Method and image data generation device for generating image data of a moving object, magnetic resonance system and computer program product

Abstract

Method for generating image data (BD) of a moving object (O), comprising the following steps: - Acquisition of raw data (RD) for an area encompassing the object (O) at various measurement times (t0, ..., t k , ...) in different phases of movement (Ph0, Ph1, ..., Ph n ) of the object (O), - Reconstruction of several interim image data sets (BBD) of the object (O) from the raw data (RD), each representing different phases of movement (Ph0, Ph1,..., Ph n ) of the object (O) are assigned, - Determination of deviation data (D0, D1,..., D j ,..., D n ) between the interim image data sets (BBD) of the different movement phases (Ph0, Ph1,..., Ph n ) of the object (O), and - Reconstruction of image data (BD) from raw data (RD) of different motion phases (Ph0, Ph1,..., Ph n ) taking into account the deviation data (D0, D1,..., D j,..., D n ).

Description

[0001] The invention relates to a method for generating image data, in particular magnetic resonance images, of a moving object, for example an organ, parts of an organ, or any target structure inside a human or animal body. Furthermore, the invention relates to an image data generation device for generating image data of an object moving inside a body, as well as to a magnetic resonance system with such an image data generation device.

[0002] In order to obtain image data (magnetic resonance images) from an area inside the body of an examination subject during magnetic resonance imaging, the body or the body part to be examined must first be exposed to a static basic magnetic field that is as homogeneous as possible (usually referred to as the B0 field). This aligns the nuclear spins in the body parallel to the direction of the B0 field, commonly referred to as the z-direction. In addition, radio-frequency antennas are used to radiate radio-frequency pulses into the examination subject. The pulses' frequency lies in the range of the resonant frequency, the so-called Larmor frequency, of the nuclei to be excited (usually hydrogen nuclei) in the existing magnetic field. These radio-frequency pulses are therefore referred to below as magnetic resonance radio-frequency pulses. The magnetic flux density of these radio-frequency pulses is usually denoted by B1.These radiofrequency pulses excite the nuclear spins of the atoms in the object under investigation in such a way that they are deflected from their equilibrium position parallel to the basic magnetic field B0 by a so-called "excitation flip angle" (commonly also called "flip angle" for short). The nuclear spins then initially precess around the z-direction and gradually relax again. The in-phase orbit of the microscopic spins around the precession cone can be viewed as macroscopic nuclear magnetization in the x / y plane (perpendicular to the z-direction). The magnetic resonance signals generated by this relaxation of the nuclear magnetization are recorded as raw data using radiofrequency receiving antennas.The magnetic resonance images of the subject under examination are finally reconstructed based on the acquired raw data. Spatial encoding is achieved using rapidly switched gradient magnetic fields superimposed on the basic magnetic field during the transmission of the magnetic resonance radiofrequency pulses and / or the acquisition of the raw data. During data acquisition, the raw data is first assigned to a spatial frequency space, known as k-space. The image data is then reconstructed using a Fourier transform of the raw data from k-space to image data space (spatial space).

[0003] In particular, if image data of an entire organ is to be generated, i.e. either volume data or a large number of closely spaced cross-sectional images are to be acquired that cover the entire area in which the organ is located, a longer measurement time is required. This is a problem when the organ is in motion, as is the case with the heart, for example. The heart not only performs its own cyclical movements due to the heartbeat, but is also subject to movement due to the patient's breathing. If raw data were simply acquired during the various movement cycles and then image data were reconstructed from this, this would lead to significant motion artifacts, such as blurring in the images, so that the generated images could often not be used for a meaningful diagnosis.

[0004] One way to avoid this is to only acquire raw data in a specific movement phase, for example by triggering the raw data acquisition in a suitable way. For example, with regard to cardiac movement, it is possible to trigger on a specific cardiac phase using an ECG. The data is preferably acquired in the diastolic phase, as cardiac movement is minimal for a relatively long time during this phase. This diastolic phase lasts approximately 100 ms. However, with many acquisition methods, particularly when acquiring the entire volume or a large number of densely spaced slices covering the heart, these 100 ms are not sufficient to acquire all of the raw data. Therefore, raw data sets or segments of k-space must be acquired in several cardiac cycles, each in the appropriate cardiac phase, in order to fill the k-space with the necessary raw data, before the image data of the volume or slice is acquired as described above.the desired slices can be reconstructed. However, the respiratory movement superimposed on the heart's movement means that the heart can be located in different locations within the body, for example, during the diastolic phases of successive cardiac cycles, depending on the current movement phase of the respiratory cycle. In principle, it would of course be possible to monitor the respiratory cycle using appropriate equipment and to additionally trigger on the respiratory cycle. In this case, however, raw data can only be acquired if the desired cardiac phase and the desired respiratory phase are reached simultaneously by chance. This would result in a total measurement time that is far too long, making such measurement methods unacceptable in practice.

[0005] A method for temporal windowing of angiographic magnetic resonance images taking respiratory phases into account is described, for example, in the article "Respiratory Self-Gated Four-Dimensional Coronary MR Angiography: A Feasibility Study" by Peng Lai et al. in Magn. Reson. Med., Vol. 59, 2008, No. 6, pp. 1378–1385.

[0006] In order to fill a sufficiently large k-space volume with measurement data to reconstruct a three-dimensional volume of a complete organ (such as the heart), it is necessary to acquire raw data at different measurement times during different movement phases of the object. This means that the scanning pattern is divided into different segments, and the segments or raw data sets are usually read out in directly consecutive cardiac cycles, which may be in different movement phases of the object due to respiration. A typical scanning pattern is a so-called phyllotactic spiral pattern, as will be explained in more detail later using the figures. This is a method with readout points that are arranged in a plane (e.g.The individual spokes, viewed from the x / y plane through k-space, lie on a kind of spoke trajectory, but the spokes extending outward from the k-space center are spirally rotated. In the direction perpendicular to the plane (e.g., the z-direction), the individual readout processes of the spokes proceed in a straight line. For example, one complete readout process in the z-direction through k-space is read out for each readout process.

[0007] Regardless of the segments in which the raw data is acquired to populate k-space, there is always the problem that the raw data originates from different movement phases, especially breathing phases. Therefore, a correction within the reconstruction process is required to compensate for these image deformations caused by the different movement phases and thus ensure that no motion artifacts appear in the final image data.

[0008] In the article "3D Radial Sampling and 3D Affine Transform-based Respiratory Motion Correction Technique for Freebreathing Whole-Heart Coronary MRA with 100% Imaging Efficiency" by Himanshu Bath et al. in Magn. Reson. Med., Vol. 65, 2011, No. 5, pp. 1269-1277, it is proposed to generate image data from the raw data from different motion phases and then to generate a motion-corrected image from these image data by appropriate summation. Similarly, in DE 197 13 846 A1 and DE 697 21 900 T2, corrected image data are each created from previously reconstructed image data.

[0009] It is therefore an object of the invention to provide a suitable method for generating image data and an image data generation device which allow the use of raw data from different movement phases, while nevertheless reducing movement artifacts and, ideally, even completely avoiding them.

[0010] This object is achieved by a method according to claim 1 and by an image data generating device according to claim 13.

[0011] According to the method according to the invention, raw data is first acquired for an area encompassing the object at various measurement times during different movement phases of the object. This acquisition of raw data can involve the direct acquisition of the raw data at the magnetic resonance system and / or the transfer of raw data already acquired at the various (previous) measurement times at any magnetic resonance system via a suitable interface.

[0012] The raw data acquired at a measurement time point can, for example, be the raw data set acquired at that specific measurement time point or the acquired segment in k-space. It is clear that the acquisition of such a raw data set or segment also requires a certain (short) period of time, so the term "measurement time point" always also refers to a short "measurement time period," for example, in a cardiac image, the approximately 100 ms in the diastolic phase.

[0013] First, several interim image datasets of the object are reconstructed from the acquired raw data, with each of these interim image datasets being assigned to a different movement phase of the object. The same raw data is used here that will later be required, along with additional raw data, to reconstruct the desired final image data. In other words, the interim image datasets differ from the subsequent image datasets not in that different raw data is used, but in that only a portion of the raw data, e.g., only raw data from a specific movement phase, is used and / or—as will be explained later—raw data from different movement phases with different weightings are used to create interim image datasets that match the movement phases.

[0014] Furthermore, deviation data, such as deformation fields (also called motion fields), between the interim image datasets of the object's various motion phases are determined. In principle, it is sufficient to determine the deviation data of the interim image datasets relative to a reference image dataset, because this also indirectly defines the deviation data between the various interim image datasets.

[0015] Finally, the desired image data is reconstructed from raw data from different movement phases, taking this deviation data into account. Preferably, all acquired raw data from all movement phases are included in order to have the largest possible database for reconstructing the image data. Depending on the reconstruction method and the type of deviation data, the previously determined deviation data can be used directly, or, for example, additional deviation data derived from it can be calculated first. For example, the deviation data can first be inverted, or inverse deformation fields can be calculated in addition to the deformation fields, which are then also included in the reconstruction.

[0016] The method according to the invention thus makes it possible to generate motion-corrected image data based on a relatively large raw data base without significant additional effort, even when only a small amount of raw data is available in the various motion phases and the motion phases lead to significant deviations in the object position and / or shape. Overall, volume image data of a moving object with sufficiently high image quality and reduced motion artifacts can be generated in shorter measurement times than before.

[0017] An image data generation device according to the invention, which can operate according to the method according to the invention, initially has a raw data interface for acquiring raw data for an area encompassing the object, wherein the raw data was acquired at different measurement times during different movement phases of the object. As mentioned above, this can be an interface that, for example, receives the raw data from a memory or from another unit.

[0018] Furthermore, the image data generation device requires an interim image reconstruction unit which is designed to reconstruct a plurality of interim image data sets which are each assigned to different movement phases of the object.

[0019] In addition, a deviation data determination unit is required which is designed to determine deviation data between the interim image data sets of different movement phases of the object or, as explained above, to reference image data sets.

[0020] In addition, an overall reconstruction unit is required, which is designed to reconstruct the desired image data from raw data of the various movement phases, taking the deviation data into account. This overall reconstruction unit can, in principle, be the same reconstruction unit as the interim image reconstruction unit, for example, with a suitable optimizer that calculates the image data matching the measured raw data using a numerical optimization process, except that different input parameters and / or different target functions may be specified. These options will be explained later.

[0021] These image data can then be output to any other components via an image data interface of the image data generation device, for example stored in a memory and / or output on a screen and / or transmitted via a network to a diagnostic station or an external storage device, etc.

[0022] An image data generation device according to the invention can in particular also be integrated as a component in a magnetic resonance system according to the invention.

[0023] In addition to such an image data generation device according to the invention, this device has, in the usual way, a measuring room in which the patient or test subject is located during the measurement, a basic field magnet system with which a basic field magnetic field is applied in the usual way in the measuring room, a transmitting antenna system, a gradient system comprising several gradient coils, a receiving antenna system and a control device which serves to control the basic field magnet system, the RF transmitting antenna system, the gradient system and the RF receiving antenna system.

[0024] The image data generation device according to the invention can also be part of another computed tomography system. Likewise, the image data generation device can be implemented on another, separate device that is connected to a suitable computed tomography system, for example, via a network or can receive the raw data in another way. Furthermore, it is possible for the image data generation device and its various components, in particular the interim image reconstruction unit, the deviation data determination unit, and the overall reconstruction unit, to be implemented in the form of software components on a suitable computer or on one or more suitable processors. This also applies to the raw data interface and, if applicable, an image data interface; these interfaces can be partially configured as hardware interfaces or combined software / hardware interfaces.The interfaces can also be pure software interfaces if the data is simply taken over by other software components.

[0025] A largely software-based implementation has the advantage that even previously used image generation devices can be easily retrofitted to operate in the manner according to the invention via a software update. In this respect, the object is also achieved by a computer program product that is stored in a portable memory and / or made available for transmission via a network and can thus be loaded directly into the memory of a programmable image generation device, with program sections for executing all steps of the method according to the invention when the program is executed in the image generation device.

[0026] The dependent claims and the following description each contain particularly advantageous embodiments and developments of the invention. In particular, the claims of one claim category can also be developed analogously to the dependent claims of another claim category. Furthermore, within the scope of the invention, the various features of different embodiments and claims can also be combined to form new embodiments.

[0027] In a particularly preferred variant of the method, the raw data are weighted differently during the reconstruction of various interim image data sets, depending on the movement phase in which they were acquired. In this way, it can be ensured that not only raw data recorded during a specific movement phase are used to determine interim image data sets, but also that additional raw data can be added, for example with a correspondingly lower weighting, in order to increase the data basis for the interim image data sets as well. It has been shown that for the subsequent determination of the deviation data, it is not highly relevant that only raw data from the assigned movement phase itself is determined; rather, a slightly larger data basis leads to better results.

[0028] Particularly preferably, the weighting of the raw data is based on a distance from a specific movement phase for which the interim image data sets are currently being reconstructed. The "distance" can be understood as a temporal distance within a movement cycle, for example, a breathing cycle. Preferably, however, this is a spatial distance, which can be determined, for example, from a deviation in the raw data or from image data reconstructed from it.

[0029] In a particularly preferred method, the raw data is assigned to different motion phase classes, i.e. the raw data is classified or sorted. In this case, for example, the distance of the raw data to a specific motion phase can also be understood as the distance between the motion phase classes, e.g. how close the motion phase classes are to one another or how many other motion phase classes lie in between. For example, the interim image datasets can then be reconstructed for different motion phase classes, and during reconstruction, the raw data is weighted depending on a motion phase class assigned to it. The further the motion phase class of the raw data is removed from the motion class for which the interim image dataset is to be reconstructed, the lower the weighting.

[0030] When reconstructing the interim image data sets for a specific movement class, the raw data of the relevant movement phase class, i.e. the raw data assigned to the relevant movement phase class, with a higher weighting, as well as the raw data of other, for example directly or indirectly adjacent, movement phase classes with a lower weighting in relation to the raw data of the relevant movement phase class for which the reconstruction of the interim image data is carried out, are used according to a weighting function, preferably a Gaussian function or a similar function.

[0031] To reconstruct the image data and / or the interim image data sets, an iterative numerical optimization method can preferably be used. A nonlinear optimization method, in particular a quasi-Newton method, as explained later, is particularly preferred. For this purpose, for example, during the subsequent reconstruction of the image data, taking the deviation data into account, the deviation data can first be inverted and then used in an objective function of the optimizer.

[0032] The classification of the raw data into the different movement phase classes can be done in different ways.

[0033] One option is to acquire so-called special navigator raw data, from which navigator images are then generated. These navigator images usually capture a volume in which respiratory movement is particularly clearly visible. For example, in cardiac images, the diaphragm is observed in the navigator image because the liver / lung junction is very easy to identify. The liver appears relatively bright in the images due to fluid filling, while the lungs appear relatively dark due to air filling. Using the current respiratory position or respiratory phase determined from the navigator images, the raw data acquired immediately after the navigator raw data was acquired and / or the image data of the desired object are then corrected for the respiratory phase.However, to perform these procedures, the acquisition of additional navigator raw data is always necessary, which requires additional effort and, in particular, additional measurement time.

[0034] For a particularly preferred method, position overview data is generated for the various measurement times based on at least a portion of the raw data itself. This means that only raw data that is also used to reconstruct the image data of the desired FoV or object is used for the position overview data, so that no additional navigator measurement is required. In order to generate position overview data for the various measurement times based on at least a portion of the raw data, the image data generation device can have an overview data generation unit. The position overview data can be image data that allows detection of the current movement phase in some way. In principle, this can be three-, two-, or even just one-dimensional data. Based on this position overview data, the raw data is then assigned to the various movement phase classes.

[0035] To create simple one-dimensional position overview data, raw data can preferably be acquired along a line (central readout process), e.g., in the z-direction (i.e., in the direction of the body's longitudinal axis) through the center of k-space when acquiring segments of k-space. This is particularly useful when, as mentioned above, raw data for k-space are being filled in to reconstruct a volume according to a phyllotactic spiral pattern. In addition to the central readout process, only raw data along further lines parallel to the central line need to be acquired, e.g., along readout processes in the z-direction that run through points on a spiral-shaped outward trajectory within an x / y plane perpendicular to the z-direction.

[0036] The central readout process is preferably selected so that it runs in a main direction of movement of the organ. On the basis of the raw data acquired along the central readout process running through the k-space center, a one-dimensional projection in the main direction of movement of the organ can then preferably be generated for each raw data set or each segment, e.g. in the case of the heart in the SI direction (SI = superior - inferior, i.e. along the body's longitudinal axis or in the z-direction). These so-called "SI projections" are one-dimensional image data, namely the projections of the image data running along the z-direction of the entire acquired volume (in the FoV = field of view) onto the z-axis. From these SI projections, the respiratory position can be determined without a navigator and the raw data and / or image data can be corrected during reconstruction. This option is described by Piccini D. et al.in „Respiratory Self-Navigation for Whole-Heart Bright-Blood Coronary MRI: Methods for Robust Isolation and Automatic Segmentation of the Blood Pool“, 2012, Magnetic Resonance in Medicine, 68: 571 bis 579, beschrieben.

[0037] To perform the classification, deviation values ​​can preferably be determined according to predefined rules for the position overview data from different measurement times to certain previously defined reference position overview data. For example, a cross-correlation of the projection data from different SI projections to the projection data of a reference SI projection (in the simplest case, the SI projection of the first measurement time) can be carried out. The raw data are then assigned to different movement phase classes depending on the deviation values. The reference position overview data can, in principle, be selected arbitrarily. For example, the values ​​at the measurement time t = 0, i.e.The first position overview data generated at the beginning of the measurement are used as the reference position overview data (zero reference), and all subsequent position overview data are then referenced to this reference position overview data. In principle, however, the position overview data could also be analyzed first to search for particularly suitable reference position overview data, for example, position overview data recorded during a very specific movement phase.

[0038] In order to further improve this classification process (also known as “binning”), the following steps can preferably be carried out: First, a measurement-time-dependent scatter of the position overview data is determined for the individual positions or spatial coordinates or position ranges, for example, using a suitable overview data analysis unit of the image data generation unit. For this purpose, the scatter of the signal value at a specific position within the position overview data can be determined as a function of time. For example, with an SI projection, the scatter of the projection signal value can be determined as a function of time at a specific location z. The scatter is measured according to a predefined scatter measure. This is preferably the variance.

[0039] Subsequently, spatial inspection areas are selected within the position overview data. For this purpose, the image data generation device can have an inspection area selection unit. In a one-dimensional projection, such as an SI projection, projection sections can be identified as inspection areas. This selection of spatial inspection areas is based on the scatter of the position overview data in the respective inspection areas. Preferably, the position areas or position sections within the position overview data can be determined precisely where the scatter is relatively high, since movement within the body should be particularly easy to detect here.

[0040] Furthermore, confidence parameter values ​​are determined for the individual test areas. As will be explained later, these confidence parameter values ​​can be used to determine, for example, in which of the test areas motion detection or the precise determination of the movement phases is likely to be most reliable. For this purpose, the image data generation device can have a test area selection unit.

[0041] Finally, the raw data are assigned to the motion phase classes, taking into account the confidence parameter values ​​of the various test areas. It has been shown that by identifying different test areas and checking which of these test areas is particularly well suited for detecting movement, the image quality can be significantly improved in most cases. While the images still contained motion artifacts in 6 out of 10 subjects when a (static) test area was defined in advance, the described procedure for automatically defining the optimal test area without using navigator measurements still produced images of equivalent quality to those otherwise achievable using additional navigator measurements.Thus, the method allows for the generation of very good volume images of the heart without the use of navigators, thus significantly reducing the overall measurement time. This is likely due to the fact that for different patients or subjects, the ideal test areas may be located at different locations within the position overview data due to anatomical differences, and the method according to the invention can automatically account for these different locations of the optimal test areas.

[0042] A wide variety of values ​​can be used as confidence parameter values ​​for the individual test areas. In a preferred variant, the confidence parameter value of a test area itself is based on the scatter of position overview data within the test area.

[0043] For this purpose, for example, the above-described classification of the position overview data or the associated raw data is first carried out separately for each of the individual inspection areas based on the deviation values ​​of the position overview data from the reference position overview data. The reference position overview data can also be determined separately for the individual inspection areas.

[0044] A measurement-time-dependent local dispersion value of the position overview data can then be determined separately, preferably for each movement phase class. This dispersion value can be determined according to the same rules, i.e., with the same degree of dispersion as the dispersion described above, according to which the test areas are initially selected, but with the difference that now only the dispersion of the position overview data of the respective movement phase class is determined around a mean value calculated from this data.

[0045] Further preferably, a sum of the scatter values ​​of several movement phase classes can then be determined as the confidence parameter value for a test area. Preferably, the confidence parameter value is a type of "total deviation value scatter," i.e., the sum of the scatter values ​​of all movement phase classes.

[0046] Particularly preferred is then the reconstruction of image data from the raw data, taking into account the assignment of the position overview data to the motion phase classes for the test area with the lowest confidence parameter value. This means, for example, that the test area with the lowest sum of the individual scatter values ​​in the motion phase classes is classified as more trustworthy. Of course, depending on how the confidence value is determined, it would also be possible to use the test area with the highest confidence parameter value.

[0047] As already mentioned above, the method is preferably used on examination subjects that are moved due to a respiratory movement. The movement phases are then correspondingly respiratory phases. The examination subject is particularly preferably or includes a heart. However, this does not preclude the method from also being used on other examination subjects and for other movement sequences, for example, to correct swallowing movements or other movements.

[0048] The invention is explained in more detail below with reference to exemplary embodiments in the accompanying figures. They show: Fig. 1 a schematic representation of the SI direction and the position of the heart and diaphragm in the human body, Fig. 2 a schematic representation of an ECG with a representation of the triggering of a raw data acquisition in a diastolic phase, Fig. 3 a representation of a k-space volume in three-dimensional Cartesian coordinates k x , k y , k z with a phyllotactic spiral scanning pattern (left a top view of the k x / k y -level and on the right a perspective view), Fig. 4 a flowchart of a possible sequence of a method according to the invention for generating image data, Fig. 5 a representation of a possible process for determining test areas in position overview data (here in the form of SI projections) and the assignment of the position overview data to movement phase classes, Fig. 6 a histogram for the number of acquired raw data sets for different movement phase classes, Fig. 7 a schematic representation of a possible procedure for reconstructing image data from raw data of different movement phases, and Fig. 8 a schematic representation of a magnetic resonance system with an image data generation device according to an embodiment of the invention.

[0049] The preferred variant of the method described below with the help of the figures involves the acquisition of a complete volume of a heart as the examination object O in a human body K. Avoiding motion artifacts when generating magnetic resonance images of the entire heart O is associated with particular challenges due to the constant movement of the heart itself and the continued superimposed respiratory movement. This is demonstrated by the Fig. 1 and Fig. 2 illustrates this.

[0050] Fig. Figure 1 shows a rough schematic of the position of the heart O in the patient's body K and also the position of the diaphragm O', which, as mentioned, marks the transition between the liver and lungs and is therefore usually clearly visible in the image data due to a strong light-dark transition. During breathing, the diaphragm O' in the patient's body is constantly raised and lowered, which leads to a strong displacement of the heart O within the body, primarily in the longitudinal direction (SI direction). This movement in the SI direction is therefore the main direction of movement R, to which the heart O as a whole is subject within the body K and which, depending on the breathing phase, is displaced in the superior direction (towards the head) or in the inferior direction (towards the feet). In addition, the heart O itself performs its own cyclical movement through the heartbeat. During a cardiac cycle, the heart goes through several so-called cardiac phases. These are Fig. 2 is shown in a schematic ECG. A cardiac cycle is usually viewed from one R-wave RZ to the next R-wave RZ (the time of the strongest contraction of the heart). Approximately in the middle range between two R-waves RZ is the so-called diastolic phase, in which the heart is at rest for a relatively long time, namely approximately 100 ms. This diastolic phase can therefore be used to acquire raw data. For this purpose, the magnetic resonance pulse sequence for raw data acquisition is triggered on the R-wave, as shown in Fig. 2 is schematically represented by the trigger time period TZ. In a suitable magnetic resonance pulse sequence, for example, a T2 preparation (TP) can be performed first, followed by a so-called fat-saturated FS, and finally the actual raw data acquisition (RDA).

[0051] In order to “fill” a three-dimensional k-space volume with raw data, from which a 3D image data volume (the FoV) can then be reconstructed, for example using a 3D Fourier transformation, in a Cartesian k-space R k several trajectories with in the z-direction (in the direction of the coordinate k z in k-space) parallel readout processes, whereby these readout processes in k-space R k related to a k x / k y -plane in a specific pattern. An example of a trajectory for such a preferred readout type is shown in Fig. 3. Here, the z-direction readouts of the trajectories are shown according to a phyllotactic spiral scanning pattern with respect to the k x / k y -level. As shown in the top view of the k x / k y-plane on the left, the pattern consists of several spirally curved spokes extending radially outwards from the center (or the z-axis) of k-space. It is therefore a mixture of a spiral trajectory pattern and a spoke trajectory pattern. On the right, a spiral spoke is shown in a three-dimensional perspective view. From this view, it can be seen that each point of the trajectory in the k x / k y -level one in k z -direction readout process along which k-space is scanned to acquire raw data. In the time window of a diastolic phase of 100 ms, approximately 30 such k z-direction readouts can be performed, which corresponds to a spiral spoke trajectory. Therefore, within a diastolic phase, a raw data segment can be read out as a raw data set, which includes the raw data RD of a spiral spoke, as shown on the right side in Fig. 3 by the theorem of longitudinal in k z -direction readout processes are shown schematically. In each case, the raw data RD Cread out through the center of k-space along the central readout process. The raw data RD is then acquired along the further readout processes, progressing from the inside out. To sufficiently fill the k-space volume for a reconstruction of meaningful image data, enough raw data sets must be acquired so that k-space is approximately 12% full. Once the required raw data for all segments have been acquired, the three-dimensional image data volume can be reconstructed. Raw data can be acquired in a similar way to capture densely spaced slices of the heart.

[0052] Unfortunately, however, the heart is spatially displaced and / or deformed by respiratory movement between the acquisition of the raw data sets or segments of the k-space data in successive cardiac cycles. Reconstructing the image data therefore requires correcting this displacement / deformation (hereinafter referred to as displacement). This is possible in principle in several ways. However, for this to happen, the displacement must be known at least approximately, i.e., it must at least be determined in which respiratory phase the individual raw data sets were acquired.

[0053] This is advantageous with the Fig. 4. In step 4.I, all raw data RD is first acquired. These are then preprocessed in step 4.II using a special sorting algorithm (also called "binning") to assign them to different movement phase classes. This step 4.II is described below using Fig. 5 is explained in more detail.

[0054] For each acquired raw data set or raw data segment, the central raw data RD C , which were acquired along the readout process of the trajectory through the k-space center (see Fig. 4), is transferred into the image data space using a one-dimensional Fourier transformation. The result is a one-dimensional so-called SI projection of the complete field of view, i.e. all image data in the FoV along the z-direction (i.e., in the main direction of movement R; see Fig. 1) projected onto the z-axis. If the field of view FoV is sufficiently large to cover not only the heart O but also the diaphragm O', as shown in Fig. As shown in Figure 1, the light-dark transition at the diaphragm is also clearly visible in these SI projections (SIP). The SI projections (SIP) are therefore particularly well-suited as position overview data (SIP).

[0055] In a first step 5.I, a measurement-time-dependent local scatter for these SI projections SIP is determined. Such a process is described in Fig. 5 in step 5.I using a diagram. This diagram shows the individual SI projections SIP close together, with the location in the z-direction along the individual SI projections SIP on the ordinate axis and the intensity value present at a location z at a specific measurement time t on the abscissa axis. The variance var(z(t)) is used here as the measurement-time-dependent local measure of dispersion, i.e. the variance of a large number of measured values ​​z(t) at a specific location z over several measurement times t is determined. This variance var(z(t)) is shown as a function of z in the uprightly displayed variance curve immediately next to the diagram in step 5.I. By determining the maxima and minima of this variance curve, possible test areas PR1, PR2 are selected in which a particularly high variance exists.For this purpose, a local maximum is searched for and two local minima are selected at a certain distance from the found local maximum, thus defining the test area around the maximum.

[0056] In the Fig. In the case illustrated in Figure 5, exactly two test areas PR1, PR2 (or actually "test sections" of the SI projections, since the SI projections are only one-dimensional) are selected along the SI projections SIP. One of these test areas PR1 covers the area in which the heart O moves, while the second test area PR2 covers an area in which the diaphragm O' moves.

[0057] In step 5.II, deviation values ​​of the position overview data SIP, ie here the SI projections SIP, which are recorded at different times t k(k = 0 to m, where m is the number of raw data records acquired) are determined from reference position overview data. Any position overview data SIP can be selected as the reference position overview data (or as the "reference SI projection"). In this case, for the sake of simplicity, the SI projection acquired during the first measurement at time t = 0 is used as the reference SI projection SIP0. To determine the deviation value of an SI projection from the reference SI projection SIP0, a cross-correlation of the signal to the reference SI projection SIP0 is determined in the individual test areas PR1, PR2. In this way, a deviation value Δz 0,k,1 and Δz 0,k,2 for each individual SI projection SIP. In step 5.III, the values ​​obtained at the different times t krecorded individual SI projections - again separately for each test area PR1, PR2 - based on their deviation values ​​Δz 0,k,1 or Δz 0,k,2 different movement phase classes BK0, BK1, BK2, ..., BK n where n is the maximum number of movement classes. For example, the movement class BK0 corresponds to a deviation value of +0.5 to a maximum of -0.5, the movement class BK1 to a possible deviation between -0.5 and a maximum of -1.5, etc. The sign or the permitted deviation values ​​for the individual movement phase classes BK0, BK1, BK2, ..., BK n can, in principle, be freely selected. In this case, they are chosen so that the first image is taken at time t = 0, when the patient has completely exhaled and the diaphragm is thus at the very top of the body K. The deviations in the subsequent breathing phases are thus characterized by a downward shift in the negative z-direction toward the feet.

[0058] In the movement class BK0, all SI projections SIP and of course the associated other raw data from the same raw data set or segment are assigned to the movement class BK0, for which only a minimal deviation from the reference SI projection SIP0 was detected in the respective test area PR1 or PR2. As the class number increases, this deviation increases, i.e. the further the recording is removed from the movement phase in which the reference SI projection SIP0 was recorded. Since this sorting was carried out separately for the two test areas PR1 and PR2, it is entirely possible that, according to the sorting in the first test area PR1, an SI projection and the associated raw data were sorted into the same movement class BK0 as the reference SI projection, and within the second test area PR2, the same raw data was sorted into a different class, e.g. the neighboring class BK1.

[0059] In principle, it can be assumed that the signal variations of the SI projections recorded in the same respiratory phase or movement phase within the respiratory cycle should be relatively small. Furthermore, if the global variations are minimal, this should also apply to each subinterval within the SI projections. Therefore, it can be assumed that the detection of respiratory movements and the sorting of the SI projections or associated raw data into the individual movement phase classes BK0, BK1, BK2, ..., BK n in the one of the two test areas PR1, PR2 in which the smallest variations within the movement phase classes BK0, BK1, BK2, ..., BK n Therefore, as described in step 5.III in Fig. 5, the time-dependent local variance of the SI projection SIP in the individual test areas PR1, PR2 is determined. This variance determination is carried out analogously to the procedure in step 5.I, but now separately for the SI projections in the individual movement phase classes BK0, BK1, BK2, ..., BK n In other words, for each movement class BK0, BK1, BK2, ..., BK n a separate dispersion value SW0, SW1, SW2, ..., SW n determined.

[0060] These dispersion values ​​SW0, SW1, SW2, ..., SW n are then summed separately for each test area PR1, PR2 to obtain a separate confidence parameter value VP1, VP2 for each of the test areas PR1, PR2. As shown in Fig. As shown in Figure 5, the confidence parameter value VP2 for the second test area PR2 is 0.22, significantly lower than the confidence parameter value VP1 for the test area PR1, which is 0.42. Therefore, it can be assumed that more reliable detection of the movement was possible in the test area PR2. Therefore, for the further reconstruction or correction of the raw data with regard to respiratory movement, the sorting of the individual raw data determined in the test area PR2 into the movement phase classes BK0, BK1, BK2, ..., BK n used.

[0061] Surprisingly, it has been found that the initial automatic identification of possible good test areas PR1, PR2 and a subsequent automatic check to determine which of these test areas PR1, PR2 has the greatest reliability and the subsequent use of the classification according to this most reliable test area PR1, PR2 leads to a dramatic improvement in the method compared to a method in which a test area is simply defined at the beginning.

[0062] Using this classification or using the deviation values ​​Δz determined in step 5.II 0,k,2Using the method explained below, the image data is then reconstructed from the raw data in various motion phases in order to generate images that are largely free of motion artifacts. However, it should be noted again that the assignment of the raw data to the motion phases could also be performed in other ways, for example, with the help of navigators.

[0063] When scanning a three-dimensional k-space with the phyllotactic spiral scanning pattern described above, as shown in Fig. As shown in Figure 3, a relatively large amount of raw data can be acquired within a given acquisition time to reconstruct image data. On the other hand, however, a relatively high degree of subsampling is still required to capture the entire volume of an organ, such as the heart, within an acceptable total acquisition time.

[0064] It is therefore desirable to use not just a small portion of the acquired raw data, for example only raw data from movement class BK0 in a movement state during full inspiration (during this phase the heart movement is in a kind of resting phase for a relatively long time), but as much, preferably all, of the acquired raw data. To do this, however, the raw data must be corrected with regard to their relative displacement during reconstruction, as explained. From the sorting process above, it is already known which raw data was acquired in which movement phases. For this purpose, the raw data has already been sorted into movement phase classes. However, it is not known how the raw data acquisition in the different movement classes / phases actually affects the displacement of the raw data or the reconstructed image data.

[0065] Reconstruction of image data from the complete raw data, taking into account the displacement of spatial structures caused by respiratory movement, could be performed using an iterative numerical optimization method, preferably a quasi-Newton method, which is particularly suitable for CS data (compressed sense data). This is a nonlinear optimization method in which operations are performed relatively quickly pixel by pixel, minimizing the following cost function: argminx ∑j∑i‖MjFCiDjx−yi,j‖22+λ|x|TV

[0066] In this cost function, the vector x represents the image data and the vector y ij the raw data acquired with a coil i in a breathing or movement phase j. C i is the coil sensitivity of the i-th coil (in matrix form). D jrepresents the deformation field (in matrix form), ie it is the model that describes the movement during the acquisition of the data y ij represents. F is the Fourier transform operator. M j represents the sampling pattern for the respective motion phase j in matrix form. The summation is done via the squares of the Euclidean norm. The additional term λ|x| TV Total Variation (TV) is a regularization term that increases the cost of excessive deviations and thus ensures sufficient convergence of the method. The factor λ can, in principle, be chosen arbitrarily; it should preferably be between 0 and 1. It depends, among other things, on the scaling of the measured data.

[0067] In order to work with this method, however, the movement field D j for each respiratory phase j. For this purpose, the Fig. 4, interim image data BBD are first reconstructed in step 4.III for the individual motion phases or the motion phase classes determined using the method described above. This can again be done using a quasi-Newton optimization method, but now using the following objective function: argminx ∑i‖W(MFCix−yi,j)‖22+λ|x|TV

[0068] As can be seen, this objective function no longer contains a summation across different movement phases j. Instead, a weighting factor W (also in matrix form) is introduced. This weighting factor W ensures that acceptable interim image data can be determined for a single movement phase. However, not only the raw data from the exact corresponding movement phase class are used, but also raw data from other, particularly directly or indirectly neighboring, movement phase classes. However, the raw data from the other movement phase classes are weighted according to the weighting W such that raw data acquired from movement phases far removed from the actually desired movement phase are only given a very low weighting compared to the raw data from the actually desired movement phase.

[0069] A suitable weighting function can be used, for example, a Gaussian weighting. This is shown schematically in Fig. 6. Here, the number #RD (see the left vertical axis of the diagram) of acquired raw data sets or segments is plotted against the detected deviation Δz of the respective raw data set in a histogram. The columns of the histogram correspond to the motion phase classes, with the motion phase class BK0 represented by the column around the value Δz = 0, the motion phase class BK5 represented by the column around the value Δz = -5, etc.

[0070] In addition, a weighting function GF (scaled to a maximum of 1 in arbitrary units, see the right vertical axis of the diagram), specifically a Gaussian function, is plotted here. This visualizes the relative weighting of the raw data from the different motion phase classes when an image for a specific motion phase j (corresponding to a specific motion phase class) is to be reconstructed using the objective function according to formula (2). As can be seen here, the raw data of the motion phase class of the desired motion phase (here, the motion phase class BK0) are included to a maximum, and the Gaussian function GF drops relatively sharply, so that essentially only the raw data of the directly adjacent motion phase class and possibly a small portion of the raw data of the next but one motion phase class are used.It should be noted at this point that such a weighting can, of course, not only be carried out with regard to the movement phase classes, but that in principle a weighting of each individual raw data set can also be carried out on the basis of its exact deviation value Δz, which can be calculated, for example, using the method developed in step 5.II in . Fig. 5. This means that each individual raw data set is taken into account when creating the interim image data BBD for a specific movement phase according to an individual weighting value, which is specified by the weighting function GF. This procedure ensures that even for movement phases in which only relatively little raw data could be generated, sufficient raw data is always available to determine reasonably meaningful interim image data for this movement phase.

[0071] If in step 4.III (see Fig. 4) the interim image data BBD for all desired movement phases or movement phase classes have been generated, in step 4.IV deviation data D j of the individual interim image data, ie the deformation field or movement field D j calculated for the individual movement phases.

[0072] For this purpose, in step 4.IV R, one of the generated interim image datasets is first selected as the reference image dataset. This is preferably the image dataset for which the most raw data was available. For example, this could be the case with the raw data as shown in the histogram in Fig. 6, the interim image data set created for the motion phase with a deviation of Δz = 1 (i.e., Δz between -0.5 and -1.5) or for the motion phase class BK1. The deviation data or the deformation field D jis then determined by registering the image data set of the respective movement phase j to the reference interim image data set of the reference movement phase. A non-rigid, symmetric diffeomorphic image registration is preferably used here. Such a registration method is explained, for example, in the publication by Avants, BB et al. "Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain", MedIA 12: 26 to 41, (2008), to which reference is made here. If, in this way, the movement field D is then determined for each movement phase j j determined, the desired image data BD can be reconstructed using all raw data using the objective function (1) explained above, regardless of the respiratory movement phase in which they were acquired.

[0073] Based on Fig. 7 the determination of the deformation field D j This is illustrated graphically once again. In the data acquisition step AC on the left, the respiratory movement and the resulting deformation or shift in the heart position are shown. The top row shows the first movement phase Ph0, the second row a second movement phase Ph1, and the bottom row a final movement phase Ph n Normally, without these movement phases or without the respiratory movement, the same object position WOL would be present as shown in the first column. However, the respiratory movement AB causes a deformation, which is determined in the individual movement phases Ph0, Ph1, ..., Ph n different and which are each determined by a deformation field or movement field D0, D1,... , D n This then leads to a relatively deformed object position DOL in the individual phases Ph0, Ph1, ..., Ph nThis deformation is also noticeable in the raw data RD in k-space. The jump into k-space is schematically represented here by the Fourier transformation FT. In k-space, however, in the individual movement phases Ph0, Ph1, ..., Ph n the acquisition of different raw data segments RD, whereby in each segment a readout process by the k-space center RD C These raw data sets RD can then be transferred back into the image data space in a reconstruction step RC using a Fourier transformation FT, where, if only the raw data RD from a single movement phase Ph0, Ph1, ..., Ph n be taken into account or as described above according to the desired movement phase Ph0, Ph1, ..., Ph nweighted, interim image data BBD are created, each showing the object in the deformed object position DOL. Through the described registration RG of the interim image data BBD, the deformations or displacements caused by the breathing movement AB, more precisely the deformation fields D0, D1,..., D n and the corresponding inverse deformation fields D0 -1 , D1 -1 , ..., D n -1 which are used in the objective function (1). If the inverse deformation fields D0 -1 , D1 -1 , ..., D n -1 applied to the interim image data BBD, theoretically the deformation would be reversed and image data BD WOL for different movement phases Ph0, Ph1, ... Ph n in the same object position, which could then result in the image data BD. However, it is pointed out again in this context that the Fig. The procedure shown in Figure 7 is intended to illustrate the principle only schematically and the mathematical reconstruction of the desired corrected image data BD is preferably carried out as explained above using the numerical quasi-Newton optimization method with the objective functions (1) and (2).

[0074] In Fig. Finally, Figure 8 shows a roughly schematic representation of a magnetic resonance system 1 according to the invention (hereinafter referred to as "MR system") with which the method according to the invention can be carried out. It comprises, on the one hand, the actual magnetic resonance scanner 2 with a measurement chamber 3 or patient tunnel extending in the z-direction, into which a patient or test subject, in whose body K the examination object O (here the heart) is located, can be moved on a couch 8.

[0075] The magnetic resonance scanner 2 is equipped in the usual way with a basic field magnet system 4, a gradient system 6 as well as an RF transmitting antenna system 5 and an RF receiving antenna system 7.

[0076] In the illustrated embodiment, the RF transmit antenna system 5 is a whole-body coil permanently installed in the magnetic resonance scanner 2, whereas the RF receive antenna system 7 consists of local coils to be arranged on the patient or subject (in Fig. 1 symbolized only by a single local coil). In principle, however, the whole-body coil can also be used as an RF receiving antenna system and the local coils as an RF transmitting antenna system, provided that these coils can be switched to different operating modes.

[0077] The MR system 1 further comprises a central control device 13, which is used to control the MR system 1. This central control device 13 includes a sequence control unit 14 for pulse sequence control. This controls the sequence of radio-frequency pulses (RF pulses) and gradient pulses depending on a selected magnetic resonance measurement sequence. The control parameters for the magnetic resonance measurement sequence can be specified, for example, within a measurement or control protocol. Several measurement or control protocols are stored, for example, in the memory 19 for selection and modification by an operator.

[0078] To output the individual RF pulses, the central control device 13 has a radio-frequency transmission device 15, which generates and amplifies the RF pulses and feeds them into the RF transmit antenna system 5 via a suitable interface (not shown in detail). The radio-frequency transmission device can comprise a variety of components, including small-signal generators, which ensure that the appropriate RF pulses are initially generated with a low amplitude, and suitable RF amplifiers to feed the RF pulses into the antennas with the required power. Furthermore, this also includes monitoring components to ensure that the RF power remains within the limits specified by the SAR (Specific Absorption Rate) standards, etc.

[0079] To control the gradient coils of the gradient system 6, the control device 13 has a gradient system interface 16. Using the components of this interface, the required gradient pulses are generated and then fed into the various gradient coils of the gradient system, or the desired gradient voltages are applied to the gradient coils.

[0080] The sequence control unit 14 communicates in a suitable manner, e.g., by transmitting sequence control data SD, with the radio-frequency transmitting device 15 and the gradient system interface 16 for transmitting the pulse sequences.

[0081] The control device 13 also has a radio-frequency receiving device 17 (which also communicates with the sequence control unit 14 in a suitable manner) for acquiring magnetic resonance signals, i.e., raw data, received in a coordinated manner from the RF transmit antenna system 7. The radio-frequency receiving device accordingly comprises a plurality of receiving channels in which the magnetic resonance signals picked up by the individual antennas of the receive antenna system and possibly further preprocessed are received and further processed, in particular amplified and digitized.

[0082] Since the acquisition of the raw data should always take place at the appropriate time and raw data should only be acquired in the diastolic phase of the heart, the central control device 13 also has an ECG interface 18, which is connected to an ECG device 11, to which the patient is connected with conventional electrodes 12. This, as in Fig. 2, the ECG curve is monitored, and a trigger command is always issued at the appropriate time after an R-wave (RZ), ensuring that the acquisition is started and ended at exactly the right time. This also allows us to determine whether the diastolic phase was maintained throughout the raw data acquisition process or whether, for example, the diastolic phase was too short due to cardiac rhythm fluctuations. In this case, the raw data can be discarded.

[0083] An image data generation unit 20 receives the acquired raw data RD and generates the desired magnetic resonance image data BD of the object O from it in the procedure explained above. These can then be stored, for example, in a memory 19 or output to a user.

[0084] To generate the image data BD in the described procedure, the image data generation unit 20 initially has a raw data interface 21, via which the raw data RD is transferred. In an overview data generation unit 23, the position overview data, or in this case specifically the SI projections, are then generated, which are then transferred to an overview data analysis unit 24, which determines the scatter of the position overview data. This overview data analysis unit 24 is here part of the test area selection unit 25, which selects potential spatial test areas, i.e., the SI projection sections PR1, PR2, based on the scatter, as described above with reference to Fig.5. In a confidence parameter value determination unit 26, the individual test areas PR1, PR2 are then evaluated in the manner described above. For this purpose, the deviation values ​​Δz of the position overview data, i.e. the SI projections, from the reference position overview data are first determined in a classification unit 27, and based on this, the position overview data and the associated raw data are then sorted into the movement phase classes. In a dispersion value determination unit 28, the dispersion, in particular the variance, of the position overview data sorted into the individual movement phase classes is then calculated, and finally, in a summation unit 29, the sum of the variances of the individual movement phase classes is formed in order to arrive at a confidence parameter value VP1, VP2 for each test area PR1, PR2.In this way, a test area is verified within the confidence parameter value determination unit 26. Finally, the information about the classifications of the raw data into the individual movement phase classes is transmitted to a reconstruction unit 30, which then reconstructs the desired image data BD based on the raw data, taking this information into account. This reconstruction unit 30 comprises, firstly, an interim reconstruction unit 31, with which, as described above, interim image data sets BBD are initially generated for each movement phase or movement phase class, which are then transferred to a deviation data determination unit 32. This determines the deviation data D. j or movement fields of the interim image data to the reference image data. The deviation data or the movement fields D determined in this way jare then transferred to a complete reconstruction unit 33, which then reconstructs the desired image data BD using all of the raw data, e.g., using the objective function (1). This image data BD can then be output again via an image data interface 22 and, for example, stored in a memory 19 and / or output to a screen 9. Likewise, this image data BD can also be stored in external memories via a network and / or output to corresponding output devices or displayed on a monitor for diagnosis.

[0085] The central control unit 13 can be operated via a terminal with an input unit 10 and a display unit 9, via which the entire MR system 1 can also be operated by an operator. The image data BD can also be displayed on the display unit 9, and measurements can be scheduled and started using the input unit 10, possibly in combination with the display unit 9.

[0086] The MR system 1 according to the invention and in particular the control device 13 can also have a large number of further components, not shown in detail here, but usually present on such devices, such as a network interface in order to connect the entire system to a network and to be able to exchange raw data and / or image data or parameter maps, but also other data, such as patient-relevant data or control protocols.

[0087] Finally, it should be noted once again that the methods and structures described in detail above are merely exemplary embodiments and that the basic principle can be varied widely by those skilled in the art without departing from the scope of the invention, as long as it is defined by the claims. For the sake of completeness, it should also be noted that the use of the indefinite articles "a" or "an" does not exclude the possibility that the relevant features may be present in multiple instances. Likewise, the term "unit" does not exclude the possibility that it consists of multiple components, which may also be spatially distributed.

Claims

[1] Method for generating image data (BD) of a moving object (O), comprising the following steps: - Acquisition of raw data (RD) for an area encompassing the object (O) at different measurement times (t0, ..., t k , ...) in different movement phases (Ph0, Ph1, ..., Ph n ) of the object (O), - Reconstruction of several interim image data sets (BBD) of the object (O) from the raw data (RD), each of which corresponds to different movement phases (Ph0, Ph1,..., Ph n ) of the object (O), - Determination of deviation data (D0, D1,..., D j ,..., D n ) between the interim image data sets (BBD) of the different movement phases (Ph0, Ph1,..., Ph n ) of the object (O), and - Reconstruction of image data (BD) from raw data (RD) of different movement phases (Ph0, Ph1,..., Ph n ) taking into account the deviation data (D0, D1,..., D j,..., D n ). [2] Method according to claim 1, wherein the raw data (RD) are determined as a function of the movement phase (Ph0, Ph1,..., Ph n ) in which they were acquired, are weighted differently when reconstructing different interim image data sets (BBD). [3] Method according to claim 2, wherein the weighting of the raw data (RD) is dependent on a distance from a movement phase (Ph0, Ph1,..., Ph n ) for which the interim image data sets (BBD) are reconstructed. [4] Method according to one of claims 1 to 3, wherein the raw data (RD) are assigned to different movement phase classes (BK0, BK1, BK2,..., BK n ) can be assigned. [5] Method according to claim 4, wherein interim image data sets (BBD) for different movement phase classes (BK0, BK1, BK2,..., BK n) are reconstructed and during reconstruction the raw data (RD) are analyzed depending on a movement phase class assigned to it (BK0, BK1, BK2,..., BK n ) weighted. [6] Method according to claim 5, wherein in the reconstruction of the interim image data sets (BBD) for a movement phase class (BK0, BK1, BK2,..., BK n ) according to a weighting function (GF) the raw data (RD) of the respective movement phase class (BK0, BK1, BK2,..., BK n ) with a higher weighting as well as raw data (RD) from other movement phase classes (BK0, BK1, BK2,..., BK n ) with a lower weighting. [7] Method according to one of the preceding claims, wherein an iterative numerical optimization method is used to reconstruct the image data (BD) and / or the interim image data sets (BBD). [8] Method according to one of the preceding claims, wherein for the different measuring times on the basis of at least one part (RD c ) of the raw data (RD) position overview data (SIP) are generated, and based on the position overview data (SIP) an assignment of the raw data (RD) to the different movement phase classes (BK0, BK1, BK2,..., BK n ) takes place. [9] Method according to claim 8, wherein deviation values ​​(Δz) for the position overview data (SIP) of different measuring times (t0,..., t k ,...) to reference position overview data (SIP0) and the raw data (RD) are assigned to different movement phase classes (BK0, BK1, BK2,..., BK n ) can be assigned. [10] Method according to claim 8 or 9, wherein the following steps are carried out: - Determination of a measurement time-dependent scatter of the position overview data (SIP), - Selection of spatial test areas (PR1, PR2) within the position overview data (SIP) depending on the dispersion, - Determination of confidence parameter values ​​(VP1, VP2) for the individual test areas (PR1, PR2), and - Assignment of the raw data (RD) to the movement phase classes (BK0, BK1, BK2,..., BK n ) taking into account the confidence parameter values ​​(VP1, VP2) of the different test areas (PR1, PR2). [11] Method according to one of the preceding claims, wherein the examination object (O) is moved due to a respiratory movement and the movement phase (Ph0, Ph1,..., Ph n ) is a breathing phase. [12] Method according to one of the preceding claims, wherein the examination object (O) comprises a heart (O). [13] Image data generating device (20) for generating image data (BD) of a moving object (O), which image data generating device (20) comprises the following components - a raw data interface (21) for acquiring raw data for an area encompassing the object (O), which at different measuring times (t0,..., t k ,...) in different movement phases (Ph0, Ph1,..., Ph n ) of the object (O) were acquired, - an interim image reconstruction unit (31) which is designed to reconstruct a plurality of interim image data sets (BBD) which each correspond to different movement phases (Ph0, Ph1,..., Ph n ) of the object (O), - a deviation data determination unit (32) which is designed to determine deviation data (D0, D1,..., D j ,..., D n ) between interim image data sets (BBD j ) different movement phases (Ph0, Ph1,..., Ph n ) of the object (O), and - an overall reconstruction unit (33) which is designed to generate image data from raw data (RD) of different movement phases (Ph0, Ph1,..., Phn ) taking into account the deviation data (D0, D1,..., D j ,.., D n ) to reconstruct. [14] Magnetic resonance system (1) with - a measuring room (3), - a basic field magnet system (4), - an RF transmitting antenna system (5), - a gradient system (6), - an RF receiving antenna system (7), - a control device (11) for controlling the basic field magnet system (4), the RF transmitting antenna system (5), the gradient system (6) and the RF receiving antenna system (7), and - an image data generating device (20) according to claim 13. [15] A computer program product which is directly loadable into a memory of a programmable image data generating device (20), comprising program code sections to carry out all steps of a method according to any one of claims 1 to 12 when the program is executed in the image data generating device (20).

Images