METHOD FOR GENERATING A MAGNETIC RESONANCE IMAGE DATA SET, COMPUTER PROGRAM PRODUCT, DATA CARRIER AND MAGNETIC RESONANCE SYSTEM

Description

[0001] The invention relates to a computer-implemented method for generating a magnetic resonance image data set.

[0002] Magnetic resonance imaging has the advantage over other clinical imaging methods that it does not use any X-rays and can image soft tissue.

[0003] One initial disadvantage was the comparatively extremely long measurement times of several minutes for a single image. Spin echo sequences and gradient echo sequences were used as imaging measurement sequences.

[0004] The first step toward reducing measurement time involved optimizing and modifying the measurement sequences. To accelerate the gradient echo sequence, for example, the flip angle of the excitation pulse was reduced. This also significantly shortened the repetition time (TR), reducing the measurement time for one image to just a few seconds. This sequence is known by the acronym FLASH.

[0005] The spin echo sequence was accelerated by repeatedly applying the refocusing pulse and subsequently reading an echo signal. This sequence is known by the acronyms FSE, TSE, or RARE.

[0006] Further acceleration by modifying the measurement sequences alone is limited by the Nyquist theorem. This states that the coding in the direction of a dimension to be represented must satisfy the following condition: Δk < 2 π / L .

[0007] Here, Δk is the distance between the signals in k-space and L is the extension of the object in the dimension to be displayed.

[0008] If you reverse the equation you get the imageable area, also called field of view: FoV = 2 π / Δk .

[0009] If you want to achieve a specific resolution, for example, in the x-direction Δx, you need to acquire corresponding data, especially by applying a specific maximum gradient moment G x . If fewer data points are acquired, the resolution decreases. If G x is exceeded, which results in an increase in Δk, aliasing artifacts occur.

[0010] The Nyquist theorem only requires a band limit for the signal to be sampled. No other prior knowledge is assumed or used.

[0011] From the acquired signals S(ω) in k-space, a signal or image S(t) in the position space can be determined by a Fourier transformation FT: S t = 1 / 2 π ∫ S ω e iωt dω .

[0012] This can also be expressed as: S t = FT ⋅ S ω .

[0013] Prior knowledge can be used in several ways to circumvent the limitation imposed by the Nyquist theorem.

[0014] A first approach is parallel imaging. This has two components. First, an undersampled data set is acquired, but with multiple coils.

[0015] This results in an image dataset with deconvolutions for each coil. Using the coil's sensitivity as prior knowledge, these images can be reconstructed into a deconvolution-free image dataset. With SENSE-based reconstruction methods, deconvolution takes place in image space, while with GRAPPA-based reconstruction methods, deconvolution takes place in k-space.

[0016] The first part of parallel imaging consists of the specific data acquisition, the second part of a special reconstruction process.

[0017] Another way to use prior knowledge and thus circumvent the limitations of the Nyquist theorem is so-called compressed sensing. This technique originates from information theory and was developed for photographic image datasets. Exemplary embodiments of methods using compressed sensing are described in the publications US 8879852 B2, US 9784812 B2, DE 102013213940 B3, US 20160146915 A1, US 9396562 B2, and US 8275294 B2.

[0018] The basic idea stems from the fact that when compressing uncompressed images, the transformation allows the image data to be represented with few coefficients. JPEG2000 compression uses the wavelet transform.

[0019] However, if you ultimately don't need a portion of the recorded data at all, the question naturally arises as to whether you can skip recording that portion of the data altogether. This is possible, provided certain conditions are met.

[0020] One prerequisite is that the signal is "sparse." This "sparsity" can be present from the outset. A well-known example is angiography datasets, where the vessels occupy only a few pixels of the image, and the other pixels, even if they do not exclusively contain noise, contain no relevant information.

[0021] Sparsity can also be generated through a transformation. For many problems, only certain image regions or edges are important. These, too, occupy only a few image elements.

[0022] In compressed sensing, only a portion of the k-space lines is acquired. Instead of a full sample S(ω), only a subsample S u (ω) is obtained, and instead of a quadratic Fourier matrix FT, a partial matrix P is obtained: S u ω = P ⋅ S t .

[0023] The inversion of equation (2) therefore does not lead to a unique solution like equation (1) but rather a solution space U is spanned.

[0024] To select a solution in this solution space, one can use an optimization method. One possibility is to select the solution with the minimum energy under the Euclidean l 2 norm from the solution space. This corresponds to a minimization: min S t 2 unter P ⋅ S t = S u ω S t

[0025] To simplify the calculation, the l 1 norm can also be used: min S t 1 unter P ⋅ S t = S u ω S t

[0026] Since real data has noise signal, the condition is changed as follows: min S t 1 unter P ⋅ S t − S u ω 2 ≤ ε S t

[0027] If equation (3) is written as a real equation and not as an inequality and in Lagrange form, we get: argmin P ⋅ S t − S u ω 2 2 + λ S t 1 S t

[0028] The parameter λ is called the regularization parameter and determines the ratio of data consistency to sparsity. If it is too high, the noise signal is removed, but too much image signal is also removed. If it is too small, hardly any noise signal is removed from the image.

[0029] The noise arises from the sampling method. With compressed sensing, the k-space lines are acquired incoherently, i.e., randomly or pseudorandomly. This means that the k-space lines are acquired at irregular intervals. This prevents the formation of convolutions, but the SNR decreases.

[0030] This noise generated by the subsampling and the acquisition strategy is then removed when the image data set is generated.

[0031] Setting the regularization parameter is a process that requires considerable experience. The correct λ also depends, for example, on the sparsity, i.e., the amount of information in the image.

[0032] From HUANG F, CHEN Y: "Self-adjusted regularization ratio for robust compressed sensing", PROCEEDINGS OF THE INTERNATIONAL SOCIETY FOR MAGNETIC RESONANCE IN MEDICINE, April 18, 2009 (2009-04-18), page 4592, XP040611823 describes a reconstruction method in which a regularization term 2 λσ 2< reduces the dependence of the data reconstruction on the regularization parameter λ.

[0033] JAN AELTERMAN ET AL: "Automatic High-Bandwidth Calibration and Reconstruction of Arbitrarily Sampled Parallel MRI," PLOS ONE, Vol. 9, No. 6, June 10, 2014 (2014-06-10), page e98937, XP055526936, DOI: 10.1371 / journal.pone.0098937 discloses an MR reconstruction algorithm for datasets acquired using parallel imaging. The reconstruction uses an iteration approach similar to compressed sensing.

[0034] KEDAR KHARE ET AL: "Accelerated MR imaging using compressive sensing with no free parameters: Accelerated MR Imaging using Compressive Sensing", MAGNETIC RESONANCE IN MEDICINE, Vol. 68, No. 5, January 20, 2012 (2012-01-20), pages 1450-1457, XP055525216, US ISSN: 0740-3194, DOI: 10.1002 / mrm.24143) presents a new method for the automatic reconstruction of subsampled data. A soft threshold is used.

[0035] KIEREN GRANT HOLLINGSWORTH: "Reducing acquisition time in clinical MRI by data undersampling and compressed sensing reconstruction", PHYSICS IN MEDICINE AND BIOLOGY, INSTITUTE OF PHYSICS PUBLISHING, BRISTOL GB, Vol. 60, No. 21, 8 October 2015 (2015-10-08), XP020290393, ISSN: 0031-9155 describes methods combining parallel imaging and compressed sensing.

[0036] It is therefore an object of the present invention to provide a method for generating a magnetic resonance image data set that allows a simplified compressed sensing reconstruction.

[0037] This task is solved by a method for generating a magnetic resonance image data set with the steps: Providing a raw data set, wherein the raw data set has been recorded in a temporally and / or spatially subsampled manner, providing measurement signals of a phase correction measurement, automated determination of a regularization parameter, and generating an image data set from the raw data set using the regularization parameter in a compressed sensing method, characterized in that the regularization parameter is determined from at least two echo signals and the echo signals from the measurement signals of the phase correction measurement are used, the recording of the raw data set and the phase correction measurement are carried out with the aid of a TSE sequence.

[0038] As is well known, magnetic resonance imaging does not record image elements directly, as is the case with photography, but rather echo signals or FIDs. These are also referred to as raw data sets. Image data sets can then be obtained from a raw data set using known reconstruction methods.

[0039] For example, in a turbo spin echo (TSE) scan, the echo signals must be re-sorted before they can be Fourier transformed. For non-Cartesian sampling of k-space, regridding can be performed.

[0040] With pseudorandomized sampling of k-space, a compressed sensing reconstruction method can be used to reconstruct an image dataset from the raw dataset or an intermediate image dataset. Temporal undersampling occurs when multiple image datasets represent different points in time of a movement or state, omitting measurement data required to satisfy the Nyquist theorem. Spatial undersampling occurs when there is insufficient k-space data available for a single image dataset.

[0041] Undersampling can also occur in both temporal and spatial dimensions simultaneously.

[0042] A compressed sensing reconstruction method typically includes two components: transform sparsity and a nonlinear, iterative reconstruction. In the transform sparsity step, the raw data set or an intermediate image data set is converted into a representation in which the relevant signal is sparse.

[0043] The noise signal can then be suppressed in the nonlinear reconstruction process. The higher the selected threshold value is set, the more noise signal but also relevant signal is removed. This can be controlled via the regularization parameter λ. To ensure that the compressed sensing reconstruction process operates error-free and optimally, the regularization parameter λ is determined automatically. "Automatically determined" means that the numerical value of the regularization parameter does not have to be entered completely freely. A completely free input, as known from the state of the art, is, for example, a numeric field in which the value of the regularization parameter is to be entered. Alternatively, a completely free input can be made using a slider, where only a lower and an upper maximum value limit the input.

[0044] In the case of automated detection, a user may still have limited influence, but this is not necessary.

[0045] The determination of the regularization parameter is therefore carried out at least partially or entirely in a control device. The described equations are stored in a data memory and can be used by software to calculate the regularization parameter.

[0046] All features described in the prior art with regard to compressed sensing also apply to the method according to the invention, with the exception of the definition of the regularization parameter λ. In particular, the k-space lines of the raw data set can be acquired incoherently, i.e., randomly or pseudorandomly.

[0047] The regularization parameter λ is determined from at least two echo signals. The echo signals can generally originate from a reference measurement or the raw data set. Reference measurements are measurements used to determine measurement parameters or correction factors.

[0048] According to the invention, the echo signals from a phase correction measurement are used. This measurement is performed anyway and therefore does not require any additional effort.

[0049] Advantageously, exactly two echo signals are used.

[0050] Alternatively, exactly three echo signals are used. Surprisingly, it turns out that the regularization parameter λ can be determined with a small number of echo signals.

[0051] When using multiple receiving coils, it must be specified that two or three measurement signals are used per receiving coil.

[0052] Preferably, a signal amplitude of the echo signal can be used in each case. In particular, the maximum signal amplitude of the echo signal can be used to determine the regularization parameter. This is usually located in the center of the echo signal. The signal amplitude in the center of k-space can also be used. Alternatively, another reference value of the echo signals can also be used. For example, the area under the echo signal, the mean value of all signal amplitudes of the echo signal, whereby the magnitude value is used in each case, or the mean value of a certain number of signal amplitudes from the center of the echo signal can also be used.

[0053] The echo signals are recorded in an echo train. This can be recorded either without phase gradients and thus without phase encoding. In this case, the echo signals reflect the T2 decay. Alternatively, a phase gradient curve can be used. This preferably corresponds to the phase gradient curve used when acquiring the raw data set.

[0054] According to the invention, a TSE phase gradient curve is used. In TSE sequences, the sampling of k-space can follow different patterns, even with Cartesian sampling. In a first embodiment, k-space can be sampled "centered." In this case, k-space is always sampled from the center outwards. Alternatively, k-space can be sampled "linearly." In this case, the sampling begins at the edge of k-space and moves back outwards via the center. Alternatively, the sampling can begin between a first edge and the center and extend beyond the other edge, containing the last echo signals of the echo train at the first edge.

[0055] Preferably, the arithmetic mean can be determined from at least two echo signals or their reference value.

[0056] Advantageously, at least two echo signals or their reference values ​​can be added together. Addition here is to be understood mathematically and includes both summation and difference value calculation.

[0057] In one embodiment, the average value can be calculated from the maximum signal amplitudes of the echo signals at the beginning and end of the echo train. The largest signal amplitude, i.e., the echo signal from the center of k-space, is added to this average value. If I max_A is the signal amplitude of the first echo in the echo train, I max_B is the signal amplitude of the last echo in the echo train, and I max_M is the signal amplitude of the echo with the largest signal amplitude in the echo train, then the regularization parameter λ is: λ = λ I max _ M + 1 / 2 I max _ A + I max _ B .

[0058] The regularization parameter is determined as a function of, or in other words, as a function of, I max_M , I max_A , and I max_B , where the mean of I max_A and I max_B is taken and added positively to I max_M . In other words, the regularization parameter is a function of I max_M , I max_A , and I max_B . This definition of the regularization parameter is used when using a turbo spin echo as the imaging sequence.

[0059] Preferably, the echo signals can be recorded with at least two different coils. Then, the regularization parameter can be additionally or alternatively weighted depending on the recording coil: λ coil = λ I max_coil .

[0060] This can be combined with the first embodiment, for example, by acquiring the echo train with each coil and calculating the regularization parameter λ for each coil as specified in equation (4). Depending on the design of the reconstruction method, the respective regularization parameter λ coil can be used for each raw data set acquired with a coil. If the coils are used for parallel imaging and a deconvolved image is calculated, an overall regularization parameter can also be determined for a number of n coils: λ coil = 1 / n ⋅ α 1 λ coil_ 1 + α 2 λ coil_ 2 + … + α n λ coil_n .

[0061] The coefficients α are weighting factors that reflect the input of the individual coil signals into the overall image.

[0062] Alternatively, an overall regularization parameter can be obtained by calculating a root mean square of the signal amplitudes: I coil_gesamt = SQRT I max_coil_ 1 2 + I max_coil_ 2 2 + … + I max_coil_n 2 and the total regularization parameter λ coil is then determined as a function of the mean value I coil_total: λ coil = λ I coil gesamt .

[0063] Alternatively or additionally, at least one of the echo signals can be recorded using a preparation module. The regularization parameter is then λ = λ I ohne − I mit .

[0064] In this negative addition, the maximum signal intensities or another reference value of the echo signals are acquired once with the preparation module and once without the preparation module. The measurement without the preparation module generates the maximum signal intensity I without . The maximum signal intensity of the echo signal acquired with the preceding preparation module is called I with . The measurement data I without can originate from a reference measurement without the preparation module or from the measurement of the image data.

[0065] Alternatively, the ratio of a measurement with and without preparation module can be used: λ = λ I ohne / I mit .

[0066] The inclusion of a preparation module can be combined with both the acquisition of an echo train and the use of multiple coils. Thus, the signal intensity determined in Equation (7) can be determined once with and once without a preparation module, and these values ​​can then be used in Equation (9).

[0067] As a further refinement, a control panel can be displayed to the user. Several settings can be displayed as a selection list. The options "sharp," "medium," and "smooth" can be selected.

[0068] Selecting the "sharp" variant then results in the automatically calculated regularization parameter λ being converted to: λ ′ = 1 / x ⋅ λ

[0069] When selecting the "medium" variant, the regularization parameter λ can be retained: λ ′ = λ

[0070] If, however, the "smooth" variant is selected, the regularization parameter λ' is set to: λ ′ = x ⋅ λ

[0071] The conversion factor x is generally freely adjustable and is preferably between 1 and 10. In this way, the user can still influence the reconstruction of the image dataset despite the automated determination of the regularization parameter λ, without having to deal with the nature of the regularization parameter or even having to know it.

[0072] Alternatively, the user can choose between a static and a dynamic measurement, which corresponds to spatial or temporal subsampling. Then, an intermediate regularization parameter can be selected using a slider, and the regularization parameter in the case of a static measurement can be determined as: λ = z ⋅ σ where z is a freely selectable, constant factor and σ is the value set on the slider. In the case of a dynamic measurement, however, the regularization parameter λ is determined as follows: λ = z ⋅ log y ⋅ σ

[0073] Here, too, y and z are freely selectable, constant factors and σ is the value set on the slider.

[0074] Preferably, the raw data set can be acquired with multiple coils. It can then be subsampled even more than would be possible with compressed sensing alone.

[0075] According to the invention, a turbo spin echo sequence is used as the measurement sequence for recording the measurement data set.

[0076] In contrast to the sequence described below, this also includes phase encoding gradients for imaging.

[0077] Preferably, k-space can be sampled Cartesian-wise when acquiring the raw data set. This is possible, for example, with spatial or static subsampling. Alternatively, k-space can be sampled radially when acquiring the raw data set. This type of subsampling is particularly advantageous for temporal subsampling. Individual spokes can be assigned to multiple raw data sets.

[0078] Alternatively, k-space can be sampled spirally when acquiring the raw data set. Spiral sampling allows for significant temporal accelerations.

[0079] Preferably, the raw data set is subsampled in spatial dimensions, i.e., statically. With this type of subsampling, the automatic specification of regularization parameters is particularly helpful.

[0080] The solution to the problem mentioned at the outset is also achieved by a computer program product or a computer program according to claim 13. In addition, the invention relates to a data carrier according to claim 14. Advantageously, the data generation unit can be an image generation unit.

[0081] The invention also relates to a magnetic resonance system with a control device. The magnetic resonance system is characterized in that the control device is designed to carry out the method as described. The implementation of the aforementioned methods in the control device can be carried out as software or as (hard-wired) hardware.

[0082] Further advantageous embodiments of the magnetic resonance system according to the invention correspond to corresponding embodiments of the method according to the invention. To avoid unnecessary repetition, reference is made to the corresponding method features and their advantages.

[0083] Further advantages, features and special features of the present invention will become apparent from the following description of advantageous embodiments of the invention.

[0084] Showing: Fig. 1 shows a magnetic resonance system, Fig. 2 shows a measurement sequence, Fig. 3 shows an echo train, Fig. 4 shows an echo signal in a first embodiment, Fig. 5 shows a first spectrum, Fig. 6 shows a second spectrum, Fig. 7 shows an echo signal in a second embodiment, and Fig. 8 shows a control panel.

[0085] Figure 1shows a magnetic resonance system 1 with a transmit coil arrangement 2. The transmit coil arrangement 2 can be configured as a body coil. However, it can also be a transmit coil array. The transmit coil arrangement 2 is shown in dashed lines.

[0086] For data acquisition, the magnetic resonance system 1 has a receiving coil arrangement 3. The receiving coil arrangement 3 is preferably a coil array with coils 4, 5, 6 and 7. The coils 4, 5, 6 and 7 read out measurement signals simultaneously and thus in parallel.

[0087] To control the experiments, the magnetic resonance system 1 has a control device 8.

[0088] The magnetic resonance system 1 further comprises a data carrier 9 as part of the control device 8 or independently thereof, on which computer programs 10 for carrying out magnetic resonance measurements are stored.

[0089] Other components of the magnetic resonance system 1, such as gradient coils or a patient bed, are not shown for the sake of clarity.

[0090] Figure 2 shows a TSE measurement sequence diagram 11. The turbo spin echo sequence is used to acquire measurement signals for phase correction. The echo train with a number of NE echoes is acquired without phase encoding and thus without gradients in the phase encoding direction GP.

[0091] ACQ denotes the axis for the radiofrequency pulses and the acquisition windows. The radiofrequency pulse 12 is typically a 90° pulse, and the refocusing pulse 13 is a 180° pulse. These generate the echo signals 14.

[0092] In the reading direction GR, a reading dephasing gradient 15 and reading gradient 16 are present.

[0093] In the slice selection direction GS, a slice selection gradient 17 is applied simultaneously with the radiofrequency pulse 12, followed by a slice rephasing gradient 18. The slice rephasing gradient 18 typically has half the gradient moment compared to the slice selection gradient 17.

[0094] Simultaneously with the refocusing pulse 13, the slice selection gradient 19 is applied. This gradient is surrounded by two spoiler gradients. The spoiler gradients 20 serve to destroy signal components that may arise due to imperfections in the refocusing pulse 13.

[0095] A preparation module 21 can optionally be used before the high-frequency pulse 12.

[0096] Figure 3 shows an echo train 22, as it is connected to the TSE sequence Figure 2 can be obtained. The echo train 22 comprises, purely as an example, four echo signals 14a, 14b, 14c and 14d.

[0097] The signal intensity or amplitude I max_A of the first echo signal 14a is indicated by arrow 23, and the signal amplitude I max_ of the last echo signal 14d is indicated by arrow 26. Arrows 24 and 25 indicate the signal amplitudes of the echo signals 14b and 14c. Arrow 24 indicates the signal amplitude I max_M, since this is, by definition, the largest signal amplitude in the echo train 22.

[0098] Based on the TSE measurement sequence according to Figure 2 the regularization parameter λ can therefore be calculated as given in formula (4): λ = λ I max _ M + 1 / 2 I max _ A + I max _ B .

[0099] Figure 4 shows the echo signal 14b enlarged. This is intended to display several reference values ​​derived from an echo signal.

[0100] As already described above, the maximum amplitude I max of the echo signal 14b can be used. Alternatively, the area of ​​the echo signal component 27 can also be considered. Alternatively, other signal components such as signal components 28 and 29 can also be included. In general, the integral of a portion of the echo signal 14b can be used.

[0101] The reference value of the echo signals 14a, 14b, 14c and 14d can be freely selected per se, but must be selected the same for all echo signals 14a, 14b, 14c and 14d.

[0102] The Figure 5 and 6 show echo signals 30 and 31. The echo signal 30 was recorded with coil 5, for example, and the echo signal 31 with coil 6. The echo signals can then be calculated to form a total signal as given in formula (7): I gesamt = SQRT I max _ coil _ 4 2 + I max _ coil _ 5 2 + I max _ coil _ 6 2 + I max _ coil _ 7 2

[0103] This is how to Figure 1described purely as an example, four receiving coils 4, 5, 6 and 7 are assumed. These also specify the indices.

[0104] Instead of individual echo signals 30 and 31, an echo train 22 can also be measured with each coil 4, 5, 6, and 7, and a reference amplitude can be determined for each coil analogous to equation (4). This reference amplitude can then be considered in equation (7).

[0105] Figure 7 shows an echo signal 14e, which is comparable in terms of recording to echo signal 14b. In contrast, the preparation module 21 was applied before generating echo signal 14e. This reduced the signal amplitude I max , represented by arrow 32. The regularization parameter λ can be determined as given in equation (9).

[0106] Figure 8shows part of a control panel 33. This contains a dropdown menu 34 for modifying the regularization parameter λ. Depending on which of the selection fields 35, 36, or 37 is selected, the regularization parameter λ is modified according to one of the equations (11) to (13).

[0107] In addition, a slider 38 may also be provided for making a basic setting of the regularization parameter λ.

Claims

1. Computer-implemented method for generating a magnetic resonance image dataset, comprising the steps: - providing a raw dataset, wherein the raw dataset has been acquired such that it is spatially and / or temporally undersampled, - providing measurement signals of a phase correction measurement, - determining a regularisation parameter (λ) in an automated manner, and - generating an image dataset from the raw dataset using the regularisation parameter (λ) in a compressed sensing technique, characterised in that - the regularisation parameter (λ) is determined from at least two echo signals (14a, 14b, 14c, 14d, 14e), wherein the echo signals from the measurement signals of the phase correction measurement are used, - the raw dataset and the phase correction measurement are acquired with the aid of a TSE sequence.

2. Method according to claim 1, characterised in that the echo signals (14a, 14b, 14c, 14d, 14e) have been acquired in an echo train (22).

3. Method according to one of claims 1 or 2, characterised in that the maximum signal amplitude (Imax_M, Imax_A, Imax_B) of the echo signals (14a, 14b, 14c, 14d, 14e) is used to determine the regularisation parameter (λ).

4. Method according to one of the preceding claims, characterised in that the echo signals (14a, 14b, 14c, 14d, 14e) have been acquired using at least two different coils (4, 5, 6, 7).

5. Method according to one of the preceding claims, characterised in that the regularisation parameter (λ) is determined by adding signal amplitudes (Imax_M, Imax_A, Imax_B).

6. Method according to one of the preceding claims, characterised in that the regularisation parameter (λ) is determined by dividing two signal amplitudes (Iwithout, Iwith).

7. Method according to one of the preceding claims, characterised in that at least one of the echo signals (14e), in particular precisely one, has been acquired using a preparatory module (21).

8. Method according to one of the preceding claims, characterised in that the regularisation parameter (λ) is determined by forming the arithmetic mean of signal amplitudes (Imax_A, Imax_B).

9. Method according to one of the preceding claims, characterised in that the regularisation parameter (λ) is determined by forming the root mean square of signal amplitudes (Imax_coil_1, Imax_coil_2, Imax_coil_n) of the echo signals (14a, 14b, 14c, 14d, 14e).

10. Method according to one of the preceding claims, characterised in that the echo signals (14a, 14b, 14c, 14d, 14e) have been acquired without a phase encoding gradient.

11. Method according to one of the preceding claims, characterised in that the regularisation parameter (λ) can be changed by means of reconstruction presets that can be selected in selection fields (35, 36, 37).

12. Method according to one of the preceding claims, characterised in that the compressed sensing technique comprises a non-linear iterative reconstruction.

13. Computer program product (10) for a controller (8) for controlling a data generating unit, in particular an image generating unit, of a magnetic resonance system (1) comprising commands which, when the program is executed by the controller, cause this to perform a method according to one of the preceding claims.

14. Computer-readable data storage medium (9) for a controller (8) for controlling a data generating unit, in particular image generating unit, of a magnetic resonance system (1), on which the computer program product according to claim 13 is stored.

15. Magnetic resonance system (1) comprising a controller (8), characterised in that the controller (8) is designed to implement a method according to one of claims 1 to 12.