, particularly one running in a readout direction of the recording process, are used. These k-space trajectories are at an angle to each other in the plane formed by the phase-encoding direction and the readout direction, and in particular, are perpendicular to each other. Specifically, these are k-space lines running through the k-space center. In a three-dimensional case, the concept described here can be extended analogously to a third k-space trajectory in the second phase-encoding direction or slab direction. Preferably, the first and second k-space trajectories are alternated across the navigation submodules according to a regular usage pattern. This ensures that both directions are regularly scanned for movements. For example, each readout module can comprise (exactly) one navigator submodule, and the first and second k-space trajectories can be changed with each new sequence segment. In a second variant of this embodiment, it is also conceivable that the first and second space trajectories are changed after each repetition. Regarding the weighting and discarding of adjacent or intervening sequence segments or repetitions, reference is made to the explanations already given above. In a first approach, the navigator orientation is rotated between echo trains. In multi-layer imaging, this rotation can occur, for example, between sequence segments of the N layers being acquired. The advantage is that the direction of the navigator, specifically the k-spatial trajectory, changes after each time interval equal to the repetition time divided by the number of layers N being acquired. This allows for the capture of rapid movements along both orientations, particularly the phase encoding direction and the readout direction, since this time interval can range from approximately 50 to 150 ms. Alternatively, in the second approach, the navigator orientation can change after each repetition, meaning after each echo train has been acquired or a sequence segment has been performed for each layer. However, it is particularly advantageous to provide that each readout module comprises two navigator submodules, one of which uses the first and one of which uses the second k-space trajectory. In this particularly preferred embodiment, at least two navigator data sets with k-space trajectories rotated relative to each other are acquired within a single sequence segment. This is particularly advantageous because it enables differentiation of the directions of motion with very high temporal resolution. The navigator submodules can follow each other directly within the readout module, but can also be separated by readout submodules. The acquisition time may be extended in some embodiments of the inventive procedure, for example, due to additional pre-phasing and / or re-phasing pulses and / or the use of multiple navigator submodules with different k-spatial trajectories per sequence segment. Therefore, a further advantageous embodiment of the invention may provide that at least one k-spatial trajectory of the navigator submodule is selected to be shorter than the acquisition trajectory of the readout submodules. This does not impair the function of the navigator, since a lower spatial resolution is sufficient for the navigator data set. On the other hand, this saves time, thus eliminating the need for additional time. Therefore, an increase in acquisition time due to a reduced resolution of the navigators can be compensated for. Exemplary implementations can provide that the navigator submodule corresponds to a readout submodule for the k-space trajectory to be sampled for the navigator data. For example, in the case of a TSE echo train, an additional TSE echo for the k-space trajectory can be acquired in the navigator submodule. However, it is also conceivable, particularly with regard to compensating for an extension of the acquisition process, to use a different echo type or sequence type for the navigator submodule, for example, measuring a gradient echo in the case of a TSE sequence. The latter case can also offer advantages with a suitable arrangement of the navigator submodule in the readout module, which will be discussed in more detail below. Preferably, in a first specific embodiment, the navigator data can be acquired at a fixed position relative to the readout submodules, in particular before or after all readout submodules. In this way, all navigator data sets are acquired under comparable conditions, which in turn improves comparability and simplifies the determination of correlation information and, if necessary, the formulation of the deviation condition and / or a correctability condition. For example, an additional echo, i.e., the navigator data, can be acquired at the beginning of each readout module in the navigator submodule, particularly in the region of the k-space center. However, the navigator submodule can also be acquired at other positions in the sequence segment, for example, at its end. A practical implementation in a turbo spin echo sequence (TSE) as a magnetic resonance sequence involves either the navigator submodule being a gradient echo submodule following all readout submodules, with the radio frequency pulse (in this case, an excitation pulse) of the navigator submodule being output with a reduced flip angle compared to the radio frequency pulses of the readout submodules and / or at a time interval from the preceding radio frequency pulse (i.e., refocusing pulse) of the last readout submodule that is shorter than the time interval between the radio frequency pulses of the readout submodules. This approach particularly benefits from the fact that a gradient echo is not bound to the time interval of the refocusing pulses specified in the TSE sequence. Thus, the additional echo can be introduced with only a slight increase in acquisition time.This is particularly advantageous when an extension, for example by at least two navigator submodules per sequence segment, needs to be at least partially compensated for. Alternatively, the navigator submodule could be a turbo spin echo submodule where the refocusing pulse has a smaller flip angle. In both cases, reducing the flip angle can reduce the SAR load. In an alternative, second concrete embodiment, for the at least partially time-neutral implementation of the acquisition of the navigator data sets, it can be provided that the navigator submodules of the same k-space trajectory cover all possible positions in the readout module and have at least partially different positions within the readout module, wherein the navigator data are additionally evaluated to determine phase evolution information about the readout module, which is used for an eddy current effect correction of the magnetic resonance data.In the prior art, it is known to provide an eddy current correction sequence section, particularly an eddy current correction echo section, prior to the first sequence segment used for acquiring magnetic resonance data, in particular the echo section in TSE imaging. In this sequence segment, an echo of only the k-space center is acquired at each temporal position, and the phase evolution between the different echoes is determined to correct for eddy current effects. The determined correction is then applied in the subsequent sequence segments for acquiring magnetic resonance data. This second embodiment of the present invention proposes to eliminate the eddy current correction sequence section and instead acquire navigator data of the k-space center for at least one of the at least one k-space trajectory at a different temporal position in each sequence segment.The navigator data can then be used both as before for correcting eddy currents between the respective temporal positions and for identifying sequence segments involving motion. In this second embodiment, if magnetic resonance data from an echo train are discarded, it is advantageous not to use the navigator data for eddy current correction either. It can then be provided that, for sequence segments whose correlation information meets the deviation condition, the phase evolution information is interpolated and / or extrapolated, omitting the navigator data for these sequence segments. This also avoids the influence of motion on eddy current effects. Correlation information can be determined, at least in part, as autocorrelation, particularly in image space. However, other correlation values, especially correlation measures, can also be used as correlation information, and other comparison methods can be employed. To determine the correlation, navigator data with the same k-space trajectory can be conveniently transformed into image space using a Fourier transform. Significantly deviating sequence segments can then be marked for discard, or appropriate weightings can be assigned. In principle, it is conceivable to use a single sequence segment as a reference sequence segment and to check the correlation of an examined sequence segment to the reference sequence segment. For example, the deviation condition can then check whether a correlation value describing the strength of the correlation or deviation (e.g., as a correlation measure) to the navigator data of the reference sequence segment, which is included in the correlation information, falls below a threshold. However, it is preferable to analyze the pairwise correlations of all sequence segments (possibly those traversed so far or covered by the same k-spatial trajectories) with each other, so that an entire recording can be avoided because the reference sequence segment was affected by movement. A preferred embodiment of the invention provides that the correlation information includes at least one correlation value, in particular a mean value for the sequence segment, wherein a reference value is determined as the mean of the correlation values ​​over all sequence segments and the deviation condition checks whether the correlation value of the sequence segment falls below a threshold value dependent on the reference value, in particular 50 to 90% of the reference value. In other words, it may be possible to check whether at least one correlation value deviates from the reference value (mean) by a threshold, particularly a percentage threshold. It should be noted that here, too, only one reference sequence segment can be used, to which the correlation value then refers. However, it is preferable to refer to all sequence segments (possibly recorded so far or covered by identical k-spatial trajectories), so that the correlation information can include an average pairwise correlation value for all these sequence segments. In this way, the sequence segments that represent outliers are easily filtered out by the deviation condition, while the (usually) larger mass of similar navigator data indicates sequence segments that are less affected by motion and can be retained for the reconstruction of a magnetic resonance image, especially with maximum weighting. Particularly when k-space trajectories are used for different directions, for example, the phase-encoding direction and the readout direction as k-space lines, it may be possible to determine the correlation information at least partially broken down according to the readout direction and the phase-encoding direction as breakdown directions. In other words, comparisons of navigator data sets acquired along a first k-space trajectory in the readout direction show motion components along the readout direction, and comparisons of navigator data sets acquired along a second k-space trajectory in the phase-encoding direction show motion components along the readout direction. Within the scope of the invention, there are embodiments in which a breakdown of the correlation information of this kind is useful.For example, it is conceivable to weight deviations in one resolution direction differently than in another. If, for instance, it is known that movements in the phase-encoding direction lead to more disruptive artifacts than in the readout direction, thresholds for discarding the magnetic resonance data can be chosen differently, or weights can scale differently with the magnitude of the deviation. In other words, different deviation conditions can also be chosen for different resolution directions where appropriate. It is also conceivable, in principle, to combine the rejection and / or weighting approach with a partial correction approach. For example, it could be provided that, if a correctability condition is met that evaluates the correlation information for a first of the resolution directions, the magnetic resonance data of one of the sequence segments, and at least if there is a greater deviation in the first resolution direction, particularly the readout direction, than in the other resolution direction, particularly the phase encoding direction, are corrected in the first resolution direction depending on the correlation information for the first resolution direction, in particular by a shift in image space and / or a phase ramp in k-space.For example, if the magnetic resonance data processing allows for easy correction of movements in the readout direction, the portion of the correlation information related to this direction can be used to correct the magnetic resonance data of a corresponding sequence segment, while the portion of the correlation information related to the phase-encoding direction can be evaluated for rejection or weighting. If the correctability condition is not met despite the presence of a deviation, for example, because the deviation is too large, rejection or reduced weighting with respect to the readout direction can, of course, also be performed. By incorporating corrections in certain cases, further improvements in image quality can be achieved. In an advantageous embodiment, an initial assessment can be made to determine whether motion along the phase-encoding direction or along the readout direction predominates. Sequence segments with strong motion along the phase-encoding direction are then discarded or given a low weighting. Remaining sequence segments where rigid motion along the readout direction is detected (specifically, for example, a translation of the navigator data along the readout direction) can be corrected, for example, by a geometric shift in image space or a phase ramp in k-space to improve image quality. A further advantageous embodiment of the invention may provide that a maximum number of magnetic resonance data sets from individual sequence segments are used to be discarded. If this number is exceeded, either the entire acquisition process is deemed invalid, or the deviation condition, in particular a threshold value, is adjusted to comply with the maximum number, and / or sequence segments whose magnetic resonance data must be reintroduced for reconstruction despite fulfilling the deviation condition are selected by means of at least one selection criterion to comply with the maximum number. The maximum number may depend on the specific acquisition process, for example, the acceleration measures, and the possibilities for compensating for missing magnetic resonance data during the reconstruction of magnetic resonance images, in particular a trained reconstruction function used.For example, the maximum number of sequence segments can be chosen depending on an acceleration factor and / or the number of sequence segments. If this limit is exceeded, meaning more sequence segments than the maximum number meet the deviation condition, the entire magnetic resonance data from the acquisition process can be discarded, and a reacquisition may be recommended. However, it is also possible, especially after confirmation by a user, to still attempt reconstruction. In this case, the deviation condition can be adjusted, for example, by increasing or decreasing a threshold value, or at least one selection criterion can be used to select magnetic resonance data that would otherwise have been discarded but should nevertheless be reintroduced. At least one of these selection criteria can address the severity of the deviation condition violation.However, it is particularly advantageous if at least one of the selection criteria relates to the coverage of the k-space to be sampled, especially the distribution of the k-space points at which measurements were taken. In a specific example, the resulting maximum distance between k-space lines of the sequence segment, particularly in the echo train, can serve as a selection criterion. That is, for echo trains with similar deviations according to the correlation information, where discarding one echo train results in a maximum undersampling of four, while discarding the other results in an undersampling of six, the echo train with an undersampling of four is chosen for retention or reintroduction. In this context, it should also be noted that other, generally known techniques for avoiding excessive (local) undersampling can be synergistically combined with the inventive method. An example of this is the method known as "Reduce Motion Sensitivity." Here, it can be provided that k-space lines (or other k-space trajectory segments) are not fixedly assigned to temporal positions of readout submodules within the readout module, but rather that the k-space lines or k-space trajectory segments to be acquired are randomly or pseudo-randomly distributed among the readout submodules within the readout module. In this way, the probability of large k-space gaps in regular undersampling can be minimized. In an advantageous group of embodiments, the determination of correlation information and the verification of the deviation condition can be performed at least partially during the acquisition process, particularly immediately after the acquisition of the navigator data and / or completion of the sequence segment. The magnetic resonance data to be discarded from a sequence segment for which the deviation condition is met are at least partially reacquired in a subsequent, appropriately adapted sequence segment. In addition to or as an alternative to an assessment after the acquisition process, particularly before reconstruction, for determining weights and / or discardable magnetic resonance data, it can therefore also be provided that the identification of discardable sequence segments occurs dynamically during the acquisition process. This makes it possible to modify the acquisition process to compensate for the loss and / or mitigate its consequences.In particular, it may be planned that the sequence segment will be repeated, especially immediately or at the end of the planned sequence. The planning of the recording protocol can also be geared towards checking for motion during recording. In this context, a suitable advanced training approach stipulates that, given a fixed or maximum specified number of sequence segments, their order within the acquisition protocol is determined such that with each sequence segment, the distance between sampled k-space trajectory segments, particularly k-space lines, is minimized. If, for example, the current sequence segment, especially the current echo path, is identified as discardable using the deviation condition, the subsequent sequence segment is adjusted to resample the k-space positions of the previously discarded sequence segment. Even if this ultimately results in the loss of sequence segments, it ensures that k-space is sampled with the smallest possible distances, thus enabling more robust and higher-quality compensation or estimation of missing magnetic resonance data during reconstruction. In this context, it may also be advantageous to choose a maximum number of possible sequence segments that is greater than the number of sequence segments required for motion-free sampling. For example, when the functionality described here is activated, time can be reserved for reserve sequence segments to allow repetitions within certain limits without dropping any sequence segments. Furthermore, it is conceivable to slightly reduce the planned undersampling to avoid negative effects from excessively large sampling gaps. In summary, it can be said that by identifying and discarding or weighting defective sequence segments, particularly in combination with deep learning reconstruction methods, improved image quality and greater insensitivity to motion, especially in the phase-encoding direction, are achieved. Compared to motion correction approaches such as SAMER, the invention has the advantage that it can be combined with the use of trained reconstruction functions and can also take pulsation effects into account. In addition to the method, the invention also relates to a magnetic resonance device comprising a main magnet unit with a main magnet for generating a main magnetic field, a gradient coil arrangement, a high-frequency coil arrangement, and a control device, which includes: - a sequence unit for controlling a recording process in a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections of a magnetic resonance sequence, wherein each sequence section comprises a preparation module and a readout module, and each readout module comprises several readout submodules in which high-frequency pulses precede readout periods for recording magnetic resonance data, wherein the sequence unit is configured to use at least one navigator submodule in each readout module for recording navigator data of the sequence section along a k-space trajectory.wherein for at least one pair of navigator submodules the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the acquisition process, - a correlation unit for determining correlation information for each sequence segment by comparing the navigator data set of the sequence segment with at least one navigator data set of another sequence segment and the same k-space trajectory, and - an evaluation unit for evaluating the correlation information to select sequence segments whose magnetic resonance data are to be discarded, and / or to assign a weight to the magnetic resonance data of at least some of the sequence segments. In particular, the evaluation unit can be designed, for example, to discard and / or reduce the weight of magnetic resonance data of at least one sequence segment for which at least one deviation condition evaluating the correlation information is met. All statements relating to the method according to the invention can be applied analogously to the magnetic resonance device according to the invention and vice versa, so that the same advantages can be obtained. The control unit can comprise at least one processor and at least one storage medium. Hardware and / or software are used to form functional units for carrying out steps of the method according to the invention; in this case, at least one sequence unit, one correlation unit, and one evaluation unit are formed. A reconstruction unit is also known in principle for control units of magnetic resonance imaging (MRI) devices and is therefore also sensibly provided according to the invention. Further functional units can, of course, also be provided, particularly with regard to the various proposed configurations. For example, an adjustment unit for adapting the recording protocol during evaluation during the recording process can be provided. The sequence unit corresponds in principle to known sequence units for control units of MRI devices that control the recording operation. A computer program according to the invention can be directly loaded into a storage medium of a control unit of a magnetic resonance device and comprises program elements such that, when the computer program is executed in the control unit, the latter is caused to carry out 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 contains 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 control unit of a magnetic resonance device, the latter is configured to carry out a method according to the invention. The data carrier is, in particular, a non-transient data carrier, for example, a CD-ROM. Further advantages and details of the present invention will become apparent from the exemplary embodiments described below and from the drawings. Figure 1 shows a flowchart of a first embodiment of the method according to the invention, Figure 2 shows an exemplary section of a sequence diagram of a TSE sequence in a first variant, Figure 3 shows an exemplary section of a sequence diagram of a TSE sequence in a second variant, Figure 4 shows a flowchart of a second embodiment of the method according to the invention, Figure 5 shows a schematic scanning scheme in the second embodiment, Figure 6 shows a schematic diagram of a magnetic resonance device according to the invention, and Figure 7 shows the functional setup of the magnetic resonance device of Figure 6. The following is a concrete example of the application of the inventive method to an acquisition process using an acquisition protocol that employs a TSE sequence. Several layers, successive in a layer selection direction, are acquired in multiple repetitions. In each repetition, echo segments of a specific length from readout submodules, here TSE submodules, are used as sequence segments to acquire magnetic resonance data. Specifically, a k-space line is acquired in each readout period following a refocusing radio frequency pulse. The k-space lines of the readout submodules are oriented in a readout direction perpendicular to a phase-encoding direction. This example, as well as the use of a TSE sequence, is purely illustrative. Fig. 1 shows a flowchart of a first embodiment of the method according to the invention. In step S1, magnetic resonance data are first acquired in the respective sequence segments, i.e., echo trains. In addition to the readout submodules, at least one navigator submodule is also used in each sequence segment to acquire navigator data. In the present embodiments, the navigator submodule always scans one of two predefined k-space trajectories through the k-space center. The first of the two k-space trajectories is a k-space line in the phase-encoding direction, and the second of the two k-space trajectories is a k-space line in the readout direction. The remaining readout submodules use different encodings, i.e., different k-space trajectory segments, in the sequence segments. Fig. 2 shows a variant of a first embodiment. For simplification, only two consecutive sequence sections 1 are illustrated, of which only a number of three readout submodules 2 are shown; in practical application, this number will be higher. Furthermore, only the radio frequency activity in an uppermost graph 3, the phase-encoding gradient pulses 4 in a middle graph 5, and the readout gradient pulses 6 in a lowermost graph 7 are indicated. Each sequence section 1 obviously comprises a preparation module 8, symbolized here by an excitation radio frequency pulse 9, and a readout module 10, which is subdivided into readout submodules 2 and a navigator submodule 11. Each readout submodule 2 and the navigator submodule 11 include radio frequency pulses 12 and 13 preceding the readout gradient pulses 6 and 6a, which define the respective readout periods. The radio frequency pulses 12 for the readout submodules 2, which are configured as TSE submodules, serve for refocusing. The navigator submodule 11 can also be configured as a TSE submodule, in which case the flip angle of the radio frequency pulse 13, which also serves for refocusing, is chosen to be smaller.However, it is also conceivable to select the navigator submodule 11 according to a different sequence type, for example as a gradient echo submodule, in which the high-frequency pulse 13 can serve to excite the gradient echo and can be closer to the high-frequency pulse 12 of the preceding readout submodule 2 than the intervals between the high-frequency pulses 12 themselves. It is evident that two different types of navigator submodules 11 are alternated in successive sequence sections 1. In the upper sequence section 1, labeled "A", the absence of phase-encoding gradient pulses 4 for the navigator submodule 11 when using a readout gradient pulse 6a indicates the measurement of a k-space line in the readout direction through the k-space center. Thus, the second k-space trajectory is used. In the lower sequence section 1, labeled "B", no gradient pulses are used in the readout direction during the readout period. Instead, a k-space line is read out in the phase-coding direction using a phase-encoding gradient pulse 4a, which defines the readout period. This means the first k-space trajectory, perpendicular to the second k-space trajectory, is used. The phase-encoding gradient pulse 4a is flanked by a pre-phasing pulse 4b and a re-phasing pulse 4c. The pre-phasing pulse 4b is output before the high-frequency pulse 13 of the navigator sub-module 11, since a readout gradient pulse 6b is already present during that period, thus allowing for temporal optimization. In any case, for both types of navigator submodule 11, the navigator echoes 14 are recorded as navigator data, so that a navigator data record exists for each sequence segment 1. In the embodiment shown in Fig. 2, the temporal position of the navigator submodule 11 in the readout module 10 is fixed at the beginning of the readout module 10. In principle, other fixed positions are also conceivable in the first embodiment, for example, at the beginning of the readout module 10. Fig. 3 schematically illustrates a second, preferred embodiment in which two navigator submodules 11a, 11b are used at the end of the respective readout module 10. The first navigator submodule 11a, recognizable by the phase-encoding gradient pulses 4a, 4b, and 4c, uses the first k-space trajectory, i.e., the k-space line in the phase-encoding direction, to record a first navigator data set assigned to this k-space trajectory. The second navigator submodule uses the second k-space trajectory, see readout gradient pulse 6a, i.e., the k-space line in the readout direction. A second navigator data set is recorded, assigned to this k-space trajectory. Since two navigator echoes are recorded, the duration of sequence segment 1 would, in principle, be extended.To at least partially compensate for this, a reduced resolution of the navigators is provided (by reducing the length of the first and second k-space trajectories compared to the k-space line segments of the readout submodules 2). Furthermore, gradient echoes with a reduced flip angle are interwoven, which further reduces the time required for the navigators. It should be noted that these time-saving measures can, of course, also be advantageously used in other embodiments. It is also conceivable to implement embodiments in which the temporal position of the at least one navigator submodule 11, 11a, 11b is not fixed but changes, such that all possible temporal positions in the readout module 10 are traversed at least by the navigator submodules 11, 11a, 11b that use the first k-space trajectory in the readout direction, but preferably by both navigator submodules 11. The corresponding navigator data sets can then also be used for eddy current correction after the phase evolution for the different temporal positions can be traced from the navigator data. This eliminates the need for a separate sequence section for the eddy current correction navigators. If, as will be explained later, only navigator data affected by motion are available for a temporal position in the readout module 10, it is possible to omit this data and perform the eddy current correction, the phase evolution, or the eddy current correction, respectively.to interpolate or extrapolate the missing navigator data. In some cases, however, redundant acquisition of navigator data for at least some of the temporal positions is possible anyway, namely whenever the number of sequence segments is greater than the echo train length. Regardless of the specific embodiment used, in step S2 of Fig. 1, after completion of the recording process, i.e., when navigator data is available for all sequence segments 1, correlation information is determined for each sequence segment 1. For this purpose, the pairwise autocorrelation of the navigator data with respect to all other sequence segments 1 in which the same k-space trajectory was used is determined for each sequence segment 1. In the preferred second embodiment according to Fig. 3, navigator data sets for the first and second k-space trajectories are available for each sequence segment, which can then be compared to determine motion in both directions with sequence segment accuracy. The navigator data of the same recording trajectory are thus compared with each other. An average correlation value is determined by statistical averaging of the individual results.Within this framework, a reference value can also be determined as the mean of all mean correlation values. In this case, the correlation value is calculated such that higher values ​​indicate a higher correlation. In step S3, the correlation information for each sequence segment 1 is evaluated using a deviation condition. In this case, the deviation condition checks whether the mean correlation value of the correlation information for the respective sequence segment 1 falls below a threshold derived as a percentage of the reference value, for example, between 50% and 90% of the reference value. It is conceivable to use different deviation conditions for different resolution directions, here the readout direction and the phase encoding direction, if movements in one of these directions could lead to stronger artifacts. If the deviation condition is met, meaning the navigator data differ significantly from the navigator data of most other sequence segments 1 with the same k-space trajectory, a motion during sequence segment 1 is inferred, which could also be a pulsation. Since the magnetic resonance data acquired in the readout submodules 2 of sequence segment 1 may also be affected by this motion, they are discarded and / or assigned a lower weight for subsequent reconstruction than magnetic resonance data for sequence segments 1 for which the deviation condition is not met. The weights are appropriately chosen depending on the position in k-space and the strength of the deviation condition violation, i.e., the motion strength. The choice of weight can also depend on the reconstruction direction.For example, the sequence segments 1 for which the deviation condition is met can be sorted according to the severity of the injury and assigned lower weighting with increasing injury severity, possibly modified with the k-space position. When using the first embodiment according to Fig. 2, rejection and / or weighting can also be applied to adjacent and / or intermediate sequence segments 1 in which a different k-space trajectory was used. For example, if a sequence segment 1 lies between two sequence segments 1 in which strong motion in the phase-encoding direction was detected, leading to their rejection, it can be assumed that this motion is also present in the intermediate sequence segment 1, which can then also be rejected. If the two sequence segments 1 with motion are merely weighted, the weighting of the magnetic resonance data of the intermediate sequence segment can, for example, be interpolated between these weights.Weighting can also be applied to adjacent sequence segments 1 (those where the deviation condition is not met), for example, in a predefined, weakened form or determined again by interpolation / extrapolation, for example, depending on the correlation information and / or relative to a weight of one. These weightings of intermediate sequence segments 1 can be cumulative for both the phase-encoding direction and the readout direction. This means that if a weight is determined for a sequence segment 1 based on movement in the phase-encoding direction and a weight is determined based on movement in the readout direction, both can be used, for example, by multiplying the weights that lie within an interval from zero to one. The remaining or weighted magnetic resonance (MRI) data are then used in step S4 to reconstruct at least one MRI image. For this purpose, a trained reconstruction function is used, which takes the MRI data in k-space as input and derives an MRI image in image space (i.e., position space) as output. Preferably, the trained reconstruction function is based on an unrolled neural network architecture. Such deep learning-based reconstruction functions can compensate for the loss of MRI data particularly well and robustly, so that despite the loss of motion-related MRI data according to step S3, high-quality MRI images are reconstructed that are unaffected or less affected by motion artifacts due to the absence of motion artifacts. If weighting is used, the trained reconstruction function can take a sampling mask of the sampled k-space as additional input data. For less weighted magnetic resonance data, the weights between zero and one are entered at the corresponding sampling points in k-space instead of one (sampled: full weighting) or zero (unsampled: not used). Weighting can be applied during reconstruction and / or to verify the consistency between the reconstruction result and the magnetic resonance data, particularly preferably at least in the latter case. It should be noted here that a maximum number of sequence segments whose magnetic resonance data may be discarded while still allowing for reliable reconstruction can also be specified. This maximum number can depend on properties of the trained reconstruction function as well as properties of the acquisition protocol, such as the acceleration factor and the number of sequence segments. If the maximum number is exceeded in step S3, the entire acquisition process can be marked as invalid due to movement, and a corresponding message can be issued to the user. If reconstruction is still desired, several approaches are conceivable. One option is to adjust the deviation condition, in this example specifically the threshold value, to avoid exceeding the maximum number.However, it is also possible to use selection criteria to reintroduce magnetic resonance data of certain sequence segments 1, whereby, for example, in addition to the strength of the violation of the deviation condition, an increase in the (possibly local) undersampling due to the omission of the magnetic resonance data can be used as a selection criterion. In step S3, a deviation condition can also be omitted, and all sequence segments 1, and thus magnetic resonance data, can be assigned a weight at least equal to the magnitude of the deviation (and thus the motion intensity) in the comparison, in particular described by the mean correlation value. For example, the sequence segments 1 can be sorted according to motion intensity, and a weighting of all sequence segments 1 can be applied that decreases linearly according to motion intensity. Fig. 4 shows a flowchart of a second embodiment of the method according to the invention. Again, in step S1, magnetic resonance data and navigator data are acquired according to the acquisition protocol. However, during the acquisition process, in step S2', the correlation information regarding previously acquired navigator data sets is determined, and in step S3', it is checked whether the deviation condition is met. If the deviation condition is met for a current sequence segment 1, the acquisition protocol can be adjusted in step S5 to at least partially reacquire the discarded magnetic resonance data of sequence segment 1, in particular to repeat sequence segment 1.It is conceivable to include time for reserve sequence segments in the acquisition protocol so that, despite discarded magnetic resonance data, complete coverage of k-space can be achieved as planned in just a few sequence segments. However, it is also advisable, and especially additionally, to plan the acquisition protocol from the outset to enable the best possible coverage with minimal gaps in the k-space being acquired. In this regard, the sequence of sequence segments can be chosen in such a way that the maximum distance between sampled points in k-space is minimized with each sequence segment. This will be explained in more detail with reference to Fig. 5, which schematically shows a section 15 of the k-space to be scanned. k-space lines 16, shown with dashed lines, are not to be scanned; however, the scanning of k-space lines 17, shown with solid lines, is planned. The numbers in parentheses to the right of the k-space lines 16 and 17 indicate the sequence section 1 for their scanning. It is evident that in the first sequence section 1, marked (1), a large distance 18 is created in k-space, which is reduced as much as possible by the acquisition in the second sequence section 1, marked (2), compare distance 19. This is also planned for the third sequence section 1, marked (3), which, however, is discarded in step S3' due to the fulfillment of the usability condition (and is therefore crossed out).Therefore, in step S5, the acquisition protocol is adjusted so that in the fourth sequence section 1, marked (4), the k-space lines 17 of sequence section 1, whose magnetic resonance data were discarded, are sampled again. This minimizes the gap between (1) and (2) as quickly as possible. The applicable statements regarding the first embodiment naturally also apply to the second embodiment. A combination of the embodiments is also conceivable, whereby, starting from the second embodiment, correlation information is determined again for all sequence segments after the recording process and evaluated for weighting, possibly using a different or modified deviation condition, or independently of such a condition. Finally, it should be noted that the functionality described here—namely, the identification of motion-laden sequence segments 1 and the discarding or weighting of their magnetic resonance data using navigator data in conjunction with a trained reconstruction function—can be activated on the magnetic resonance imaging (MRI) system being used. Activating this functionality then also allows for the inclusion of the aforementioned reserve sequence segments. Furthermore, it may be advantageous to slightly reduce the undersampling to counteract negative effects caused by excessively large sampling gaps resulting from discarded MRI data. Fig. 6 shows a schematic diagram of a magnetic resonance imaging (MRI) device 20 according to the invention. The MRI device 20 comprises a main magnet unit 21, which includes a superconducting main magnet (not shown in detail) and a cylindrical patient receptacle 22 into which a patient can be moved for an imaging procedure using a patient table (not shown in detail). A gradient coil arrangement 23 and a high-frequency coil arrangement 24 are provided surrounding the patient receptacle 22. The operation of the magnetic resonance device 20 is controlled by a control unit 25, the functional structure of which is described in more detail in Fig. 7. Accordingly, the control unit 25 initially comprises a storage medium 26 in which various data, in particular magnetic resonance data, navigator data, correlation information and the like, can be stored. The control unit 25 further comprises a sequence unit 27, which controls the acquisition operation of the magnetic resonance device 20, in particular using an acquisition protocol according to step S1. In a correlation unit 28, correlation information can be determined according to steps S2, S2'. In an evaluation unit 29, sequence segments containing motion are then identified by evaluating the deviation condition according to steps S3, S3', and their magnetic resonance data are discarded. In an adaptation unit 30, the acquisition protocol can be adapted according to step S5. A reconstruction unit 31 serves to reconstruct magnetic resonance images from the (possibly remaining) magnetic resonance data according to step S4 using a trained reconstruction function, which is based in particular on an unrolled neural network architecture. Regardless of the grammatical gender of a particular term, persons with male, female or other gender identities are included.

Claims

Computer-implemented method for operating a magnetic resonance device (20) during a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections (1) of a magnetic resonance sequence, wherein each sequence section (1) comprises a preparation module (8) and a readout module (10) and each readout module (10) comprises several readout submodules (2) in which radio frequency pulses (12, 13) precede readout periods for recording magnetic resonance data, wherein in each readout module (10) at least one navigator submodule (11, 11a, 11b) is used for recording navigator data of the sequence section (1) along a k-space trajectory, wherein for at least one pair of navigator submodules (11, 11a, 11b) the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process.- for each sequence segment (1), correlation information is determined by comparing the navigator data set of the sequence segment (1) with at least one navigator data set of another sequence segment (1) and the same k-space trajectory, and - the correlation information is evaluated for the selection of sequence segments (1) whose magnetic resonance data are discarded, and / or for assigning a weight to the magnetic resonance data of at least a part of the sequence segments (1) before a reconstruction of a magnetic resonance image, characterized in that a first k-space trajectory and a second k-space trajectory are used which are at an angle to each other in the plane formed by the phase encoding direction and the readout direction. Method according to claim 1, characterized in that the navigator data for all sequence segments (1) are recorded along at least one identical k-space trajectory and / or that each k-space trajectory includes the k-space center. Method according to claim 1 or 2, characterized in that the first k-space trajectory runs in the phase encoding direction and the second k-space trajectory runs in a readout direction of the recording process. Method according to one of the preceding claims, characterized in that the first k-space trajectory and the second k-space trajectory are perpendicular to each other. Method according to one of the preceding claims, characterized in that the first and the second k-space trajectory are alternated according to a regular usage pattern via the navigation submodules (11, 11a, 11b). Method according to claim 5, characterized in that each readout module (10) comprises a navigator submodule (11) and the first and second k-space trajectories are changed with each new sequence segment (1), or that the first and second space trajectories are changed after each repetition, or characterized in that each readout module (10) comprises two navigator submodules (11a, 11b), one of which uses the first and one of which uses the second k-space trajectory. Method according to one of the preceding claims, characterized in that the at least one k-space trajectory of the navigator submodule (11, 11a, 11b) is selected to be shorter than the acquisition trajectory of the readout submodules (2). Method according to one of the preceding claims, characterized in that the navigator submodules (11, 11a, 11b) cover all possible positions in the readout module (10) and have at least partially different positions within the readout module (10), wherein the navigator data are additionally evaluated to determine phase evolution information about the readout module (10), which is used for eddy current effect correction of the magnetic resonance data. Method according to one of the preceding claims, characterized in that, if at least one deviation condition evaluating the correlation information is met for one of the sequence sections (1), the magnetic resonance data of at least the sequence section (1) are discarded or, in a subsequent reconstruction of a magnetic resonance image, are weighted less than the magnetic resonance data of sequence sections (1) that do not meet the deviation condition. Method according to one of the preceding claims, characterized in that for reconstruction a reconstruction function, in particular a trained reconstruction function, compensating for missing magnetic resonance data, in particular a reconstruction function based on an unrolled neural network architecture, is used which takes magnetic resonance data in k-space as input data and provides at least one magnetic resonance image in image space as output data. The method according to claim 10, characterized in that the trained reconstruction function uses, in addition to the magnetic resonance data as input data, a sampling mask that describes the distribution of the sampled sampling points in the sampled k-space, wherein - for magnetic resonance data to be discarded, the corresponding sampling points are marked as not sampled and / or - for magnetic resonance data to be weighted differently, the weighting is entered into the sampling mask at the corresponding sampling points, wherein the reconstruction function is configured to use the weighting in the reconstruction and / or to check the consistency between the reconstruction result and the magnetic resonance data. Computer-implemented method for operating a magnetic resonance device (20) during a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections (1) of a magnetic resonance sequence, wherein each sequence section (1) comprises a preparation module (8) and a readout module (10) and each readout module (10) comprises several readout submodules (2) in which radio frequency pulses (12, 13) precede readout periods for recording magnetic resonance data, wherein in each readout module (10) at least one navigator submodule (11, 11a, 11b) is used for recording navigator data of the sequence section (1) along a k-space trajectory, wherein for at least one pair of navigator submodules (11, 11a, 11b) the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process.- for each sequence segment (1), correlation information is determined by comparing the navigator data set of the sequence segment (1) with at least one navigator data set of another sequence segment (1) and the same k-space trajectory, and - the correlation information is evaluated for the selection of sequence segments (1) whose magnetic resonance data are discarded, and / or for assigning a weight to the magnetic resonance data of at least a part of the sequence segments (1) before a reconstruction of a magnetic resonance image, wherein the navigator data are recorded at a fixed position relative to the readout submodules (2), characterized in that, in a turbo spin echo sequence in which the radio frequency pulses (12) of the readout submodules (2) are refocusing pulses, the magnetic resonance sequence is either the navigator submodule (11, 11a, 11b) or a gradient echo submodule after all readout submodules. (2) is,wherein an excitation pulse (13) of the navigator submodule (11, 11a, 11b) is output with a reduced flip angle compared to the high-frequency pulses (12) of the readout submodules (2) and / or at a time interval to the preceding high-frequency pulse (12) of the last readout submodule (2) that is less than the time interval between the high-frequency pulses (12) of the readout submodules (2), or the navigator submodule (11, 11a, 11b) is a turbo spin echo submodule in which the refocusing pulse has a smaller flip angle. Method according to claim 12, characterized in that the navigator data is recorded before all readout submodules (2) or after all readout submodules (2). Computer-implemented method for operating a magnetic resonance device (20) during a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections (1) of a magnetic resonance sequence, wherein each sequence section (1) comprises a preparation module (8) and a readout module (10) and each readout module (10) comprises several readout submodules (2) in which radio frequency pulses (12, 13) precede readout periods for recording magnetic resonance data, wherein in each readout module (10) at least one navigator submodule (11, 11a, 11b) is used for recording navigator data of the sequence section (1) along a k-space trajectory, wherein for at least one pair of navigator submodules (11, 11a, 11b) the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process.- for each sequence segment (1), correlation information is determined by comparing the navigator data set of the sequence segment (1) with at least one navigator data set of another sequence segment (1) and the same k-space trajectory, and - the correlation information is evaluated for the selection of sequence segments (1) whose magnetic resonance data are discarded, and / or for assigning a weight to the magnetic resonance data of at least a part of the sequence segments (1) before a reconstruction of a magnetic resonance image, characterized in that the correlation information is determined at least partially broken down according to the readout direction and the phase encoding direction as breakdown directions, wherein, upon fulfillment of a correctability condition that evaluates the correlation information for a first of the breakdown directions,for one of the sequence segments (1) and at least in the case of a greater deviation in the first resolution direction than in the other resolution direction, the magnetic resonance data of this sequence segment (1) are corrected in the first resolution direction depending on the correlation information for the readout direction. Method according to claim 14, characterized in that the first decoding direction is the readout direction and the other decoding direction is the phase encoding direction, and / or characterized in that the magnetic resonance data of this sequence segment (1) are corrected in the first decoding direction by a shift in image space and / or a phase ramp in k-space. Magnetic resonance device (20) comprising a main magnet unit (21) with a main magnet for generating a main magnetic field, a gradient coil arrangement (23), a high-frequency coil arrangement (24), and a control unit (25), which comprises: - a sequence unit (27) for controlling a recording process according to a recording protocol, which in at least one repetition comprises several sequence sections (1) of a magnetic resonance sequence, wherein each sequence section (1) comprises a preparation module (8) and a readout module (10), and each readout module (10) comprises several readout submodules (2) in which high-frequency pulses (12, 13) precede readout periods for recording magnetic resonance data, wherein the sequence unit (27) is configured to include at least one navigator submodule (11, 11a, 11b) in each readout module (10) for recording navigator data of the sequence section (1) along a to use k-space trajectories,wherein for at least one pair of navigator submodules (11, 11a, 11b) the k-space trajectory used in these runs at least partially along at least one phase-encoding direction of the recording process, a correlation unit (28) for determining correlation information for each sequence segment (1) by comparing the navigator data set of the sequence segment (1) with at least one navigator data set of another sequence segment (1) and the same k-space trajectory, and an evaluation unit (29) for evaluating the correlation information to select sequence segments (1) whose magnetic resonance data are to be discarded, and / or to assign a weight to the magnetic resonance data of at least a part of the sequence segments (1), characterized in that the magnetic resonance device is configured to carry out a method according to one of the preceding claims. Computer program which has programming means such that when the computer program is executed on a control device (25) of a magnetic resonance device (20), the latter is caused to carry out the steps of a method according to one of claims 1 to 15. Electronically readable data carrier on which a computer program according to claim 17 is stored.

Images