Method for predicting magnetic field perturbation maps for magnetic resonance imaging - Patents.com

Abstract

The proposed concept therefore aims to provide schemes, solutions, concepts, designs, methods and systems related to predicting a magnetic field perturbation map for a magnetic resonance imaging (MRI) of a subject. In particular, the invention aims to provide a magnetic field perturbation map of a subject without requiring additional time-consuming scanning of the subject. An accurate magnetic field perturbation map is necessary to obtain a high quality MRI scan of the subject. Therefore, a synthetic computed tomography (CT) image is generated by inputting an initial MRI intensity image of the subject into an image transformation machine learning algorithm. A weighted susceptibility map of the subject is then determined based on the synthetic CT image and the initial MRI intensity image, which is then used to determine the magnetic field perturbation map of the subject.

Description

[Technical field]

[0001] The present invention relates to the field of magnetic resonance imaging (MRI), and in particular to obtaining a magnetic field perturbation map of a subject for MRI. [Background technology]

[0002] During an MRI scan, magnetic susceptibility variations within a subject cause magnetic field inhomogeneities that result in reduced image quality and integrity. As a result, it is common to spend a significant amount of scanning time prior to the actual MRI scan to obtain the magnetic field perturbation information that is used to compensate for these variations.

[0003] Instead of spending scan time to obtain this magnetic field perturbation information, it is possible to predict the magnetic field perturbation map of the subject. However, magnetic field inhomogeneity is difficult to predict in vivo because the magnetic susceptibility variation depends on the subject's anatomy. In addition, the subject's orientation also affects the magnetic field inhomogeneity, and thus the subject's motion also affects the magnetic field inhomogeneity. Moreover, the image distortion caused by the magnetic field inhomogeneity also depends on the scan settings. Therefore, the magnitude of the observed imaging artifact depends on all three independent variables, namely the subject's anatomy, motion, and scan settings.

[0004] An additional problem arises if the subject moves between the magnetic field perturbation map acquisition and the actual MRI scan. This movement causes changes in the magnetic field perturbation map, which leads to degradation of image quality. For example, when the subject's head moves within the magnetic bore of the MRI scanner, the magnetic susceptibility field also moves. Hence, the magnetic field perturbation map changes and needs to be updated for the new head position.

[0005] There are known methods for predicting magnetic field perturbation maps using analytical phase-based methods. The methods require tissue shape, type, and corresponding magnetic susceptibility values ​​to be used as input. Therefore, in order to perform a magnetic field perturbation map prediction for a subject, it is necessary to obtain information about the subject's tissue shape, structure, and tissue type information. Generally, current methods obtain this "shape and tissue type" information from a subject's computed tomography (CT) images. Summary of the Invention

[0006] The invention is defined by the claims.

[0007] According to an example according to an aspect of the present invention, a method for predicting a magnetic field perturbation map for magnetic resonance imaging (MRI) of a subject is provided.

[0008] The method includes the steps of obtaining an initial MRI intensity image of the subject, generating a synthetic computed tomography (CT) image based on the initial MRI intensity image by using an image transformation machine learning algorithm, determining a weighted susceptibility map based on the synthetic CT image and the initial MRI intensity image, and determining a magnetic field perturbation map of the subject based on the weighted susceptibility map.

[0009] The proposed concept aims to provide schemes, solutions, concepts, designs, methods and systems related to predicting magnetic field perturbation maps for e.g. MRI. In particular, the embodiments aim to reduce the time spent scanning a subject prior to an MRI scan while still ensuring high image quality and completeness. This is achieved by predicting the subject's magnetic field perturbation map instead of spending additional scan time mapping magnetic field inhomogeneities caused by the subject. The prediction is based on information typically acquired during an MRI scan, thus eliminating the need for additional scans.

[0010] Specifically, the step of generating a synthetic CT image of the subject by using an image transformation machine learning algorithm based on the initial MRI intensity image means that it is not necessary to perform a CT scan to predict the magnetic field perturbation map. The synthetic CT image is then used together with the initial MRI intensity image to determine the weighted susceptibility map and, ultimately, the magnetic field perturbation map of the subject. The initial MRI intensity image is generally acquired during a survey scan of the subject, which is always performed during an MRI session. As a result, the acquisition of this information also does not add any additional time to the scanning process.

[0011] In other words, embodiments of the present invention avoid the need to obtain a magnetic field perturbation map directly by scanning the subject, and further overcome the time cost required to obtain the subject's shape and tissue type information necessary to predict the subject's magnetic field perturbation map. By utilizing image transformation machine learning algorithms, there is no need to perform additional scans of the subject to obtain the shape and tissue type information.

[0012] In some embodiments, the image transformation machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and known outputs, where the training inputs include MRI images and the known outputs include CT images, and the MRI images and CT images are provided in pairs.

[0013] In some embodiments, determining the magnetic field perturbation map is further based on analytical methods.

[0014] In some embodiments, the method further comprises identifying susceptibility partitions of the subject based on the initial MRI intensity image and the synthetic CT image, where determining the weighted susceptibility map is based on the identified susceptibility partitions to account for partial volume effects across boundaries.

[0015] In this way, the composite CT image and the initial MRI intensity image are used to accurately segment the subject's major susceptibility compartments, such as the head compartment, etc. Hence, the weighted susceptibility map takes into account partial volume effects across the subject's tissue boundaries, thereby ensuring a high quality weighted susceptibility map, which ultimately results in a more accurate magnetic field perturbation map.

[0016] In some embodiments, the method further comprises regularizing the magnetic field perturbation map at the subject's tissue interface based on the phase of the initial MRI intensity image.

[0017] In some embodiments, the method further comprises correcting imperfections in the magnetic field perturbation map by using an image reconstruction machine learning algorithm.

[0018] In this way, known imperfections in the magnetic field perturbation map are corrected by a machine learning algorithm that identifies and corrects imperfections that arise due to failures in the algorithm that converts the weighted susceptibility map into a magnetic field perturbation map.

[0019] In some embodiments, the method further comprises determining movement of the subject, updating the weighted susceptibility map based on the movement of the subject, and updating the magnetic field perturbation map based on the updated weighted susceptibility map.

[0020] In other words, in some embodiments, if it is detected that the subject has moved during a section of the MRI scan, the weighted susceptibility map is updated to reflect this, thereby improving the quality of the resulting MRI images as the magnetic field perturbation map remains as accurate as possible despite possible changes in magnetic field susceptibility.

[0021] In some embodiments, updating the weighted susceptibility map comprises updating an orientation of the weighted susceptibility map based on movement of the subject.

[0022] According to an example embodiment of the present invention there is provided a computer program comprising code means for performing a method according to any one of the preceding claims when said computer program is executed on a processing system.

[0023] According to another aspect of the present invention, a system for predicting a magnetic field perturbation map for magnetic resonance imaging (MRI) of a subject is provided.

[0024] The system includes an interface configured to obtain an initial MRI intensity image of the subject, an image transformation component configured to generate a synthetic computed tomography (CT) image based on the initial MRI intensity image by using an image transformation machine learning algorithm, a susceptibility mapping component configured to determine a weighted susceptibility map based on the synthetic CT image and the initial MRI intensity image, and a perturbation mapping component configured to determine a magnetic field perturbation map of the subject based on the weighted susceptibility map.

[0025] In some embodiments, the image transformation component includes a training unit configured to train an image transformation machine learning algorithm that is trained using a training algorithm configured to receive an array of training inputs and known outputs, where the training inputs include MRI images and the known outputs include CT images, and where the MRI images and CT images are provided in pairs.

[0026] In some embodiments, the perturbation mapping component includes an analysis unit configured to determine a magnetic field perturbation map based on an analysis method.

[0027] In some embodiments, the susceptibility mapping component includes a segmentation unit configured to identify susceptibility zones of the subject based on the initial MRI intensity image and the synthetic CT image, where the susceptibility mapping component is further configured to determine a weighted susceptibility map based on the identified susceptibility zones to account for partial volume effects across boundaries.

[0028] In some embodiments, the perturbation mapping component further includes an interface correction unit configured to regularize the magnetic field perturbation map at a tissue interface of the subject based on a phase of the initial MRI intensity image.

[0029] In some embodiments, the perturbation mapping component further includes a defect correction unit configured to correct defects in the magnetic field perturbation map by using an image regeneration machine learning algorithm.

[0030] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0031] For a better understanding of the present invention, and to show more clearly how the same may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which: [Brief description of the drawings]

[0032] [Figure 1] 1 is a flow diagram of an exemplary embodiment of a method for predicting a magnetic field perturbation map of a subject. [Diagram 2] 11 is a flow diagram of a further exemplary embodiment of a method for predicting a magnetic field perturbation map of a subject including motion detection, phase-based regularization, and imperfection correction. [Diagram 3] FIG. 1 is a simplified block diagram of a system in which a predicted magnetic field perturbation map of a subject is used in the future. [Figure 4] FIG. 1 is a simplified block diagram of an exemplary embodiment of a system for predicting a magnetic field perturbation map of a subject. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0033] The present invention will now be described with reference to the figures.

[0034] It should be understood that the detailed description and specific examples, while illustrating exemplary embodiments of the devices, systems, and methods, are for illustrative purposes only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the devices, systems, and methods of the present invention will become better understood from the following description, the appended claims, and the accompanying drawings. It should be understood that the figures are merely schematic and not drawn to scale. It should also be understood that the same reference numbers are used throughout the figures to denote the same or similar parts.

[0035] Variations to the disclosed embodiments can be understood and effected by those skilled in the art, from a study of the drawings, the disclosure, and the appended claims, in practicing the claimed invention. In the claims, the words "comprising," "including," "having" do not exclude other elements or steps, and singular elements do not exclude a plurality.

[0036] It should be understood that the figures are only schematic and are not drawn to scale. It should also be understood that the same reference numbers are used throughout the figures to denote the same or similar parts.

[0037] Embodiments according to the present disclosure relate to various techniques, methods, schemes, and / or solutions related to predicting a magnetic field perturbation map for magnetic resonance imaging (MRI) of a subject. According to the proposed concepts, several possible solutions are implemented separately or jointly. That is, although these possible solutions are described separately below, two or more of these possible solutions may be implemented in one combination or another.

[0038] The disclosed invention eliminates the need for a separate magnetic field perturbation map acquisition step when acquiring MRI images of a subject by predicting the magnetic field perturbation map using a combination of analytical physics-based methods and artificial intelligence (AI) algorithms.

[0039] Accordingly, the present invention uses an image transformation machine learning algorithm that generates a synthetic CT image. In some embodiments, this takes into account partial volumetrics across tissue interfaces to obtain accurate shape and tissue type information. Analytical methods are then used to predict a personalized magnetic field perturbation map. The synthetic CT image is generated using a survey scan that is always acquired during the MRI scanning procedure for planning purposes.

[0040] Even given physically accurate shape and tissue type information, analytical magnetic field perturbation map prediction methods still have inherent limitations in prediction accuracy, especially at tissue interfaces. In an embodiment, this deviation is learned using an image reconstruction machine learning algorithm to improve the prediction accuracy of the magnetic field perturbation map. In another embodiment, the phase component of the survey scan is used to regularize the magnetic field perturbation map prediction.

[0041] Embodiments of the present invention provide the following important advantages over conventional systems: (a) Time saved by avoiding the acquisition of multiple magnetic field perturbation maps in a scanning session. The present invention does not require the acquisition of magnetic field perturbation maps. Thus, the present invention provides an improvement in the efficiency of scanning in terms of time. (b) Improved image quality by correcting for unexplained motion during the scan. A magnetic field perturbation map is not always acquired before each individual scan in an MRI session. Even if a magnetic field perturbation map is acquired, subject motion can occur after the prescan. This results in a misalignment between the magnetic map and the actual magnetic field in the magnetic bore. All of these factors result in artifacts in the reconstructed image. The proposed embodiment overcomes this challenge, as unexplained subject motion is captured via prediction of changes in the magnetic field perturbation map. (c) A personalized magnetic field perturbation mapping is provided. Some embodiments utilize survey scans acquired at all times during a session to extract the information required for personalized magnetic field perturbation map prediction.

[0042] Furthermore, the present invention includes the following main elements: (i) AI to generate synthetic CT images from initial MRI intensity images for magnetic field perturbation map prediction. (ii) Using phase data from the initial MRI intensity images to regularize the magnetic field perturbation map predictions. (iii) improved analytical prediction by identifying distinct regions of interest in a subject and optimizing susceptibility assignments across those regions of interest. (iv) To extract optimized geometry and tissue type, and hence susceptibility map, information from synthetic CT images for improved magnetic field perturbation map prediction at tissue interfaces. (v) Correcting known deficiencies in analytical magnetic field perturbation map predictions through the use of AI, which results in improved accuracy of the predicted magnetic field perturbation maps.

[0043] Referring now to FIG. 1, a flow diagram of an exemplary embodiment of a method 100 for predicting a magnetic field perturbation map for an MRI of a subject is shown.

[0044] In step 102, an initial MRI intensity image of the subject is obtained. This is obtained during a survey scan, which is always performed before the main scan session during an MRI scan of the subject. Alternatively, this may be obtained during the main scan session, which may be the first scan, second scan, or any other MRI scan acquired.

[0045] In step 104, a synthetic CT image based on the initial MRI intensity image is generated. This is generated by inputting the initial MRI intensity image into an image transformation machine learning algorithm. The synthetic CT image is nearly identical to a CT image obtained by directly performing a CT scan of the subject. As a result of step 104, it is not necessary to obtain a CT image of the subject to determine the weighted susceptibility map and ultimately predict the magnetic field perturbation map. As a result, the total scan time is reduced.

[0046] The image transformation machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and known outputs, where the training inputs include MRI images and the known outputs include CT images, where the MRI images and the CT images are known MRI-CT image pairs.

[0047] As a result, a synthetic CT image of the subject is obtained by inputting the subject's initial MRI intensities into a trained image transformation machine learning algorithm.

[0048] The image transformation machine learning algorithm can be an artificial neural network (or simply a neural network). The structure of an artificial neural network is inspired by the human brain. A neural network is composed of layers, each layer containing multiple neurons. Each neuron contains a mathematical operation. In particular, each neuron contains a different weighted combination of a single type of transformation (e.g., the same type of transformation, such as sigmoid, but with different weightings). In the process of processing the input data, the mathematical operation of each neuron is performed on the input data to produce a numerical output, and the output of each layer of the neural network is sequentially fed to the next layer. The final layer provides the output.

[0049] There are several types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). An embodiment of the present invention utilizes a CNN-based learning algorithm because CNNs have proven particularly successful in analyzing images and can classify images with a much lower error rate than other types of neural networks.

[0050] CNNs generally contain several layers, including convolutional layers, pooling layers, fully connected layers, and softmax layers. Convolutional layers consist of a set of learnable filters and extract features from the input. Pooling layers are a form of nonlinear downsampling and reduce data size by combining the outputs of multiple neurons in one layer into a single neuron in the next layer. Fully connected layers connect each neuron in one layer to every neuron in the next layer. Softmax layers determine the probability distribution of each output.

[0051] Methods for training machine learning algorithms are well known. In general, such methods include obtaining a training data set including training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and the corresponding training output data entry is used to correct the machine learning algorithm. This process is repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%). This is generally known as a supervised learning technique.

[0052] For example, the weights of the mathematical operations of each neuron are modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation algorithms, etc.

[0053] The training input data entries for the multiple CNN-based learning algorithms used in step 104 correspond to MRI images of the subject. The training output data entries correspond to CT images of the corresponding subject. As a result, the MRI images and the CT images are known MRI-CT image pairs.

[0054] Several CNN-based learning algorithms are created by modifying existing CNN-based learning algorithms such as VGG, Inception, and ResNet.

[0055] In step 106, a weighted susceptibility map of the subject is determined. This determination is based on the synthetic CT image and the initial MRI intensity image. This step is performed by known analytical methods.

[0056] In an exemplary embodiment, the region of interest is segmented into sections based on the initial MRI image and the synthetic CT image. For example, the subject's head is segmented into susceptibility sections. A weighted susceptibility map is then determined based on the identified susceptibility sections. This method of determining the weighted susceptibility map takes into account partial volume effects across boundaries. Partial volume effects are defined as the loss of apparent activity in small objects or regions due to the limited resolution of the imaging system.

[0057] A magnetic field perturbation map of the subject is determined based on the weighted susceptibility map, step 108. The magnetic field perturbation map is derived from the weighted susceptibility map by known analytical methods.

[0058] The magnetic field perturbation map is used to set appropriate shim currents in the MRI machine, which reduce or eliminate the magnetic field inhomogeneities. Alternatively or additionally, the magnetic field perturbation map is used to reconstruct the MRI image. In this way, artifacts in the subject's MRI image are suppressed and higher quality images are obtained.

[0059] As a further example, a flow diagram of an exemplary embodiment of a method 200 for predicting a magnetic field perturbation map for an MRI of a subject, including motion detection, phase-based regularization, and imperfection correction, is shown in FIG.

[0060] In step 202, an initial MRI intensity image is obtained and input to an image transformation machine learning algorithm. As described above, the image transformation machine learning algorithm takes as input the initial MRI intensity image at the beginning of the scan, and the image transformation machine learning algorithm is trained on a large number of MRI-CT pairs to generate a synthetic CT image of the subject.

[0061] In step 204, the composite CT image and the initial MRI intensity image are used for accurate segmentation of the subject's major susceptibility compartments and for refining the boundaries between the subject's various interfaces.

[0062] A weighted susceptibility map is determined from the composite CT image and the initial MRI intensity image in step 206. Instead of assigning a predefined set of susceptibility values ​​to the identified compartments, partial volume effects across the interface boundary are taken into account by weighting the susceptibility map based on the identified susceptibility compartments of the subject.

[0063] In step 208, the subject's motion is determined based on the motion parameters received from a motion detection and estimation unit monitoring the subject within the bore of the MRI machine. If the motion parameters indicate that the subject has moved beyond a threshold minimum motion, it is determined that motion is detected.

[0064] In response to determining that the subject has moved, method 200 proceeds to step 210. The subject is frequently monitored before and during an active scanning session, so this step occurs any time before or during the active scanning session. In step 210, the weighted susceptibility map is updated based on the subject's movement. Updating the weighted susceptibility map includes updating the orientation of the weighted susceptibility map based on the subject's movement. As a result, the already generated weighted susceptibility map is updated and modified without the need to determine the weighted susceptibility map from updated synthetic CT images and initial MRI intensity images of the subject.

[0065] In other words, the weighted susceptibility map is generated from the initial MRI intensity image, which is generated only once. If the system detects motion during an active scanning session, the already generated weighted susceptibility map is updated using the motion parameters received from the motion detection and estimation unit.

[0066] In response to determining that the subject is not moving, or upon updating the weighted susceptibility map, method 200 proceeds to step 212. In step 212, a magnetic field perturbation map of the subject is determined from the weighted susceptibility map, or possibly from the updated weighted susceptibility map. The magnetic field perturbation map is derived from the weighted susceptibility map using analytical methods.

[0067] In step 214, it is determined whether the phase of the initial MRI intensity image is available. This depends on the scan type used to obtain the MRI intensity image, or on the MRI machine itself. The survey scan is a gradient echo scan or an asymmetric spin echo scan.

[0068] In response to determining that the phase of the initial MRI intensity image is available, the method 200 proceeds to step 216. In step 216, the magnetic field perturbation map is regularized at the interface. This regularization is performed based on the phase of the initial MRI intensity image. This regularization results in a higher quality magnetic field perturbation map.

[0069] In response to determining that the initial MRI intensity phase is not available or upon regularizing the magnetic field perturbation map, the method 200 proceeds to step 218. In step 218, defects in the magnetic field perturbation map are corrected. These defects result from the inherent limitations of the analytical approach of determining the magnetic field perturbation map based on the weighted susceptibility map, as the analytical technique cannot be generalized sufficiently. To correct these defects, an image regeneration machine learning algorithm is used. The image regeneration machine learning algorithm processes the magnetic field perturbation map for further refinement. As a result, the magnetic field perturbation map becomes more accurate and is used for a more accurate reconstructed image.

[0070] In step 220, the predicted magnetic field perturbation map is output. The predicted magnetic field perturbation map is subsequently used to set appropriate shim currents in the MRI machine or to reconstruct the MRI scan. In this way, artifacts in the subject's MRI images are suppressed and higher quality images are obtained.

[0071] As a further example, FIG. 3 shows a simplified block diagram of a system 300 in which a predicted magnetic field perturbation map of a subject may be used in the future.

[0072] The magnetic field perturbation map prediction module 306 predicts the magnetic field perturbation map directly based on the survey scan 302, which includes an initial MRI intensity image and, if available, a phase image (depending on the scan type). Throughout the main scanning session 304, the subject in the bore is monitored for motion by a motion estimation unit 308. If motion is detected, the motion parameters are fed back to the magnetic field perturbation map prediction module 306 to reconstruct the magnetic field perturbation map for the subject's latest orientation.

[0073] In other words, before the main scanning session 304, a smart survey scan 302 is performed on the subject in the bore of the MRI machine. It includes information such as an initial MRI intensity image, and further includes a phase image. This information is used to create a magnetic field perturbation map of the subject in the bore, which is fed back to the main scanning session 304. If a phase image has also been acquired, the phase image is used to regularize the magnetic field perturbation map of the subject. The magnetic field perturbation map is used by the main scanning session 304, for example, to modify the shim currents input to the shim coils of the MRI machine. As a result, the images included in the reconstruction stage from the MRI have fewer artifacts and are therefore of higher quality. By using the initial MRI image to create the magnetic field perturbation map, there is no need to perform additional scans to directly obtain the magnetic field perturbation map.

[0074] Throughout the main scanning session 304, the subject is actively monitored in the bore by the motion estimation unit 308 to determine if motion has occurred. If motion has occurred, this will change the magnetic field inhomogeneity in the bore and therefore an updated weighted susceptibility map is required to obtain an accurate magnetic field perturbation map. Therefore, the measured motion parameters are used to update the magnetic field perturbation map, which is then used for the main scanning session 304. If the magnetic field perturbation map changes at various times during the main scanning session 304, this is recorded and further used retrospectively in the reconstruction stage to remove artifacts present in the scan image and improve image quality.

[0075] FIG. 4 illustrates a block diagram of an example embodiment of a system 400 for predicting a magnetic field perturbation map of a subject.

[0076] An interface 410 is provided and configured to obtain an initial MRI intensity image of the subject. The initial MRI intensity image is obtained from an initial survey scan. In some embodiments, the initial MRI intensity image further includes a phase image.

[0077] Further, an image transformation component 420 is provided, which receives an initial MRI intensity image and generates a synthetic CT image based on the initial MRI intensity image from the interface 410. The image transformation component 420 generates a synthetic CT image by inputting the initial MRI image to an image transformation machine learning algorithm. The image transformation component 420 includes a training unit 425 configured to train the image transformation machine learning algorithm, which is trained using a training algorithm configured to receive an array of training inputs and known outputs. The training inputs are MRI images and the known outputs are CT images, where the MRI images and CT images are provided in pairs, each corresponding to the same subject.

[0078] In addition, a susceptibility mapping component 430 is provided, which is configured to determine a weighted susceptibility map based on the composite CT image from the image transformation component 420 and the initial MRI intensity image. In some embodiments, the susceptibility mapping component 430 includes a segmentation unit 435 configured to identify susceptibility zones of the subject based on the initial MRI intensity image and the composite CT image. In other words, the segmentation unit 435 segments regions of the subject that are at interface boundaries between different tissue types or major shapes. The susceptibility mapping component 430 is then further configured to determine a weighted susceptibility map based on the identified susceptibility zones to account for partial volume effects across the boundaries.

[0079] A perturbation mapping component 440 is further provided, which is configured to determine a magnetic field perturbation map of the subject based on the weighted susceptibility map provided by the susceptibility mapping component 430. In an exemplary embodiment, the perturbation mapping component 440 further includes an analysis unit 442 configured to determine the magnetic field perturbation map based on an analysis method.

[0080] In some embodiments, the perturbation mapping component 440 includes a defect correction unit 446. The defect correction unit 446 corrects defects in the magnetic field perturbation map by using an image reconstruction machine learning algorithm. In this manner, known limitations of analytical methods for determining the magnetic field perturbation map from the magnetic susceptibility map are overcome, especially where the known limitations include a lack of ability to generalize adequately.

[0081] In a further exemplary embodiment, the perturbation mapping component 440 further includes an interface correction unit 448 configured to regularize the magnetic field perturbation map at the interface based on the phase of the initial MRI intensity image, which is provided by the interface 410. This regularization improves the quality of the perturbation map at the interface boundary.

[0082] The system further includes a motion estimator 450. The motion estimator 450 is configured to determine the motion of the subject within the bore of the MRI machine during the scanning session. The motion estimator 450 is configured to provide the detected motion parameters to the perturbation mapping component 440. The perturbation mapping component further includes a perturbation map updater 444 configured to update the weighted susceptibility map received from the susceptibility mapping component 430 in response to receiving the motion parameters. The updating includes updating the orientation of the weighted susceptibility map based on the motion parameters. The perturbation map updater 444 then updates the subject's magnetic field perturbation map based on the updated weighted susceptibility map. With the motion estimator 450 and the perturbation map updater 444 provided, the motion of the subject occurring after the initial MRI intensity image is obtained and taken into account to provide improved image quality throughout the scanning session.

[0083] The present invention is a method, system, and / or computer program product that includes a computer readable storage medium (or media) having computer readable program instructions for causing a processor to perform aspects of the present invention.

[0084] A computer readable storage medium is a tangible device that holds and stores instructions for use by an instruction execution device, such as, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing.

[0085] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0086] A single processor or other unit may fulfill the functions of several items recited in the claims.

[0087] It will be understood that the disclosed methods are computer implemented methods, so also the concept of a computer program is proposed, comprising code means for performing any of the described methods when said computer program is run on a processing system.

[0088] Those skilled in the art will be able to readily develop a processor to perform the methods described herein, and each step of the flow diagram thus represents a different action performed by the processor, which is executed by a respective module of the processing processor.

[0089] As discussed above, the system utilizes a processor to perform data processing. The processor may be implemented in numerous ways using software and / or hardware to perform the various functions required. The processor typically utilizes one or more microprocessors that are programmed using software (e.g., microcode) to perform the functions required. The processor may be implemented as a combination of dedicated hardware to perform some functions and one or more programmed microprocessors and associated circuitry to perform other functions.

[0090] Examples of circuitry that may be utilized in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0091] In various embodiments, the processor is associated with one or more storage media, such as volatile and non-volatile computer memories, such as RAM, PROM, EPROM, and EEPROM. The storage media are coded with one or more programs that, when executed by the one or more processors and / or controllers, perform the necessary functions. The various storage media may be fixed or portable within the processor or controller, such that the one or more programs stored thereon are loaded into the processor.

[0092] Modifications to the disclosed embodiments can be understood and effected by those skilled in the art from a study of the drawings, the disclosure, and the appended claims in practicing the claimed invention. In the claims, the words "comprising, including, having" do not exclude other elements or steps, and singular elements do not exclude a plurality. A single processor or other unit fulfills the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage. The computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid medium, provided together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. It should be noted that when the term "adapted to" is used in the claims or description, the term "adapted to" is intended to be equivalent to the term "configured to". Any reference signs in the claims should not be interpreted as limiting the scope.

Claims

1. 1. A method for predicting a magnetic field perturbation map for magnetic resonance imaging (MRI) of a subject, comprising: obtaining an initial MRI intensity image of the subject; generating a synthetic computed tomography (CT) image based on the initial MRI intensity image by using an image transformation machine learning algorithm; determining a weighted susceptibility map based on the composite CT image and the initial MRI intensity image; determining the magnetic field perturbation map of the subject based on the weighted susceptibility map; The method comprising:

2. 2. The method of claim 1, wherein the image transformation machine learning algorithm is trained using a training algorithm to receive an array of training inputs and known outputs, the training inputs including MRI images, the known outputs including CT images, and the MRI images and the CT images are provided in pairs.

3. The method of claim 1 or 2, wherein the step of determining the magnetic field perturbation map is further based on analytical methods.

4. identifying susceptibility zones of the subject based on the initial MRI intensity image and the composite CT image; and 4. The method of claim 1, wherein the step of determining the weighted susceptibility map is based on the identified susceptibility partitions to take into account partial volume effects across boundaries.

5. The method of claim 1 , further comprising the step of regularizing the magnetic field perturbation map at tissue interfaces of the subject based on a phase of the initial MRI intensity image.

6. The method according to claim 1 , further comprising the step of correcting imperfections in the magnetic field perturbation map by using an image reconstruction machine learning algorithm.

7. determining movement of the subject; updating the weighted susceptibility map based on the movement of the subject; updating the magnetic field perturbation map based on the updated weighted susceptibility map; The method of claim 1 , further comprising:

8. The method of claim 7 , wherein updating the weighted susceptibility map comprises updating an orientation of the weighted susceptibility map based on the movement of the subject.

9. A computer program comprising code means for carrying out the method according to any one of claims 1 to 8 when the computer program is executed on a processing system.

10. 1. A system for predicting a magnetic field perturbation map for magnetic resonance imaging (MRI) of a subject, comprising: an interface for obtaining an initial MRI intensity image of the subject; an image transformation component for generating a synthetic computed tomography (CT) image based on the initial MRI intensity image by using an image transformation machine learning algorithm; a susceptibility mapping component for determining a weighted susceptibility map based on the composite CT image and the initial MRI intensity image; a perturbation mapping component for determining the magnetic field perturbation map of the subject based on the weighted susceptibility map; Including, the system.

11. 11. The system of claim 10, wherein the image transformation component includes a training unit for training the image transformation machine learning algorithm, the image transformation machine learning algorithm being trained using a training algorithm to receive an array of training inputs and known outputs, the training inputs including MRI images, the known outputs including CT images, and the MRI images and the CT images being provided in pairs.

12. The system of claim 10 or 11, wherein the perturbation mapping component comprises an analysis unit for determining the magnetic field perturbation map based on an analytical method.

13. the susceptibility mapping component includes a segmentation unit for identifying susceptibility compartments of the subject based on the initial MRI intensity image and the composite CT image; 13. The system of claim 10, wherein the susceptibility mapping component further determines the weighted susceptibility map based on the identified susceptibility partitions to account for partial volume effects across boundaries.

14. 14. The system of claim 10, wherein the perturbation mapping component further comprises an interface correction unit for regularizing the magnetic field perturbation map at tissue interfaces of the subject based on a phase of the initial MRI intensity image.

15. The system of claim 10 , wherein the perturbation mapping component further comprises a defect correction unit for correcting defects in the magnetic field perturbation map by using an image regeneration machine learning algorithm.

Images