Methods for updating magnetic field maps

By using the B0 prediction model and spherical harmonic coefficients to predict changes in the B0 map caused by the motion of an MRI object, the problem of image quality degradation caused by object movement in MRI is solved, and efficient and accurate magnetograph updates and image quality improvement are achieved.

CN116685857BActive Publication Date: 2026-11-17KONINKLIJKE PHILIPS NV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202180089730.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2021-01-08
Filing Date
2021-12-29
Publication Date
2026-11-17
Estimated Expiration
2041-12-29

AI Technical Summary

Technical Problem

Existing MRI techniques have difficulty accurately updating the magnetograph (B0 map) when the object moves, resulting in a decrease in image quality. Existing methods cannot effectively model changes in the B0 map caused by complex motion, and the additional scanning time is costly.

Method used

By combining the B0 prediction model with spherical harmonic coefficients, the changes in the B0 diagram caused by the motion of the object are predicted using the initial B0 diagram and motion data. Global and local changes are separated, and training algorithms and synthetic B0 diagrams are used to improve modeling accuracy and dynamically correct magnetic field inhomogeneity.

Benefits of technology

It reduces scan time, improves MRI image quality, enhances robustness to complex motion, ensures magnetic field homogeneity, reduces artifacts, and supports prospective and retrospective corrections.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116685857B_ABST
    Figure CN116685857B_ABST
Patent Text Reader

Abstract

The proposed concept aims to provide plans, schemes, ideas, designs, methods, and systems related to updating the magnetic field (BO) map of an object during magnetic resonance imaging (MRI) examinations or image reconstruction. Specifically, the magnetic field inside the magnetic cavity changes when the object moves. As a result, any BO map obtained before the object's movement may be inaccurate. Therefore, the initial BO map is updated to reflect the changes in the BO map caused by the object's movement. This can be achieved by determining the object's BO map based on a BO prediction model without incurring additional scan time to acquire another BO map.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging (MRI), and more particularly to updating the magnetic field map during MRI examination or reconstruction. Background Technology

[0002] Magnetic resonance imaging (MRI) reconstruction assumes that the magnetic field (B0) is spatially homogeneous. This condition is rarely met, leading to degraded image quality and artifacts. In particular, changes in magnetic susceptibility within an object during an MRI scan cause B0 inhomogeneities. Therefore, producing high-quality MRI images depends on correcting these B0 inhomogeneities. Consequently, a significant amount of scan time is typically spent before the actual MRI scan in order to obtain a B0 map. This B0 map can be used to compensate for B0 inhomogeneities, for example, by shimming or by compensating for B0 inhomogeneities during reconstruction.

[0003] However, further problems arise if the subject moves between the B0 map acquisition and the actual MRI scan. This movement causes changes in the B0 map, leading to a deterioration in image quality. For example, when the subject's head moves within the magnetic cavity of the MRI scanner, the magnetic susceptibility field also shifts. Therefore, the magnetic field changes, necessitating an update of the B0 map for the new head position. Unfortunately, performing a B0 pre-scan before acquiring each volume is impractical.

[0004] During rigid motion, the magnetic susceptibility field follows the object's movement. However, for larger movements, this estimation is poor, especially around axes not parallel to the magnetic field. A method is needed to model the variation of B0 with respect to motion. Existing methods focus on the effects of isolated rotations along a single axis and are limited in terms of the number of locations considered and generalizability across objects. Summary of the Invention

[0005] According to an example of one aspect of the invention, a method is provided for updating a magnetic field (BO) map during a magnetic resonance imaging (MRI) examination of an object or during MRI reconstruction.

[0006] The method includes: obtaining an initial Boolean map of the object associated with the start of the MRI examination; obtaining motion data of the object during the MRI examination; and determining a motion-updated Boolean map of the object based on a Boolean prediction model, the initial Boolean map of the object, and the motion data. The Boolean prediction model is configured to model changes in the Boolean map as a (linear or nonlinear) function of the object's motion parameters. Furthermore, the changes in the Boolean map are approximated using a set of spherical harmonic coefficients.

[0007] Therefore, the proposed concept aims to provide plans, schemes, ideas, designs, methods, and systems related to updating the BoO map of an object during MRI examination or reconstruction. Specifically, embodiments are designed to update the BoO map to reflect changes in the BoO map caused by (rigid and non-rigid) movement of the object during MRI examination. This can be achieved by determining the object's BoO map based on a BoO prediction model without incurring additional scan time to acquire another BoO map.

[0008] In other words, when an object moves, the magnetic field inside the magnetic cavity also moves. Consequently, any B0 map obtained before the object moves may be inaccurate. This invention solves this problem by determining a motion-updated B0 map. This is achieved using a B0 prediction method along with the initial B0 map of the object and motion data captured during the MRI examination. The motion-updated B0 map can accurately represent the magnetic field present in the magnetic cavity of the MRI machine after the object moves. As a result, methods for mitigating magnetic field inhomogeneities using B0 maps may be more effective. Ultimately, high-quality MRI images of the object with fewer artifacts can be obtained.

[0009] Furthermore, changes in the B0 diagram caused by the movement of an object can be modeled, and these changes are not limited to rotation around a single axis. The inventors have recognized that changes in the B0 diagram can be obtained by modeling this movement as a linear or nonlinear function of the object's rotational motion. Additionally, a simplified yet accurate model of the changes in the B0 diagram can be obtained by approximating this motion as a set of spherical harmonic coefficients.

[0010] In some embodiments, the changes in the B0 diagram include global changes (for a given anatomical structure) that are common across different objects and local changes specific to each object. In this case, determining the motion-updated B0 diagram includes: determining the global changes in the initial B0 diagram of the object based on the lower-order spherical harmonics and the motion data; and determining the local changes in the initial B0 diagram of the object based on the higher-order spherical harmonics and the motion data.

[0011] In this way, the predicted changes in the B0 graph are divided into two levels—local changes and global changes. This means separating parameters that cannot be generalized across structures (local changes) from those that can be generalized (global changes). As a result, the model can be trained and developed separately. The predicted changes can then be combined to create the final prediction.

[0012] In some embodiments, the spherical harmonic coefficients are determined based on a training algorithm of an array configured to receive training inputs. The training inputs include Boolean (B0) maps of multiple different objects at multiple different locations and motion parameters corresponding to the Boolean maps.

[0013] In other words, B0 maps collected from multiple objects in different motion states can be used to obtain spherical harmonic coefficients using training algorithms.

[0014] In some embodiments, the training input further includes a synthetic Bo map generated by the motion frame and the motion parameters corresponding to the synthetic Bo map.

[0015] In addition to B0 maps acquired from multiple objects in different orientations, synthetic B0 maps can be generated and used to obtain spherical harmonic coefficients. Due to equipment limitations (especially in high-field conditions), B0 maps acquired from different objects can be susceptible to measurement errors. In contrast, synthetic B0 maps are more accurate and robust. Therefore, using synthetic B0 maps to determine spherical harmonic coefficients yields more accurate results. Using synthetic B0 maps allows for the generation of virtually unlimited data across object motion states and structures with significant variations, thus further benefiting the learning process.

[0016] In some embodiments, generating a synthetic BO diagram includes: obtaining a structural image of a base object; generating a motion-simulated structural image based on the structural image and a set of motion parameters, the motion parameters including rotation, stretching, and tilting; and generating a synthetic BO diagram by inputting the motion-simulated structural image into an analytical BO calculation method.

[0017] In some embodiments, generating the motion simulation structure image includes: obtaining a point cloud of the structure image; generating an initial mesh based on the point cloud; identifying structural regions of the initial mesh that include rigidly moving structural regions, structural regions that include non-rigidly moving structural regions, and structural regions that include fixed regions; generating a set of moving meshes based on the identified structural regions and the set of motion parameters, according to the initial mesh; and generating the motion simulation structure image based on the initial mesh and the moving meshes.

[0018] In some embodiments, generating a moving mesh from the set of moving meshes based on the initial mesh includes: obtaining a first portion of the moving mesh by transforming a rigid moving region based on a motion parameter from the set of motion parameters; obtaining a second portion of the moving mesh by keeping the fixed region unchanged; and obtaining a third portion of the moving mesh by transforming a non-rigid moving region to make the moving mesh a smoothly changing surface mesh.

[0019] In some embodiments, the structural image is a computed tomography (CT) image or an MRI image of the underlying object.

[0020] In some embodiments, the training algorithm is a curve fitting model or a neural network.

[0021] In other words, the training algorithm used to obtain the spherical harmonic coefficients can be a curve fitting model or a neural network.

[0022] In some embodiments, obtaining the initial B0 map includes: acquiring the initial B0 map from a B0 pre-scan performed at the start of the inspection; or predicting the initial B0 map based on an investigation scan.

[0023] In some embodiments, the determination of the motion update BO map of the object is also based on an initial MRI image obtained at the start of the examination.

[0024] By using additional initial MRI images, a more accurate motion-updated B0 map can be determined.

[0025] In some embodiments, the method further includes applying an updated shimming coil current to the shimming coil of an MRI scanner used during the MRI examination. In this case, the updated shimming coil current is configured to correct for magnetic field inhomogeneities of the MRI scanner based on the motion-updated B0 map.

[0026] Therefore, by using motion-updated B0 maps, the magnetic field inside the MRI machine's bore can be homogenized. MRI images acquired after adjusting the shimming coils can have higher quality and fewer artifacts.

[0027] In some embodiments, the method further includes retrospectively reconstructing MRI images of the object acquired during the MRI examination based on the motion-updated BO map.

[0028] This method can be used as an alternative or supplementary method for adjusting the shimming coil.

[0029] According to an example of one aspect of the invention, a computer program comprising code units is provided for implementing the method according to the invention when the program is run on a processing system.

[0030] These and other aspects of the invention will become apparent and elucidated with reference to one or more embodiments described below. Attached Figure Description

[0031] To better understand the invention and to more clearly illustrate how it can be practiced, reference will now be made to the accompanying drawings by way of example only, in which: Figure 1A This is a flowchart of an exemplary embodiment of a method for updating a Bo map during an MRI examination of an object; Figure 1BThis is a flowchart of another exemplary embodiment of a method for updating a BO diagram, including methods for directly acquiring the BO diagram or notifying an operator; Figure 2 It is a simplified block diagram of the B0 prediction model used to determine the B0 graph for motion updates; Figure 3 This is a flowchart of the method for generating a composite B0 diagram; Figure 4 It is a graph representing the input, output, and functions of the motion simulation framework; Figure 5 This is a simplified block diagram of a system for prospective B0 change estimation and motion correction; and Figure 6 This is a simplified block diagram of a system for retrospective B0 change estimation and motion correction. Detailed Implementation

[0032] The invention will be described with reference to the accompanying drawings.

[0033] It should be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, they 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 apparatus, system, and method of the invention will be better understood from the following description, claims, and drawings. It should be understood that these drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used throughout all the drawings to indicate the same or similar parts.

[0034] Based on a study of the accompanying drawings, the disclosure, and the appended claims, those skilled in the art will be able to understand and implement variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality.

[0035] It should be understood that the accompanying drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used throughout all the drawings to indicate the same or similar parts.

[0036] The embodiments of this disclosure relate to various related technologies, methods, plans, and / or schemes. Based on the proposed concepts, many possible schemes can be implemented individually or in combination. That is, while these possible schemes may be described individually below, two or more methods of these possible schemes may be implemented in one combination or another.

[0037] Embodiments of the present invention aim to update the B0 map to reflect changes in the B0 map caused by object movement during an MRI examination. This can be achieved by determining the object's B0 map based on a B0 prediction model without incurring additional scan time to acquire another B0 map.

[0038] In summary, the embodiments can provide the following key advantages over existing methods: (i) Reduced scanning time. Scanning time can be reduced by avoiding the acquisition of multiple B0 images during the scanning process. By using only a single B0 image acquired at the beginning of the process and predicting changes in the B0 image due to motion, the latest B0 image for the entire process can be obtained. Therefore, this improves scanning efficiency in terms of time.

[0039] (ii) Improved quality of acquired MRI images. Correcting for unconsidered motion of the subject during scanning improves the final acquired MRI scan. Typically, B0 map scans are not always performed before each scan, meaning that subject movement can distort the scan. In fact, even if a B0 map is acquired before the acquisition volume, motion can occur during the scanning process. Therefore, a mismatch may exist between the B0 map and the actual magnetic field in the bore. These can all lead to artifacts in the reconstructed images.

[0040] (iii) The B0 map updates are robust to motion and structural changes. The embodiments are robust to a wide range of motion parameters, such as rotation, tilt, and stretching. The proposed embodiments are also robust to anatomical changes between objects, meaning that accurate motion-updated B0 maps can be acquired.

[0041] (iv) The B0 map update is robust to changes caused by motion involving interactions of rotation in various directions. Existing methods for B0 map prediction study the effects of isolated rotations along a single axis (which is highly impractical). Embodiments of the present invention consider higher-order interactions between various motion parameters to account for more complex B0 map changes caused by motion. As a result, more accurate B0 maps can be predicted, which in turn improves the quality of the reconstructed MRI scans.

[0042] (v) Magnetic field correction via motion detection results in a uniform field throughout the entire examination. In some embodiments, the motion sensor actively senses any occurring motion and provides a method for predicting changes in the BO map. Thus, by employing dynamic shimming, the magnetic field can remain uniform throughout the entire scanning process, independent of motion.

[0043] (vi) B0 map prediction can also be performed retrospectively, which allows for iterative motion correction reconstruction.

[0044] In some embodiments, the present invention includes a B0 map prediction method to determine a motion-updated B0 map that is robust to anatomical changes. This motion-updated B0 map is based on an initial B0 map acquired at the start of the scan, and may also be based on an initial MRI image.

[0045] In one embodiment of the B0 prediction method, a B0 prediction model is provided in which global (lower-order) changes that are common across objects are predicted. The method can also predict increasingly object-specific local (higher-order) changes.

[0046] Alternatively, in some other embodiments of the B0 prediction model, a motion simulation framework is provided. This framework can be used to generate a synthetic B0 map, which can then be used to create a B0 prediction model using structural data.

[0047] In another embodiment of the hybrid B0 prediction method, the two methods described above can be combined to improve the prediction method.

[0048] The embodiments may also provide a system for proactively correcting object motion. This system can actively capture motion data by monitoring the object's motion and can ensure that the magnetic field is uniform throughout the scanning process, independent of motion. Specifically, the system according to some embodiments may include: (i) Motion sensors that actively monitor objects and identify when object motion occurs in order to trigger motion estimation.

[0049] (ii) Motion estimator, which identifies motion and estimates motion quality (including whether magnetic field correction is required).

[0050] (iii) B0 graph updater, which is configured to predict and update the B0 graph given motion, position and structural parameters.

[0051] (iv) B0 graph scanner, which is used to determine when magnetic field correction is needed, but cannot estimate motion or predict changes in the B0 graph.

[0052] (v) A shimming coil controller configured to determine the required shimming coil current based on the actual or predicted B0 diagram / if the MRI scanner allows shimming at that given time, the shimming coil controller triggers shimming.

[0053] (vi) A component that triggers a warning to the user when a large movement occurs during a scan in an inspection.

[0054] In another embodiment, the B0 map prediction method is used to retrospectively correct motion within an iterative reconstruction framework. Changes in the B0 map are iteratively estimated and corrected until convergence.

[0055] Now for reference Figure 1A The flowchart depicts an exemplary embodiment of a method 100 for updating a BO map during an MRI examination of an object.

[0056] In step 102, an initial B0 image of the subject is obtained. The initial B0 image is associated with the start of the MRI examination. The initial B0 image of the subject can be obtained at the start of the examination. Alternatively, the initial B0 image can be obtained before or during the examination.

[0057] Acquiring an initial B0 image can involve acquiring it from a preparatory scan performed at the start of the inspection. This involves scanning the object and constructing the B0 image using known methods. Alternatively, obtaining an initial B0 image can involve predicting it based on an initial image (e.g., a survey scan). Predicting the B0 image based on a survey scan can mean a reduction in total scan time because no additional preparatory scan is required.

[0058] In step 104, motion data of the object during the MRI examination is acquired. Motion data may include rotation, tilt, translation, and / or stretching of the entire object. Alternatively, motion data may also include rotation, tilt, translation, and / or stretching of only a region of interest of the object (e.g., the head). Motion data may be acquired by a motion sensor that monitors the object within the magnetic cavity of the MRI machine. The motion sensor can identify when object motion occurs and, in response to determining that object motion has occurred, proceed to step 106. The motion sensor may also capture a set of motion parameters in response to determining that object motion has occurred. Motion data may be captured by a system (e.g., VitalEye) installed in the MRI machine or by another suitable technique.

[0059] In step 106, a motion-updated B0 map of the object is determined. The motion-updated B0 map is based on a B0 prediction model, the object's initial B0 map, and motion data. The determination of the motion-updated B0 map can also be based on an initial MRI image. The initial MRI image may include an MRI amplitude image and / or an MRI phase image.

[0060] Providing accurate B0 maps is a fundamental requirement for reconstructing high-quality MRI images of an object. However, movement of the object within the magnetic cavity causes changes in the magnetic field inside the cavity, resulting in inaccurate B0 maps. This can be particularly problematic when the object is young or elderly. Furthermore, performing additional scans to determine updated B0 maps consumes significant scan time and often cannot be performed during scans of the object's volume. This invention overcomes these problems by determining motion-updated B0 maps, which record the magnetic field inside the MRI machine's magnetic cavity after object movement. Because motion-updated B0 maps are determined based on B0 prediction models, additional scans can be performed to determine B0 maps in response to object movement. References will follow below. Figure 2 A more detailed description of the B0 prediction model.

[0061] Finally, the B0 graph updated by motion can be used to perform shimming, as shown in the reference. Figure 5 As described in more detail. Additionally, the motion-updated B0 map can be retrospectively used to reconstruct the magnetic resonance images of the object, as in the reference... Figure 6 As described in more detail. Ultimately, a motion-updated B0 map can be used to obtain an MRI image of the object, which produces fewer artifacts compared to using the initial B0 map.

[0062] exist Figure 1B An alternative exemplary embodiment of a method 150 for updating a BO map during an MRI examination of an object is shown.

[0063] Steps 102, 104, and 106 can be referenced above. Figure 1A The steps described are the same or substantially the same. Additionally, steps 152, 154, and 156 are also included.

[0064] In this embodiment, after obtaining the object's motion data in step 104, the method proceeds to step 152. In step 152, it is determined whether the object's motion can be reliably estimated. In other words, the accuracy of the motion data is evaluated. The accuracy of the motion data can be evaluated by monitoring the co-registration quality. If it is determined that the motion can be reliably estimated, the method proceeds to step 154. If it is determined that the motion cannot be reliably estimated, the method proceeds to step 156.

[0065] In step 154, it is determined whether the B0 map can be reliably predicted. This determination may be based on the obtained motion data. For example, if the motion predicted based on the motion data exceeds a predetermined maximum threshold, it can be determined that the B0 map cannot be reliably determined. If it is determined that the B0 map can be reliably estimated, the method proceeds to step 106. If it is determined that the B0 map cannot be reliably estimated, the method proceeds to step 156.

[0066] In step 156, a B0 image is acquired in response to the determination that object motion or a B0 image cannot be accurately determined. A B0 image can be acquired when feasible (e.g., between scans). If acquiring a B0 image is not feasible, the operator of the MRI machine can be notified. This notification may include a warning about excessive object motion.

[0067] In this way, Method 150 provides a means by which an accurate B0 graph can always be obtained, but there is also the possibility of inaccurate motion data or situations where the B0 graph updated by motion cannot be accurately determined.

[0068] Figure 2 A simplified block diagram 200 depicts a B0 prediction model for determining a motion-updated B0 map. An initial B0 map and motion data are input into a B0 prediction model 210, which determines the motion-updated B0 map. Alternatively, an initial MRI image can be input into the B0 prediction model 210.

[0069] B0 prediction model 210 can be configured to model changes in the B0 graph as a fitting result to the motion parameters of the object. This fitting result can be linear or non-linear. As a result, using an initial B0 graph and the object's motion data as input, B0 prediction model 210 can model changes in the initial B0 graph caused by the object's motion. For example, changes in the initial B0 graph can be modeled using the following formula (I): B0 = a + b rot x +c rot y +d rot z (I) In addition, the changes in the B0 graph input into the B0 prediction model 210 can be approximated by a set of spherical harmonic coefficients.

[0070] The B0 prediction module 210 can categorize the identification results of changes in the initial B0 graph into global changes and local changes. That is, changes common across different objects are global changes, while object-specific changes are local changes. In this case, the B0 prediction module 210 determines the global and local changes separately. Specifically, determining the motion-updated B0 graph may include determining the global changes of the initial B0 graph based on lower-order spherical harmonic coefficients and motion data. Conversely, determining the motion-updated B0 graph may include determining the local changes of the initial B0 graph based on higher-order spherical harmonic coefficients and motion data.

[0071] The global ΔB0 predictor 220 can predict global changes in the initial B0 map caused by the object's motion by obtaining the initial B0 map, the object's motion data, and inputting this data into a trained global model 240. Additionally, the global ΔB0 predictor 220 can use the initial MRI image along with the trained global model 240. The trained global model 240 can use lower-order spherical harmonics to determine the global changes in the initial B0 map.

[0072] The local ΔB0 predictor 230 can predict local changes in the initial B0 map caused by the object's motion by obtaining the initial B0 map, the object's motion data, and inputting this data into a trained local model 250. Additionally, the local ΔB0 predictor 230 can use the initial MRI image along with the trained local model 250. The trained local model 250 can use higher-order spherical harmonic coefficients to determine local changes in the initial B0 map.

[0073] The predicted local changes can then be combined with the predicted global changes to obtain the B0 map of motion updates.

[0074] The spherical harmonic coefficients can be determined based on a training algorithm that is configured to receive training inputs, wherein the training inputs include B0 graphs of multiple different objects at multiple different positions (or motion states) and motion parameters corresponding to the B0 graphs.

[0075] In other words, the training algorithm receives the B0 graph and an array of associated motion parameters. The updated B0 graph is then approximated as a set of spherical harmonic coefficients by the training algorithm.

[0076] Similarly, the spherical harmonic coefficients of the trained local model 250 and the trained global model 230 can be determined based on a training algorithm configured to receive training inputs, wherein the training inputs include B0 graphs of multiple different objects at multiple different positions (or motion states).

[0077] Additionally, training input can include synthetic B0 maps generated by the motion frame and the corresponding motion parameters. Acquiring B0 maps of multiple different objects at multiple different locations (or motion states) may require significant scan time. In contrast, synthetic B0 maps can be acquired from the motion frame without any scan time. Furthermore, due to hardware limitations or other reasons, the acquired B0 maps may have many artifacts or inconsistencies. These will be referenced below. Figure 3 and Figure 4 The generation of the synthetic B0 graph is described in more detail.

[0078] As a result, a motion-updated B0 map can be obtained by using spherical harmonic coefficients, the initial B0 map, and the object's motion data.

[0079] The training algorithm can be a known curve-fitting model. Alternatively, the training algorithm can be any machine learning algorithm, such as a neural network. The structure of artificial neural networks is inspired by the human brain. A neural network consists of multiple layers, each containing multiple neurons. Each neuron contains mathematical operations. In particular, each neuron can include different weighted combinations of a single type of transformation (e.g., transformations of the same type, sigmoid, etc., but with different weights). In processing the input data, each neuron performs mathematical operations on the input data to produce a numerical output, and the output of each layer in the neural network is fed sequentially into the next layer. The last layer provides the output. Embodiments of the present invention employ an ANN-based learning algorithm to learn the spherical harmonic function of motion updates in order to reduce the space and time complexity of the model.

[0080] As described above, in order to determine the motion-updated B0 map, magnetic field inhomogeneities must be identified. These magnetic field inhomogeneities are characterized by the magnitude and location of the magnetic field offset with respect to movement (from the initial B0 map). To identify these inhomogeneities, it is necessary to understand how the object's shape, positioning, and motion affect the magnetic field variations in different regions of the object. In some embodiments, this can be understood using polynomial coefficients for each voxel, with motion parameters as input. The B0 prediction model 210 can parametrically learn the changes in the B0 map caused by motion.

[0081] Previous efforts to estimate changes in the B0 diagram due to motion have been limited. However, known methods do not account for both lower-order and higher-order interactions between motion parameters on each axis (which is typical of the motion involved in practice). Furthermore, known methods do not address the generalizability of the approach across objects for structural variations.

[0082] The B0 prediction model 210 can take an initial MRI image, an initial B0 map, and motion data as input. The model can divide the prediction into two levels to isolate parameters that cannot generalize across objects. These two levels can be: a method for predicting global (lower-order) changes that are general across objects; and a method for predicting increasingly object-specific local (higher-order) changes. The model can output a motion-updated B0 map, which can be approximated as a set of spherical harmonic coefficients.

[0083] For global variations of the B0 map that are universal across objects, the trained B0 map (in harmonic space) can be reconstructed solely from lower-order harmonic coefficients. The B0 map of each voxel estimated from each lower-order harmonic can be modeled as the interaction between motions on each axis. The input features can be rotational parameters of a given image and their interactions, and the output can be the predicted magnetic field for a given voxel. For example, this can be represented by the following equation (II): B0 = a + b rot x +c rot y +d rot z +e rot x rot y …+j rot x 2 rot z +… (II) The coefficients of the above formulas can be learned using traditional curve fitting models or neural networks. Then, the coefficient map of each voxel for each harmonic order of the B0 graph can be obtained and stored as a trained global model 240.

[0084] For object-specific local (higher-order) variations in the B0 map, the training B0 map (in harmonic space) is reconstructed only from the higher-order harmonic coefficients. Differential isomorphic co-registration is performed between the anatomically closest training B0 map and each training sample in the training set. Each resulting transformation function is then applied to the atlas coefficient map. Each resulting transformed coefficient map is trained together with spherical harmonics obtained from the initial B0 map to obtain an object-specific transformed coefficient map from within. Subsequently, object-specific transformed higher-order B0 coefficients can be obtained.

[0085] To learn about global and local B0 variations, a set of computed tomography (CT) images and / or magnetic resonance images, along with the coefficient plots of the corresponding B0 plots, can be used.

[0086] Using the initial MRI images, the “n” most similar samples in terms of structure and location, along with their B0 maps, are retrieved from the trained local models 250. This can be used to determine local variations in the initial B0 maps.

[0087] The predicted global changes and predicted local changes from the initial B0 graph can then be combined to create a B0 graph with motion updates.

[0088] Figure 3 A flowchart 300 depicts a method for generating synthetic B0 maps. This method can be used to generate B0 maps of objects at different locations. For example, the method can be used to generate samples of B0 maps of a head at different locations. These samples can then be used to train a B0 prediction model for head locations. In other words, synthetic B0 maps are used to learn how B0 maps change with motion and can ultimately be used to predict B0 maps updated by motion.

[0089] In step 302, a structural image can be obtained. This structural image can be a structural image of the base object, and can be a structural image of a portion of the base object. For example, the structural image can be a structural image of the head of the base object. The structural image can be a CT image or an MRI image of the base object. This structural image can be obtained from medical records or directly from the base object.

[0090] In steps 304 to 320, a motion-simulated structural image can be generated. The motion-simulated structural image can be based on the obtained structural image and a set of motion parameters. This set of motion parameters can include a set of translations, rotations, stretches, and tilts, or a complete deformation field. The motion parameters can be a desired set of motion parameters, including reasonably expected motion of the object. For example, if the structure is a head, this set of motion parameters could be a set of head rotations, but not exceeding 180 degrees on any axis. 0 .

[0091] Specifically, in step 304, a point cloud of the structural image is obtained. A point cloud is a set of data points in space, therefore a 3D structural image can be represented as a set of data points. In step 306, an initial mesh can be obtained from the point cloud. The initial mesh can be determined by inputting the point cloud into many well-known point cloud meshing algorithms. The initial mesh can be, for example, a triangular mesh.

[0092] In steps 308-312, structural regions of the initial mesh can be identified. Structural regions can include rigidly moving regions, non-rigidly moving regions, and fixed regions. For example, when obtaining the initial mesh from a point cloud of structural images representing the head, neck, and shoulders of a base object, the head can be a rigidly moving region, the neck a non-rigidly moving region, and the shoulders a fixed region.

[0093] In steps 314-318, a moving mesh can be generated based on the initial mesh. The moving mesh can be generated based on the identified structural region and one of the motion parameters in this set of motion parameters. Based on one of the motion parameters in this set of motion parameters, a surface modeling method can be used to generate the moving mesh from the initial mesh. In fact, by repeating steps 314-318 for each motion parameter in this set of motion parameters, a set of moving meshes can be obtained.

[0094] Specifically, in step 314, the first part of the moved mesh can be obtained by transforming the rigidly moving region of the initial mesh to a new position based on one of the motion parameters in this set of motion parameters. For example, if the motion parameter is 15 about the x-axis... 0 If rotated, the first part of the moving mesh could be rotated 15 degrees on the x-axis. 0 The rigid moving region of the initial mesh.

[0095] In step 316, the second part of the moving mesh can be obtained by keeping the fixed region of the initial mesh unchanged. Since the fixed region is unaffected by the object's movement (because it is fixed), it is not necessary to transform the initial mesh in any way.

[0096] In step 318, the third part of the moving mesh can be obtained by transforming the non-rigid moving region of the initial mesh so that the moving mesh is a smoothly changing surface mesh.

[0097] In step 320, a motion-simulated structural image can be generated. The motion-simulated structural image can be based on an initial mesh and a moving mesh.

[0098] In step 322, the structural images from the motion simulation can be input into the B0 calculation method. For example, the B0 calculation method can determine a susceptibility map based on each structural image from the motion simulation. Then, based on the susceptibility map, the calculated B0 map can be determined using known methods.

[0099] In step 324, a synthetic B0 map generated by the analysis B0 calculation method based on the structural image of motion simulation can be output.

[0100] As another example Figure 4 Figure 400 presents a method for using the motion simulation framework 410.

[0101] Structural images 420, such as CT images, can be used to generate synthetic B0 maps at various locations. Synthetic B0 maps can represent the head of different objects at various locations. Note that synthetic B0 maps can simulate not only data at different locations but also different patients, i.e., by inducing anatomical variability through stretching and tilting. Synthetic B0 maps can be used to learn how B0 maps change. In general, synthetic B0 maps can be used to predict updated magnetic field maps, which the scanner will use to set the shimming current during scanning and / or retrospective correction of MRI images.

[0102] In an exemplary embodiment of the motion simulation framework 410, the motion simulation framework may perform the following steps: (i) A structural image 420 (CT or MRI) and a set of motion parameters 430 (including rotation, stretching and tilt) can be provided together as input to the motion simulation frame 410.

[0103] (ii) The motion simulation framework 410 then creates a point cloud 440, and subsequently creates an initial mesh 450 based on the structure image 420. The initial mesh may be a triangular mesh.

[0104] (iii) Then the mesh 450 is divided into rigid moving regions, non-rigid moving regions, and fixed regions. In some examples, these could be the head, neck, and shoulders, respectively.

[0105] (iv) Using surface modeling method 460 and the identified region, the mesh of the rigidly moving region is transformed to the new location based on motion parameters.

[0106] (v) The fixed area remains unchanged.

[0107] (vi) Allows non-rigid moving regions to transform to optimal conditions, resulting in a smoothly varying surface mesh.

[0108] (vii) Use the initial mesh and the new mesh to generate a structural image 470 for motion simulation.

[0109] (viii) A synthetic B0 map can be generated using the structural image 470 from motion simulation and the B0 calculation method from analysis.

[0110] Go to Figure 5 A simplified block diagram 500 of a system for prospective B0 map change estimation and motion correction is depicted. Specifically, the system detects the movement of an object in the bore, determines the motion-updated B0 map, and updates the shimming coil current of the MRI accordingly.

[0111] The method for updating the B0 map during an MRI examination of an object, as detailed in Figures 1-4, may further include applying an updated shimming coil current to the shimming coil of the MRI scanner used during the MRI examination. In this case, the updated shimming coil current can be configured to correct for magnetic field inhomogeneities of the MRI scanner based on the motion-updated B0 map. Figure 5 The described system 500 can determine the motion-updated B0 map and accordingly set the shimming coil current of the MRI scanner.

[0112] System 500 can control an MRI scanner to reduce magnetic field inhomogeneities caused by object motion during the main scanning phase 510. The system includes a monitoring unit 520 configured to acquire an initial BO map and an initial MRI image before scanning. System 500 may also include a motion sensor 530 configured to acquire motion data of the object during scanning, and a BO map updater 540 configured to determine a motion-updated BO map based on the initial magnetic field map, the initial MRI image, and the motion data. System 500 may also include a shimming coil controller 550 configured to control the operating parameters of the shimming coils of the MRI scanner based on the motion-updated BO map.

[0113] Therefore, when acquiring MRI images of the subject from system 500, fewer artifacts are present because the operating parameters of the shimming coils can be adjusted for the subject's movement. This is because if the adjusted shimming coils are based on an accurate BoO map, the adjusted shimming coils can reduce the presence of magnetic field inhomogeneities. Typical systems assume that the BoO map is static in time, an assumption that does not hold true for long-duration scans. This is especially true when the subject is a child, for example. Therefore, by using a motion sensor 530 that continuously monitors the subject's motion and a BoO map updater 540 capable of generating motion updates for shimming, the shimming coils can be adjusted to eliminate magnetic field inhomogeneities.

[0114] Additionally, system 500 may include a motion estimator 560 configured to determine the accuracy of motion data acquired from motion sensor 530. Motion estimator 560 may also be configured to control B0 map scanner 570 to acquire B0 maps based on the determined accuracy being below a predetermined threshold accuracy. As a result, even when motion sensor 530 does not capture accurate motion data, a B0 map that can be used by shimming coil controller 550 to perform shimming motion updates can be obtained. When the accuracy of the motion data is sufficient, it may not be necessary to acquire a B0 map from B0 map scanner 570, as B0 map updater 540 can instead determine a B0 map, thereby reducing scanning time.

[0115] Additionally, the motion estimator 560 can be configured to determine the feasibility of predicting motion updates to the B0 map based on motion data. In this case, the motion estimator 560 can also be configured to control the B0 map scanner 570 to acquire the B0 map based on the determined feasibility being below a predetermined threshold.

[0116] In other words, the motion estimator 560 can also assess the accuracy of motion data by monitoring the quality of co-registration. If the co-registration has acceptable quality, the motion estimator 560 can trigger the B0 map updater 540. If the quality of co-registration is unacceptable, the B0 map scanner 570 can be triggered to directly acquire the B0 map. Similarly, given motion data, the motion estimator 560 can also automatically determine whether a prediction of the motion-updated B0 map is feasible. If not, the B0 map scanner 570 can be triggered to directly acquire the B0 map. In all cases, the system retrieves the (predicted or scanned) B0 map reflecting the latest system state for updating the shimming current and performing shimming. In this case, the effectiveness of the motion estimation can also be measured. Using this information, the system allows triggering a B0 map scan when the prediction confidence is low, thereby ensuring that the B0 map estimation is accurate even when motion cannot be accurately estimated.

[0117] The motion sensor 560 can also be configured to determine whether an object has moved and acquire motion data in response to determining that the object has moved. In this way, unnecessary updates to the B0 map can be avoided.

[0118] The operating parameter of the shimming coil can be the shimming coil current. In this case, the shimming coil controller 550 can be configured to apply an updated shimming coil current to the shimming coil, wherein the updated shimming coil current is configured to correct magnetic field inhomogeneities in the magnetic bore based on a motion-updated B0 diagram.

[0119] In another embodiment, the motion sensor 530 can also be configured to determine whether significant movement has occurred based on motion data. If significant movement is detected, an alert can be sent to a user. The user could be a professional operating the MRI machine or an object inside the magnetic cavity. This can provide an alarm to the user, thereby reducing movement.

[0120] In summary, monitoring component 520 can be configured to perform an initial survey scan and acquire an initial B0 map and an initial MRI image. The system can then perform a main scan phase 510. The main scan phase 510 may not include any additional B0 map scans. Throughout the scan phase, motion sensor 530 monitors the object in the bore. Motion sensor 530 can be any motion sensor capable of tracking the movement of the object in the bore, such as a VitalEye. If motion is detected by motion sensor 530 at any point, the motion sensor can trigger motion estimator 560. Motion estimator 560 can reconstruct intermediate images based on subsequent k-space data, calculate the registration between the intermediate images and the initial B0 map scan, and compute motion data.

[0121] The B0 map updater 540 then acquires motion data, the initial B0 map, and the initial MRI image to predict changes in the B0 map relative to the initial B0 map. This can be performed using the method described above with reference to Figures 1-4. The shimming coil controller 550 then uses the motion-updated B0 map field map to calculate a new shimming current, which can then be used for shimming.

[0122] The shimming coil allows the shimming field to be set only at the start of a scan or at any point during the scan. This can be determined by a flag. This flag can be set based on the MRI scanner's ability to update higher-order shims or based on user requirements. If the MRI scanner is able to update higher-order shims in each scan, the currently running scan can be paused until shimming is performed, and then the scan can resume once the shimming has been updated. Motion sensors can notify the user of any movement requiring a rescan via an alarm on the monitor.

[0123] Finally, the raw data from the main scanning unit 510, along with the corresponding B0 residuals obtained, can be sent to the reconstructor 580 for the final reconstruction of the magnetic resonance image.

[0124] Figure 6 A simplified block diagram of a system 600 for retrospective Bo map change estimation and motion correction is depicted. Raw data from the main scan stage 610, along with the initial Bo map 612 and intelligent survey 614, can be provided to a comprehensive reconstruction framework that outputs an image suitable for observation.

[0125] B0 prediction model 622 can be used to retrospectively estimate changes in the B0 map, which can be used for retrospective motion correction in the comprehensive reconstruction framework 620.

[0126] Raw data and an initial B0 map from the main scan phase can be supplied to the reconstructor 624. The raw data may include k-space data and raw motion data of the object captured during the main scan phase 610. Assuming the object did not move at all during the main scan phase 610, the reconstructor 624 can generate MRI images. The motion estimator 626 can then determine the motion data based on the raw data from the main scan phase 610. The B0 prediction model 622 can use the motion data, raw data, and the initial B0 map to determine a motion-updated B0 map. The high-fidelity builder 628 can acquire the initial B0 map, raw data, motion data, hardware defect data from the hardware defect log 630, and the motion-updated B0 map, and generate images for observation.

[0127] In some embodiments, motion data of the object, along with an initial B0 map and an initial MRI image, can be input into a B0 prediction model 622 to predict a motion-updated B0 map. This can then be used to update the predicted magnetic field inhomogeneities. A second reconstruction can be performed using the improved B0 map estimate to improve image quality. Motion data can be updated based on the new image to further improve the B0 map used for reconstruction. This process can be repeated a fixed number of iterations. Alternatively, this process can be repeated until a convergence criterion is met. For example, a convergence criterion can be met when the root mean square deviation between two successive reconstructions is below a threshold. While data-driven motion estimation may be used in some embodiments, motion data captured during examination may also be used in other embodiments.

[0128] The present invention can be a method, system, and / or computer program product. A computer program product may include one or more computer-readable storage media having computer-readable program instructions thereon for causing a processor to execute aspects of the present invention.

[0129] A computer-readable storage medium can be a tangible device capable of retaining and storing instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, 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.

[0130] This document describes various aspects of the invention with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks of the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0131] A single processor or other unit can perform the functions of several items as described in the claims.

[0132] It should be understood that the disclosed method is a computer-implemented method. Therefore, a concept for a computer program is also proposed, comprising code units for implementing any described method when the program is run on a processing system.

[0133] Technicians will be able to easily develop processors for performing any of the methods described herein. Therefore, each step of the flowchart can represent a different action performed by the processor, and can be executed by the corresponding module processing the processor.

[0134] As discussed above, this system utilizes a processor to perform data processing. A processor can be implemented in various ways using software and / or hardware to perform a variety of required functions. A processor typically employs one or more microprocessors, which can be programmed using software (e.g., microcode) to perform the desired functions. A processor can be implemented as a combination of dedicated hardware for performing some functions and one or more programmed microprocessors and associated circuitry for performing other functions.

[0135] Examples of circuits that may be used in various embodiments of this disclosure include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0136] In various implementations, the processor may be 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 may be encoded with one or more programs that, when run on one or more processors and / or controllers, perform the required functions. The various storage media may be fixed within the processor or controller, or may be portable, allowing one or more programs stored thereon to be loaded into the processor.

[0137] Based on a study of the accompanying drawings, the disclosure, and the appended claims, those skilled in the art will be able to understand and implement variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude multiple. A single processor or other unit can perform the functions of several items recited in the claims. The fact that certain measures are recited in mutually different dependent claims does not indicate that combinations of these measures cannot be advantageously used. Computer programs can be stored / distributed on suitable media, such as optical storage media or solid-state media provided with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. If the term "suitable" is used in the claims or description, it should be noted that the term "suitable" is intended to be equivalent to the term "configured as." No reference numerals in the claims should be construed as limiting the scope.

Claims

1. A method for updating a magnetic field B0 map (100) during or during a magnetic resonance imaging (MRI) examination of an object, comprising: Obtain an initial Boolean image of the object associated with the start of the MRI examination; Obtain motion data of the object during the MRI examination; and Based on the B0 prediction model (210), the initial B0 map of the object, and the motion data, a motion-updated B0 map of the object is determined, wherein the B0 prediction model is configured to model changes in the B0 map as a fitting result to the motion parameters of the object, and wherein the changes in the B0 map are approximated by a set of spherical harmonic coefficients. The changes in the B0 graph include global changes common to different objects and local changes specific to each object, and the determination of the B0 graph for the motion update includes: The global changes of the initial B0 diagram of the object are determined based on the lower-order spherical harmonic coefficients and the motion data; and The local variations of the initial B0 diagram of the object are determined based on the higher-order spherical harmonic coefficients and the motion data.

2. The method according to claim 1, wherein, The spherical harmonic coefficients are determined based on a training algorithm of an array configured to receive training inputs, wherein the training inputs include B0 maps of multiple different objects at multiple different locations and motion parameters corresponding to the B0 maps.

3. The method according to claim 2, wherein, The training input also includes a synthetic Bo map generated by the motion frame and the motion parameters corresponding to the synthetic Bo map.

4. The method according to claim 3, wherein, Generating the composite B0 diagram includes: Obtain the structural image of the basic object; A motion-simulated structural image is generated based on the structural image and a set of motion parameters, including rotation, stretching, and tilting; and A synthetic B0 image is generated by inputting the structural image of the motion simulation into the B0 calculation method.

5. The method according to claim 4, wherein, Generating the structural image of the motion simulation includes: Obtain the point cloud of the structure image; Generate an initial mesh based on the point cloud; The initial mesh is identified as including structural regions that exhibit rigid movement, structural regions that exhibit non-rigid movement, and structural regions that exhibit fixed regions. Based on the identified structural regions and the set of motion parameters, a set of moving meshes is generated according to the initial mesh; and The structural image of the motion simulation is generated based on the initial grid and the moving grid.

6. The method according to claim 5, wherein, Generating a mobile grid from the set of mobile grids based on the initial grid includes: The first part of the moving mesh is obtained by transforming the rigid moving region based on one of the motion parameters in the set of motion parameters; The second part of the moving grid is obtained by keeping the fixed area unchanged; and The third part of the moving mesh is obtained by transforming the non-rigid moving region to make the moving mesh a smoothly changing surface mesh.

7. The method according to any one of claims 4-6, wherein, The structural image is a computed tomography (CT) image or an MRI image of the underlying object.

8. The method according to any one of claims 2-6, wherein, The training algorithm is a curve fitting model or a neural network.

9. The method according to any one of claims 1-6, wherein, Obtaining the initial B0 diagram includes: The initial B0 map is acquired from the B0 pre-scan performed at the start of the inspection; or The initial B0 map is predicted based on the survey scan.

10. The method according to any one of claims 1-6, wherein, The B0 map of motion update for the object is also determined based on the initial MRI image obtained at the start of the examination.

11. The method according to any one of claims 1-6, further comprising: An updated shimming coil current is applied to the shimming coil of the MRI scanner used during the MRI examination, wherein the updated shimming coil current is configured to correct for magnetic field inhomogeneities of the MRI scanner based on the motion-updated B0 map.

12. The method according to any one of claims 1-6, further comprising: The MRI images of the subject acquired during the MRI examination are retrospectively reconstructed based on the motion-updated B0 map.

13. A computer program comprising code units configured to implement the method according to any preceding claim when the program is run on a processing system.

Citation Information

Patent Citations

  • Magnetic Resonance Imaging

    US20150362578A1

  • Method and magnetic resonance apparatus for shimming the basic magnetic field by operation of a shim device

    US20180238979A1