Method and system for magnetic resonance imaging processing

Abstract

According to the invention, a method for processing acquired magnetic resonance images is provided, comprising the following steps: S1) providing a first magnetic resonance image, S2) providing a second magnetic resonance image, S3) providing a first operator, S4) providing a second operator, s5) applying a second operator to the first magnetic resonance image and applying the first operator to a magnetic resonance image that has been received by applying the second operator to the first magnetic resonance image, a magnetic resonance image expected in the case where the first magnetic resonance image has been acquired in the first main magnetic field under shimming conditions for shimming the first main magnetic field to be uniform is retrieved instead of the first magnetic resonance image.

Description

Technical Field

[0001] The present invention relates to the field of magnetic resonance imaging. More specifically, the present invention relates to the field of processing magnetic resonance images acquired using a magnetic resonance imaging device. Background Art

[0002] High-temperature superconductors enable the magnet to tilt very quickly, which in turn enables the construction of a magnetic resonance imaging device with a switchable magnetic flux density of the main magnetic field. This allows the internal examination magnetic flux density of the main magnetic field to be switched to another magnetic flux density of another main magnetic field of the magnetic resonance imaging device. For example, imaging is performed under a 1.5 T main magnetic field. Subsequent switching of the high-temperature superconductor will cause the magnetic flux density to change to another main magnetic field of 0.6 T and enables additional magnetic resonance images to be acquired at this lower magnetic flux density. However, shimming is still set so that the main magnetic field at, for example, 1.5 T is uniform. The additional main magnetic field shows non-uniformity due to the change in magnetic flux density because shimming is not performed for magnetic flux densities different from the first magnetic flux density.

[0003] Generally, shimming is a process of providing the maximum possible homogenization of the main magnetic field. A distinction is made between active and passive shimming. Passive shimming is performed by placing and positioning ferromagnetic materials inside the magnet that forms the main magnetic field and is used to minimize the non-uniformity caused by the design of the magnetic resonance imaging device. Active shimming can be performed using shim coils provided for this purpose. Active shimming homogenizes the non-uniformity in the main magnetic field caused by the patient load.

[0004] The non-uniformity caused by the change in the magnetic flux density of the main magnetic field results in image artifacts in the magnetic resonance imaging device, which cannot be sufficiently reduced by active shimming.

[0005] The article "Feasibility study of novel rapid ramp-down procedure in MgB2 MRI magnet using persistent current switch with high off-resistivity" by Kodama et al. (Superconductor Science and Technology, 34 (2021) 074003 (13 pp)) describes a dry magnet with high-temperature superconductors and MgB2, which is provided with a novel rapid shutdown method that can replace the controlled shutdown in an emergency. For this purpose, a power supply is established by means of a persistent current switch, which turns off when heated, and the power supply is interrupted by means of a circuit breaker. The energy stored in the solenoid is consumed at an external resistor.

[0006] The accurate homogenization of the main magnetic field is crucial for reducing inhomogeneous artifacts in magnetic resonance images in magnetic resonance imaging. However, with existing tiltable magnetic resonance imaging scanners and the ability to vary the main magnetic field, the homogenization of the main magnetic field by shimming is insufficient.

[0007] According to US Patent Application US2020 / 0400764, a magnetic resonance imaging system is known that is configured for imaging near metal implants, and in which the magnetic field can be tilted within a clinically acceptable time. Summary of the Invention

[0008] The object of the present invention is to provide a time- and memory-efficient method for reducing the influence of magnetic field inhomogeneities in magnetic resonance images.

[0009] According to the present invention, this object is solved by the subject matter of the independent claims. Preferred embodiments of the present invention are described in the dependent claims.

[0010] Thus, according to the present invention, a method for processing acquired magnetic resonance images is provided. The method comprises the steps of: providing a first magnetic resonance image that has been acquired using a magnetic resonance imaging device under shimming conditions for homogenizing a second main magnetic field to uniformity, in a first main magnetic field having a first magnetic flux density, wherein the second main magnetic field has a second magnetic flux density different from the first magnetic flux density; providing a second magnetic resonance image that has been acquired using the magnetic resonance imaging device under shimming conditions for homogenizing the second main magnetic field to uniformity, in the second main magnetic field; providing a first operator for transforming a magnetic resonance image that has been acquired using the magnetic resonance imaging device under shimming conditions for homogenizing the first main magnetic field to uniformity, in the first main magnetic field, into a magnetic resonance image expected under shimming conditions for homogenizing the second main magnetic field to uniformity, in the second magnetic field; providing a second operator for transforming a magnetic resonance image that has been acquired using the magnetic resonance imaging device under shimming conditions for homogenizing the second main magnetic field to uniformity, in the first main magnetic field, into a magnetic resonance image expected under shimming conditions for homogenizing the first main magnetic field to uniformity, in the first magnetic field; retrieving the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under shimming conditions for homogenizing the first main magnetic field to uniformity, in the first main magnetic field, rather than the first magnetic resonance image, by applying the second operator to the first magnetic resonance image and applying the first operator to the magnetic resonance image received after applying the second operator to the first magnetic resonance image.

[0011] An operator is typically a mapping or function that acts on elements of a space to produce elements of another space. The term can be used as a synonym for a function or for applying a machine learning algorithm to elements of a space. The actual elements in this space are a first magnetic resonance image and a second magnetic resonance image having respective first and second operators.

[0012] The term "shimming the main magnetic field to homogeneity" refers to applying such shimming, where the image distortion caused by field inhomogeneity is reduced. However, while such shimming can help greatly reduce field inhomogeneity, such shimming will never be able to achieve a perfectly homogeneous main magnetic field.

[0013] Furthermore, it seems that in practice it will be impossible to generate an operator that works perfectly, i.e., an operator that can accurately calculate the expected image based on a given image. If this is the case, then the second operator alone will be sufficient to reconstruct the correctly shimmed image from an image with poor shimming, which is an image that has already been acquired under shimming conditions for another main magnetic field with a higher or lower magnetic flux density. Thus, according to the present invention, the second operator is used together with the first operator, the first magnetic resonance image, and the second magnetic resonance image in order to reconstruct the magnetic resonance image that would be expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the correct shimming conditions for shimming the first main magnetic field to homogeneity, rather than the first magnetic resonance image. Since a ground truth formed by the second magnetic resonance image is provided (i.e., acquired), which has been acquired using the magnetic resonance imaging device under shimming conditions for shimming the second main magnetic field to homogeneity, the operation of the first operator provides access to the consistency of the operation of the second operator, which can be corrected or adapted accordingly. The operation of the first operator can also be used as a constraint on the operation of the second operator to improve the consistency of the result of the action of the first operator.

[0014] This allows for a time-efficient and resource-saving method to unlock all the functions of a magnetic resonance imaging device including switchable magnetic flux density. The first magnetic resonance image provides a time-efficient method using the transformation of two operators to examine the interior of a patient in a magnetic resonance imaging device switchable between two different modes (two magnetic flux densities) without adjusting passive or active shimming between the two magnetic flux densities. In particular, passive shimming, which involves changing the main magnetic field by introducing ferromagnetic materials to achieve homogenization of the magnetic field, is a resource-intensive shimming method. Using operators to transform magnetic resonance images enables new magnetic resonance imaging possibilities, such as reducing susceptibility artifacts by magnetic resonance image acquisition at different magnetic flux densities. Additionally, by forming the operators, computationally and memory-efficient methods are specifically selected. Once the operators have been constructed, the training data of the magnetic resonance imaging device from which the first magnetic resonance image and the second magnetic resonance image are derived are no longer required, thus achieving a synergistic effect with the first and second operators since the first operator does not necessarily have to be locally generated at the magnetic resonance imaging device under discussion.

[0015] According to the invention, various algorithms can be used for the last step of applying the operators. However, according to a preferred embodiment of the invention, the method for performing the last step includes a trained machine learning algorithm representing the actions of the first operator and the second operator, which, when inputting the first magnetic resonance image and the second magnetic resonance image, generates the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the first main magnetic field in a shimming condition for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image.

[0016] In particular, this means that when applying the first operator, the intermediate result may not necessarily be available to the user of the system but remains within the trained machine learning algorithm, and only the final result is accessible. Thus, these operators can be execution steps that cannot be accessed within the trained machine learning algorithm. For example, in a trained artificial neural network, these operators can be the activation functions in the corresponding-level nodes of the trained artificial neural network.

[0017] In principle, the training dataset may include various data. However, according to a preferred embodiment of the invention, the method further includes a training dataset for developing the first operator in the trained machine learning algorithm, the training dataset including magnetic resonance images acquired using the first main magnetic field under a shimming condition for shimming the first main magnetic field to uniformity, and magnetic resonance images acquired under the second main magnetic field in a shimming condition for shimming the second main magnetic field to uniformity.

[0018] Magnetic resonance images acquired using a first main magnetic field under shimming conditions for shimming the first main magnetic field to homogeneity and magnetic resonance images acquired under shimming conditions for shimming a second main magnetic field to homogeneity and under the second main magnetic field are fed in pairs, thereby constructing a training dataset into a machine learning algorithm. To this end, in-situ and in-vivo magnetic resonance images of a humanoid object, an animal object, or a special magnetic resonance imaging phantom intended to replace in-vivo magnetic resonance imaging can be acquired at different magnetic flux densities. The training dataset can include corresponding magnitude images as well as phase images.

[0019] Generally, a training dataset can be generated for a first operator in various ways. However, according to a preferred embodiment of the present invention, the method further includes using a generative adversarial network including two artificial neural networks to generate a training dataset for developing a first operator in a trained machine learning algorithm.

[0020] Magnetic resonance images acquired using a first main magnetic field under shimming conditions for shimming the first main magnetic field to homogeneity and magnetic resonance images acquired under shimming conditions for shimming a second main magnetic field to homogeneity and under the second main magnetic field can be generated by employing a generative adversarial network including two artificial neural networks to generate training data for a first operator without additional in-vivo or in-situ acquisition of magnetic resonance images.

[0021] According to a preferred embodiment of the present invention, the method further includes using electromagnetic simulations to generate a training dataset for developing a first operator in a trained machine learning algorithm.

[0022] Using these electromagnetic simulations, for example, Maxwell's equations for a given problem can be solved in a simulation environment to correspondingly derive a first operator.

[0023] A training dataset can be generated for a second operator in various ways. However, according to a preferred embodiment of the present invention, the method further includes: the training dataset for developing a second operator in a trained machine learning algorithm includes magnetic resonance images of a magnetic resonance imaging phantom acquired using a first main magnetic field under shimming conditions for shimming the second main magnetic field to homogeneity, and magnetic resonance images of the magnetic resonance imaging phantom acquired under shimming conditions for shimming the second main magnetic field to homogeneity and under the second main magnetic field.

[0024] Therefore, the training dataset contains the difference between a first magnetic resonance image and a second magnetic resonance image and thus a direct measure of the magnetic field inhomogeneity generated by the change in magnetic flux density. A machine learning algorithm can be trained using a second operator to compensate for these inhomogeneities. This effect is specific to the magnetic resonance imaging device according to the magnetic resonance imaging device, wherein the first magnetic resonance image and the second magnetic resonance image are derived from the magnetic resonance imaging device.

[0025] Preferably, the trained machine learning algorithm for the second operator is a trained artificial neural network.

[0026] In principle, different methods can be provided to approximate the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the first main magnetic field under the shimming condition for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image. However, according to a preferred embodiment of the present invention, the method further includes: in a final step, by minimizing a loss function, retrieving the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the first main magnetic field under the shimming condition for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image, the loss function using the first magnetic resonance image and the second magnetic resonance image as well as the first operator and the second operator.

[0027] In the case of minimizing the loss function, the first magnetic resonance image converges to the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the first main magnetic field under the shimming condition for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image. This is a process that can never be perfect, and thus iterative asymptotic approximation is regularly performed. Therefore, it can be stipulated that when a certain iteration threshold that can be defined by the user is reached, the loss function stops calculating the magnetic resonance image expected in the case where the magnetic resonance imaging device has acquired the first magnetic resonance image under the first main magnetic field under the shimming condition for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image. Similarly, the minimization of the loss function may result in a local minimum rather than a global minimum. However, basically, the artifacts in the expected magnetic resonance image are reduced compared to the first magnetic resonance image.

[0028] Different embodiments for minimizing the loss function can be adopted. However, according to a preferred embodiment of the present invention, the method further includes the following formula applicable to the loss function LF:

[0029] LF = operator1(operator2(image1)) - image2,

[0030] where operator1 is the first operator, operator2 is the second operator, image1 is the first image and image2 is the second image.

[0031] The loss function iteratively minimizes the distance between the second magnetic resonance image and the first magnetic resonance image to which the first operator and the second operator are applied. In this minimization of the loss function, the first operator and the second operator are input as initial values, which can be changed during the iterative minimization using the loss function.

[0032] Generally, different magnetic resonance imaging devices can be used to acquire magnetic resonance images for a first operator. However, according to a preferred embodiment of the present invention, the method further comprises: obtaining a first operator by means of magnetic resonance image characteristics of magnetic resonance images acquired under a first main magnetic field including shimming conditions for shimming the first main magnetic field to uniformity, and magnetic resonance image characteristics of magnetic resonance images acquired under a second main magnetic field including shimming conditions for shimming the second main magnetic field to uniformity, wherein the magnetic resonance images acquired under the first main magnetic field and the second main magnetic field respectively have not been acquired by means of the magnetic resonance imaging device, and the first magnetic resonance image and the second magnetic resonance image are derived from the magnetic resonance imaging device.

[0033] Therefore, the first operator is independent of the magnetic resonance imaging device from which the first magnetic resonance image and the second magnetic resonance image are derived, and it is not necessary to determine the operator for each specific magnetic resonance imaging device. That is, the first operator only needs to be determined once.

[0034] According to a preferred embodiment of the present invention, the method further comprises obtaining a second operator by means of magnetic resonance images acquired from a magnetic resonance imaging phantom simulating a patient load under a first main magnetic field and a second main magnetic field, and the difference between the magnetic resonance images of the magnetic resonance imaging phantom acquired under the first main magnetic field and the second main magnetic field includes the influence of the non-uniformity of the first main magnetic field having a first magnetic flux density under shimming conditions for shimming the second main magnetic field to uniformity.

[0035] Furthermore, according to the present invention, there is provided a computer program for magnetic resonance imaging modification, which comprises instructions that, when the program is executed by a computer, cause the computer to execute the method as described above.

[0036] The present invention also provides a system of a magnetic resonance imaging device and a computing unit, the magnetic resonance imaging device having a switch by means of which the main magnetic field is switchable, wherein the first main magnetic field includes a first magnetic flux density having shimming conditions for shimming the second main magnetic field to uniformity, wherein the second main magnetic field includes a second magnetic flux density having shimming conditions for shimming the second main magnetic field to uniformity, wherein the second magnetic flux density is different from the first magnetic flux density, and wherein the computing unit is adapted to execute the method as described above.

[0037] By allowing the magnetic flux density to change to a lower or higher flux density, artifacts in magnetic resonance images depending on the magnetic flux density can be compensated for.

[0038] In principle, different superconducting magnets can be employed. However, according to a preferred embodiment of the present invention, a superconducting magnet is provided for generating the first main magnetic field and the second main magnetic field, and the superconducting magnet is a dry magnet.

[0039] Preferably, the superconducting magnet for generating the first main magnetic field and the second main magnetic field is a high-temperature superconductor.

[0040] The high-temperature superconductor allows the superconducting magnet to quickly switch between the first magnetic flux density and the second magnetic flux density.

[0041] Details regarding the training of (i) a first operator for converting an image under a field / homogeneity setting of one field strength to an image under a field / homogeneity setting of another field strength and (ii) a second operator for converting an image under a field setting with a non-corresponding homogeneity setting to an image under the same field setting and its corresponding homogeneity setting. These first and second operators can be trained based on training image pairs with ground truth annotations converted by the respective first and second operators into one another or based on ground truth developed by computer simulation. These operators can be implemented as neural networks trained to return a desired image from an input image. The second operator can be trained on a large set of image pairs (e.g., of a phantom under the field tilts and correction-associated homogeneity settings corresponding thereto), or by Bloch simulation. The first operator can be implemented as a trained neural network by training on images of a phantom model under one field tilt and correct / incorrect homogeneity settings. Alternatively, it is sufficient to collect data only under incorrect homogeneity settings, and based on knowledge of the encoding distortion caused by the incorrect homogeneity settings, corresponding images under equal field ramps and correct homogeneity settings can be generated.

[0042] To train the described TH- (transformation between contrasts at different field strengths) and US- (unmixing) networks, real data can be employed. As a general rule, the more data provided for training, the better the result. To improve training without having to measure more real data, artificial augmentation of the training data set by means of a GAN (generative adversarial network) can be considered. The GAN itself needs to be trained, but once trained, for example, based on a limited initial training set, it can generate an almost infinite number of TH and US feature / label pairs, which can be used to train the above-described TH- and US-networks.

[0043] Even more preferably, the superconducting magnet for generating the first main magnetic field and the second main magnetic field is a magnesium diboride magnet. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] These and other aspects of the invention will be apparent from the embodiments described below. However, such embodiments do not necessarily represent the full scope of the invention, and reference is therefore made to the claims and the present text for the purpose of interpreting the scope of the invention.

[0045] In the drawings:

[0046] Figure 1 A scheme of a method according to a preferred embodiment of the invention is schematically depicted; and

[0047] Figure 2 Schematically depicts a system of a magnetic resonance imaging device and a computing unit according to a preferred embodiment of the present invention.

[0048] List of reference numerals

[0049] 1. Magnetic resonance imaging device

[0050] 2. Computing unit

[0051] 3 Patient positioning system

[0052] 4. Treatment table

[0053] 5. Magnetic resonance imaging phantom

[0054] 6. Superconducting magnet

[0055] 7. Switch Detailed implementation manners

[0056] Figure 1 Schematically depicts a scheme of a method according to a preferred embodiment of the present invention. The method includes steps S1 to S5.

[0057] S1: First, provide a first magnetic resonance image that has been acquired using the magnetic resonance imaging device 1 under a first main magnetic field having a first magnetic flux density under shimming conditions for shimming the second main magnetic field to uniformity, wherein the second main magnetic field has a second magnetic flux density different from the first magnetic flux density.

[0058] S2: Second, provide a second magnetic resonance image that has been acquired using the magnetic resonance imaging device 1 under the second main magnetic field under shimming conditions for shimming the second main magnetic field to uniformity.

[0059] S3: Third, provide a first operator that is used to transform a magnetic resonance image that has been acquired using the magnetic resonance imaging device 1 under a first main magnetic field under shimming conditions for shimming the first main magnetic field to uniformity into a magnetic resonance image expected under the second magnetic field under shimming conditions for shimming the second main magnetic field to uniformity.

[0060] S4: Fourth, provide a second operator that is used to transform a magnetic resonance image that has been acquired using the magnetic resonance imaging device 1 under the second main magnetic field under shimming conditions for shimming the second main magnetic field to uniformity into a magnetic resonance image expected under the first magnetic field under shimming conditions for shimming the first main magnetic field to uniformity.

[0061] S5: Fifth, by applying a second operator to the first magnetic resonance image and applying a first operator to the magnetic resonance image that has been received by applying the second operator to the first magnetic resonance image, retrieve the magnetic resonance image expected in the case where the magnetic resonance imaging device 1 has acquired the first magnetic resonance image under the first main magnetic field in a shimming condition for shimming the first main magnetic field to be uniform, rather than the first magnetic resonance image.

[0062] The first operator is generated by a trained machine learning algorithm that has been trained using magnetic resonance images acquired under the first main magnetic field in a shimming condition for shimming the first main magnetic field to be uniform and magnetic resonance images acquired under the second main magnetic field in a shimming condition for shimming the second main magnetic field to be uniform. These images have not been acquired using the magnetic resonance imaging device 1, and the first magnetic resonance image and the second magnetic resonance image are derived from the magnetic resonance imaging device 1. The first operator is derived only once and is valid for a combination of a first magnetic flux density and a second magnetic flux density for a given set of scan parameters that has been determined. In the case where the scan parameters change, for example, if the repetition time (TR) or the echo time (TE) will change and thus different image contrasts will be achieved, a new determination will be required.

[0063] The second operator is also generated by a trained machine learning algorithm that has been trained using training data including magnetic resonance images of the magnetic resonance imaging phantom 5 acquired under the first main magnetic field in a shimming condition for simulating a patient load for shimming the second main magnetic field to be uniform and magnetic resonance images of the magnetic resonance imaging phantom 5 acquired under the second main magnetic field in a shimming condition for making the second main magnetic field uniform with a corresponding set of given scan parameters.

[0064] Then, the first operator and the second operator are implemented as a loss function in order to retrieve, by minimizing the loss function, the magnetic resonance image expected in the case where the magnetic resonance imaging device 1 has acquired the first magnetic resonance image under the first main magnetic field in a shimming condition for shimming the first main magnetic field to be uniform, rather than the first magnetic resonance image, the loss function using the first magnetic resonance image and the second magnetic resonance image and the first operator and the second operator.

[0065] Figure 2Schematically depicts a system of a magnetic resonance imaging device 1 and a computing unit 2 according to a preferred embodiment of the present invention. The main magnetic field in the bore of the magnetic resonance imaging device 1 is generated by a superconducting magnet 6, which is a high-temperature superconductor. The magnetic flux density of the main magnetic field generated by the superconducting magnet 6 can be switched between a first main magnetic field having a first magnetic flux density and a second main magnetic field having a second magnetic flux density different from the first magnetic flux density under a shimming condition for shimming the second main magnetic field to uniformity by a switch 7.

[0066] A magnetic resonance imaging phantom 5 is arranged on a treatment table 4 of a patient positioning system 3. The switch 7 is connected to the computing unit 2 and can be controlled by it. In addition, the computing unit 2 is adapted to perform the method of steps S1 - S5, and thus obtains an expected magnetic resonance image instead of the first magnetic resonance image in the case where the magnetic resonance imaging device has acquired a first magnetic resonance image under a first main magnetic field under a shimming condition for shimming the first main magnetic field to uniformity by applying a second operator to the first magnetic resonance image and then applying a first operator to the magnetic resonance image received after applying the second operator to the first magnetic resonance image, and then calculates the magnetic resonance image to the first magnetic resonance image according to the first operator and calculates the magnetic resonance image to the second magnetic resonance image according to the second operator.

[0067] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the present invention is not limited to the disclosed embodiments. Other variations of the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention by studying the drawings, the disclosure, and the claims. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality. The fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope. In addition, for clarity, not all elements in the drawings have reference signs.

Claims

1. A method for processing acquired magnetic resonance images, the method comprising the following steps: S1) Provide a first magnetic resonance image that has been acquired using a magnetic resonance imaging device (1) under a first main magnetic field having a first magnetic flux density under shimming conditions for shimming a second main magnetic field to uniformity, wherein the second main magnetic field has a second magnetic flux density different from the first magnetic flux density, S2) Provide a second magnetic resonance image that has been acquired using the magnetic resonance imaging device (1) under the second main magnetic field under shimming conditions for shimming the second main magnetic field to uniformity, S3) Provide a first operator for transforming a magnetic resonance image that has been acquired using the magnetic resonance imaging device (1) under the first main magnetic field under shimming conditions for shimming the first main magnetic field to uniformity into a magnetic resonance image expected under the second magnetic field under shimming conditions for shimming the second main magnetic field to uniformity, S4) Provide a second operator for transforming a magnetic resonance image that has been acquired using the magnetic resonance imaging device (1) under the first main magnetic field under shimming conditions for shimming the second main magnetic field to uniformity into a magnetic resonance image expected under the first magnetic field under shimming conditions for shimming the first main magnetic field to uniformity, S5) Retrieve the magnetic resonance image expected in the case where the magnetic resonance imaging device (1) has acquired the first magnetic resonance image under the first main magnetic field under shimming conditions for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image, by applying the second operator to the first magnetic resonance image and applying the first operator to the magnetic resonance image received after applying the second operator to the first magnetic resonance image.

2. The method according to claim 1, wherein, To perform step S5, provide a trained machine learning algorithm representing the actions of the first operator and the second operator, which, when input with the first magnetic resonance image and the second magnetic resonance image, generates the magnetic resonance image expected in the case where the magnetic resonance imaging device (1) has acquired the first magnetic resonance image under the first main magnetic field under shimming conditions for shimming the first main magnetic field to uniformity, rather than the first magnetic resonance image.

3. The method according to claim 2, wherein, The training data set for developing the first operator in the trained machine learning algorithm includes the magnetic resonance images acquired using the first main magnetic field under shimming conditions for shimming the first main magnetic field to uniformity, and the magnetic resonance images acquired under the second main magnetic field under shimming conditions for shimming the second main magnetic field to uniformity.

4. The method according to claim 3, wherein, Use a generative adversarial network including two artificial neural networks to generate the training data set for developing the first operator in the trained machine learning algorithm.

5. The method according to any one of claims 2 to 4, wherein, Use electromagnetic simulation to generate the training data set for developing the first operator in the trained machine learning algorithm.

6. The method according to any one of claims 2 to 5, wherein, The training dataset for developing the second operator in the trained machine learning algorithm includes: magnetic resonance images of a patient load magnetic resonance imaging phantom (5) acquired using the first main magnetic field under a shimming condition for shimming the second main magnetic field to uniformity, and magnetic resonance images of the magnetic resonance imaging phantom (5) acquired under the second main magnetic field under a shimming condition for shimming the second main magnetic field to uniformity.

7. The method according to any one of the preceding claims, wherein, In step S5, by minimizing a loss function using the first magnetic resonance image and the second magnetic resonance image and the first operator and the second operator, the magnetic resonance image expected in the case where the magnetic resonance imaging device (1) has acquired the first magnetic resonance image under the first main magnetic field under a shimming condition for shimming the first main magnetic field to uniformity is retrieved instead of the first magnetic resonance image.

8. The method according to claim 7, wherein The following formula applies to the loss function LF: LF = operator1(operator2(image1)) - image2, where operator1 is the first operator, operator2 is the second operator, image1 is the first image and image2 is the second image.

9. The method according to any one of the preceding claims, wherein, The first operator has been obtained by including magnetic resonance image characteristics of magnetic resonance images acquired under the first main magnetic field under a shimming condition for shimming the first main magnetic field to uniformity and magnetic resonance image characteristics of magnetic resonance images acquired under the second main magnetic field under a shimming condition for shimming the second main magnetic field to uniformity, wherein the magnetic resonance images acquired under the first main magnetic field and the second main magnetic field respectively have not been acquired using the magnetic resonance imaging device (1), and the first magnetic resonance image and the second magnetic resonance image are derived from the magnetic resonance imaging device (1).

10. The method according to any one of the preceding claims, wherein, The second operator is obtained by acquiring magnetic resonance images from a magnetic resonance imaging phantom (5) simulating a patient load under the first main magnetic field and the second main magnetic field, and the difference between the magnetic resonance images of the magnetic resonance imaging phantom (5) acquired under the first main magnetic field and the second main magnetic field includes the influence of the non-uniformity of the first main magnetic field having the first magnetic flux density under a shimming condition for shimming the second main magnetic field to uniformity.

11. A computer program for magnetic resonance imaging modification, comprising instructions which, when the program is executed by a computer, cause the computer to perform the method according to any one of claims 1 - 10.

12. A system of a magnetic resonance imaging device (1) and a computing unit (2), the magnetic resonance imaging device (1) having a switch (7) by means of which the main magnetic field is switchable, wherein, The first main magnetic field includes a first magnetic flux density having a shimming condition for shimming the second main magnetic field to uniformity, wherein the second main magnetic field includes a second magnetic flux density having a shimming condition for shimming the second main magnetic field to uniformity, wherein, the second magnetic flux density is different from the first magnetic flux density, wherein, the computing unit (2) is adapted to perform the method according to any one of claims 1 to 11.

13. The system according to claim 12, wherein, A superconducting magnet (6) is provided for generating the first main magnetic field and the second main magnetic field, and the superconducting magnet is a dry magnet.

Images