Adaptive reconstruction of magnetic resonance images

Abstract

The present disclosure relates to a method comprising: providing a trained machine learning model. The trained machine learning model is configured to reconstruct an image based on input data. The method comprises: receiving (201) a multidimensional matrix comprising M-dimensional acquired data; determining (205) a subset of values ​​of at least one K selected dimension of the matrix; for each value of the subset, determining an M-K dimensional matrix corresponding to the acquired data, resulting in a set of M-1 dimensional matrices; inputting (209) the set of M-K dimensional matrices to the trained machine learning model, and receiving a reconstructed image from the trained machine learning model.

Description

Technical Field

[0001] The present invention relates to a scanning imaging system, in particular to a medical analysis system for reconstructing magnetic resonance images. Background Art

[0002] Magnetic resonance imaging (MRI) scanners rely on a large static magnetic field (B0) to align the nuclear spins of atoms as part of the process of producing images inside a patient. These images can reflect various quantities or properties of the object. Reconstructing images from undersampled k-space data plays an important role in MRI. In particular, deep learning has shown the potential to significantly speed up MRI reconstruction with reduced measurements. ISMRM-2018 abstract (page 2796): 'Integrating spatial and temporal correlation into a deep neural network for low-delay reconstruction of highly undersampled radial dynamic images' (H. Takeshima) discloses a dynamic reconstruction method using a deep neural network (DNN), in which a single image is reconstructed from M consecutive frames and N adjacent frame slices. Summary of the Invention

[0003] Various embodiments provide medical analysis systems, methods, and computer program products for reconstructing magnetic resonance images.

[0004] Embodiments of the present invention may provide a unit for reconstructing an actual (center) line, slice, or slab, wherein correlations with adjacent lines, slices, or slabs are taken into account, respectively. The present invention may achieve a reduction in the dimensionality of the optimization problem in reconstruction, which reduces the computational workload and time for MR image reconstruction. This may be achieved by reducing the dimensionality of a multidimensional matrix of dimension M of the MR image data by one or more dimensions (K>=1 dimension), obtaining a reduced matrix of dimension MK (M minus K), selecting one or more sub-parts of the reduced matrix, wherein each sub-part includes a correlation matrix, and using the sub-parts as input to a model that has been trained to reconstruct an image based on data having the reduced dimensionality MK.

[0005] In one aspect, the present invention relates to a medical analysis system for reconstructing magnetic resonance images. The medical analysis system includes a processor and at least one memory storing machine-executable instructions. The processor is configured to control the medical analysis system. The medical analysis system includes a trained machine learning model, wherein the trained machine learning model is configured to reconstruct an MR image based on input data. The running of the machine-executable instructions causes the processor to: receive a multi-dimensional matrix containing M-dimensional acquisition data, determine a subset of values of at least one selected dimension of the M-dimensional matrix (K selected dimensions, where K >= 1 and K < M), for each value of each subset in the at least one subset, determine an M-K-dimensional matrix containing the acquired data corresponding to the value, obtain a set of M-K-dimensional matrices, input the set of M-K-dimensional matrices into the trained machine learning model, and receive an image reconstruction output from the trained machine learning model.

[0006] For example, if two dimensions are selected, i.e., K = 2, two subsets of values can be determined, one subset for each selected dimension. For each pair of a first value of one of the two subsets and a second value of the other of the two subsets, an M-2-dimensional matrix can be determined. For example, if the first subset has two values and the second subset has three values, six possible pairs can be defined from these two subsets, and thus six M-2 matrices can be determined for these six pairs. If three dimensions are selected, i.e., K = 3, three subsets of values can be determined, one subset for each selected dimension. For each triple of a first value of one of the three subsets and second and third values of the other two of the three subsets, an M-3-dimensional matrix can be determined. For example, if one dimension is selected, i.e., K = 1, one subset of values can be determined. For each value of the subset, an M-1-dimensional matrix can be determined, and so on.

[0007] When increasing the dimension of the problem, a trained model (e.g., a deep learning model) may perform better. For example, a 3D problem can be solved by using pure 3D reconstruction to process 3D undersampled Cartesian scans. This may result in better image quality compared to solving the �D problem slice by slice (i.e., N 2D problems). However, solving N 2D problems may be more practical than solving the entire 3D problem at once. This may be more practical in terms of computational time and memory during training and inference, thus allowing the application of more powerful deep learning solutions.

[0008] The present subject matter can enable decomposition of an M-dimensional problem into multiple MK-dimensional problems. An example is accelerated Cartesian 3D scanning, where k-space is fully measured in the readout direction. The latter allows splitting the optimization problem along the readout direction, thus decomposing the 3D reconstruction problem into Nread_out 2D problems. By decomposing the problem into smaller problems, the trained machine learning model may have to process less data, thus potentially reducing computational and memory load. Computational load and memory consumption can be important factors during inference (e.g., during reconstruction of images on the scanner) and during training. This can be particularly advantageous for deep neural networks for the following reasons. The size of the network(s) in a deep learning solution is limited by hardware constraints during training and on the scanner (e.g., the availability of computational power of GPU memory). Since performance in terms of output quality typically improves with increasing model size, it can be beneficial to use as large a network as possible. Therefore, solving smaller problems may allow the use of larger models and therefore improve output quality. The present subject matter can also optimize the solution of reduced multiple MK-dimensional problems. In particular, the present subject matter can employ an "N-and-a-half" approach to use correlations between N(MK)-dimensional data to solve an M-dimensional problem (for M=3, the approach can be a 2.5D approach). By using information from a few related slabs or slices (e.g., adjacent 2D slices may be related slices) rather than the entire M-dimensional volume, the correlations between slices or slabs can be used to improve the image quality of the actual individual slices or slabs. This "N-and-a-half" approach can be used when the problem can be separated along one or more dimensions.

[0009] For example, an MRI scanner can be used to scan or image a target volume in a subject (e.g., a brain) to acquire data. This may result in M-dimensional acquired data. The acquired data may include k-space data (e.g., undersampled k-space data) or other image data, such as aliased image data. K-space data can be defined herein as the recorded measurement results of radio frequency signals emitted by atomic spins using the antenna of a magnetic resonance device during a magnetic resonance imaging scan.

[0010] The trained machine learning model can be configured to provide an image reconstruction output in response to receiving a set of X-dimensional matrices of data. The image reconstruction output is the result of reconstructing the image data of a single X-dimensional (XD) matrix in the set of X-dimensional matrices and taking into account the correlation between the single XD matrix and the remaining XD matrices in the set of XD matrices. The X dimension is provided so that MK=X. The machine learning model can be trained to directly learn the mapping between the input (e.g., undersampled k-space data or aliased images) and the output MR image. For example, the image reconstruction output can be a magnetic resonance image, such as a reconstructed one-dimensional, two-dimensional, or three-dimensional visualization of anatomical data contained within the k-space data. For example, the undersampled k-space data is used as the input to the trained machine learning model, and the desired image from the fully sampled k-space data can be the reconstructed image output.

[0011] Each matrix in the set of MK matrices may comprise at least one dimension representing spatial frequency information in one of three directions of the object. According to one embodiment, K=1, so that one dimension is selected and a subset of values ​​along the selected dimension is determined.

[0012] According to one embodiment, execution of the machine executable instructions further causes the processor to determine a subset of values ​​such that the correlation between the MK-dimensional matrices associated with the values ​​of the subset (or a given value) and the remaining MK-dimensional matrices in the set is higher than a predefined threshold.

[0013] Correlations along dimensions can be very strong, especially in a local manner (e.g., adjacent slices), so not accounting for them can limit the output quality.

[0014] This embodiment may enable solving partial multidimensional problems where data from all dimensions is not fully considered, but only "local" data, i.e., when optimizing for a specific multidimensional problem at a point in the (MK) dimension, the data in the domain of that point is also used as input to train the machine learning model. For example, instead of passing data for a single point in the readout direction (e.g., through a single ky / kz space), data for a range in the readout direction is provided to the trained machine learning model. The output of the trained machine learning model may still be a solution to the original multidimensional problem, but now the machine learning model can exploit the (local) correlations that exist along the readout direction.

[0015] The correlation between MK-dimensional matrices can, for example, be position-dependent. For example, due to the lack of inter-slice gaps, 2D adjacent slices of a 3D acquisition may be highly correlated. In particular, due to the slow spatial changes of the scanned object, adjacent 2D slices may be very similar. For example, if M = 3, and a 3D k-space k(x, y, z) matrix is ​​provided with a selected dimension x (and x = x1, x2, x3...), then the slices k(x1, y, z), k(x2, y, z) and k(x3, y, z) corresponding to successive values ​​of x1, x2 and x3 can be related slices. The 2D slices k(x1, y, z), k(x2, y, z) and k(x3, y, z) can therefore be a set of two-dimensional matrices. In another example, if M = 4, and a 4D k-space k(x, y, z, t) matrix is ​​provided, where the selected dimension is time t (and t = t1, t2, t3, ...), then the 3D slabs k(x, y, z, t1) and k(x, y, z, t4) can be correlated sets of three-dimensional matrices. Correlations can occur between slabs acquired at different time points, which are not necessarily consecutive. For example, when dealing with cardiac imaging, correlations are strongest between adjacent phases in the cardiac cycle, which may not be measured consecutively in time.

[0016] According to one embodiment, execution of the machine executable instructions further causes the processor to determine additional subsets of values ​​and repeat step d) of determining the MK-dimensional matrix and step e) of inputting for each additional subset in the additional subsets. Steps d) and e) may be repeated such that an image reconstruction output may be provided for each MK-dimensional matrix of the M-dimensional matrix. For example, if M=3 and K=1, each 2D slice along the selected dimension may be input to the trained machine learning model along with one or more slices associated with each 2D slice. For example, if the number of 2D slices along the selected dimension is 100, steps d) and e) may be repeated 100 times, resulting in 100 reconstructed image outputs (2D slice images).

[0017] According to one embodiment, the subsets are non-overlapping subsets. This can further save processing resources while still providing accurate reconstruction results. For example, if M=3 and K=1, only one slice of the relevant slices of the 2D slice subset can be reconstructed and can represent all slices of the subset. If the number of 2D slices along the selected dimension is 100, steps d) and e) can be repeated less than 100 times (which is the number of all subsets of values ​​of the selected dimension), for example, if the first selected subset has 3 slices, the next subset can be selected from the remaining 100-3 slices, if the next subset includes 5 slices, the next subset can be selected from the remaining 100-8 slices, and so on.

[0018] According to one embodiment, the M-dimensionally acquired data is 3D k-space data (M=3), wherein the selected dimension is the readout direction (K=1), wherein the set of M-1 dimensional matrices represents a corresponding set of 2D slices.

[0019] For example, M=3 dimensions correspond to the spatial dimensions kx, ky and kz of k-space.The 3D acquired data may be k-space data.

[0020] This embodiment may be advantageous for the following reasons. By using a 2D neural network model to squeeze out the maximum performance from the 2D neural network, the present subject matter can be used with limited memory on the GPU. This is because using a 3D neural network model may quickly run into memory issues when passing a full image or patch through a 3D neural network.

[0021] According to one embodiment, the subset of values ​​includes a predefined odd number of values. This can enable highly correlated slices to be obtained, particularly if the actual value is the center value of the determined subset. In another example, the subset of values ​​can include a randomly selected number of values ​​that is less than a predefined maximum number.

[0022] According to one embodiment, determining a subset of values ​​includes identifying a central value and two or more surrounding values ​​of the central value along a selected dimension. The subset includes the central value and the surrounding values. This allows for a systematic approach to selecting the subset, which can further improve the performance of the subject matter. This can be particularly advantageous because adjacent slices can be highly correlated.

[0023] According to one embodiment, the trained machine learning model is a deep neural network (DNN) model. For example, the input layer of the DNN includes nodes, each of which is configured to receive the value of an element of a matrix from a set of MK-dimensional matrices. In another example, the trained machine learning model can be an ISTA-net model. This can use a deep learning neural network to rapidly reconstruct magnetic resonance images from sparsely sampled MR data in k-space.

[0024] In another aspect, the present invention relates to a method comprising: providing a trained machine learning model, wherein the trained machine learning model is configured to reconstruct an image based on input k-space data; receiving a multidimensional matrix comprising M-dimensional acquired data; determining a subset of values ​​of K selected dimensions of the matrix, wherein K is an integer greater than or equal to 1 and less than M; for each value of the subset, determining an MK-dimensional matrix corresponding to the value in the acquired data, thereby generating a set of MK-dimensional matrices; inputting the set of MK matrices into the trained machine learning model, and receiving a reconstructed image from the trained machine learning model.

[0025] In another aspect, the present invention relates to a computer program product comprising machine-executable instructions for execution by a processor, wherein execution of the machine-executable instructions causes the processor to perform the method of any one of the preceding embodiments.

[0026] In one aspect, the present invention relates to a medical analysis system for reconstructing magnetic resonance images. The medical analysis system includes a processor and at least one memory storing machine-executable instructions. The processor is configured to control the medical analysis system. The medical analysis system includes a trained machine learning model, wherein the trained machine learning model is configured to reconstruct an image based on input data. The execution of the machine-executable instructions causes the processor to: receive a dimensional matrix containing M-dimensional multidimensional acquisition data, determine a subset of values ​​of a selected dimension of the M-dimensional matrix, determine an M-1 dimensional matrix containing acquisition data corresponding to each value of the subset, obtain a set of M-1 dimensional matrices, input the set of M-1 dimensional matrices into the trained machine learning model, and receive an image reconstruction output of the trained machine learning model.

[0027] It should be understood that one or more of the aforementioned embodiments of the present invention may be combined as long as the combined embodiments are not mutually exclusive. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Preferred embodiments of the present invention will hereinafter be described, by way of example only, and with reference to the accompanying drawings, in which:

[0029] Figure 1 is a schematic diagram of a control system according to the present subject matter.

[0030] Figure 2 A flowchart of a method for reconstructing a magnetic resonance image according to an example of the present subject matter,

[0031] Figure 3 depicts a graph illustrating the expected gain in image quality (IQ), increase in memory cost, and computational load using the present subject matter,

[0032] Figure 4 A cross-section and functional diagram of an MRI system is shown.

[0033] Reference Signs List

[0034] 100 Medical System

[0035] 101 Scanning Imaging System

[0036] 103 processor

[0037] 107 Memory

[0038] 108 Power Supply

[0039] 109 bus

[0040] 111 Control System

[0041] 121 Software

[0042] 125 Display

[0043] 129 User Interface

[0044] 133 Database

[0045] 150 AI components

[0046] 160 Machine Learning Models

[0047] 201-209 Methods and Steps

[0048] 301 Storage Cost

[0049] 303 Calculating Load

[0050] 700 MRI system

[0051] 704 magnet

[0052] 706 Magnet Bore

[0053] 708 Imaging Area

[0054] 710 Magnetic Field Gradient Coil

[0055] 712 Magnetic Field Gradient Coil Power Supply

[0056] 714 RF Coil

[0057] 715 RF Amplifier

[0058] 718 objects DETAILED DESCRIPTION

[0059] In the following, elements with the same number in the drawings are either similar elements or perform equivalent functions. If the functions are equivalent, elements that have been discussed previously will not necessarily be discussed in later drawings.

[0060] Various structures, systems and devices are schematically depicted in the drawings for purposes of explanation only and so as to not obscure the present invention with details that are well known to those skilled in the art. Nevertheless, the drawings are included to describe and explain illustrative examples of the disclosed subject matter.

[0061] Figure 1is a schematic diagram of a medical analysis system 100. The medical analysis system 100 includes a control system 111 configured to be connected to a scanning imaging system (or acquisition component) 101. The control system 111 includes a processor 103 and a memory 107, each of which is capable of communicating with one or more components of the medical system 100. For example, the components of the control system 111 are coupled to a bidirectional system bus 109.

[0062] It will be understood that the methods described herein are at least partially non-interactive and automated by a computer system. For example, the methods may be further implemented in software 121 (including firmware), hardware, or a combination thereof. In an exemplary embodiment, the methods described herein are implemented in software as an executable program and run by a special-purpose or general-purpose digital computer (e.g., a personal computer, workstation, minicomputer, or mainframe computer).

[0063] The processor 103 is a hardware device for executing software, particularly software stored in the memory 107. The processor 103 can be any custom or commercially available processor, a central processing unit (CPU), a secondary processor among multiple processors associated with the control system 111, a semiconductor-based microprocessor (in the form of a microchip or chipset), a microprocessor, or any device generally used to run software instructions. The processor 103 can control the operation of the scanning imaging system 101.

[0064] The memory 107 may include any one or a combination of volatile memory elements (e.g., random access memory (RAM, such as DRAM, SRAM, SDRAM, etc.) and non-volatile memory elements (e.g., ROM, erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), and programmable read-only memory (PROM)). Note that the memory 107 may have a distributed architecture in which various components are remote from one another but can be accessed by the processor 103. The memory 107 may store instructions or data related to at least one other component of the medical analysis system 100.

[0065] The control system 111 may further include a display device 125, which displays characters and images on a user interface 129. The display device 125 may be a touch screen display device.

[0066] The medical analysis system 100 may further include a power supply 108 for powering the medical analysis system 100. The power supply 108 may be, for example, a battery or an external power source, such as power supplied by a standard AC outlet.

[0067] The scanning imaging system 101 may include at least one of an MRI, CT, and PET-CT imager. The control system 111 and the scanning imaging system 101 may or may not be integrated. In other words, the control system 111 may or may not be external to the scanning imaging system 101.

[0068] Scanning imaging system 101 includes components controllable by processor 103 to configure scanning imaging system 101 to provide image data to control system 111. Configuration of scanning imaging system 101 may enable operation of scanning imaging system 101. Operation of scanning imaging system 101 may, for example, be automatic. Figure 4 An example of components of a scanning imaging system 101 is shown as an MRI system.

[0069] The connection between the control system 111 and the scanning imaging system 101 may include, for example, a general Ethernet connection, a WAN connection, or an Internet connection.

[0070] In one example, the scanning imaging system 101 can be configured to provide output data, such as images, in response to specified measurements. The control system 111 can be configured to receive data, such as MR image data, from the scanning imaging system 101. For example, the processor 103 can be adapted to receive information (automatically or upon request) from the scanning imaging system 101 in a compatible digital form so that the information can be displayed on the display device 125. Such information can include operating parameters, warning notifications, and other information related to the use, operation, and function of the scanning imaging system 101.

[0071] The medical analysis system 100 can be configured to communicate with other scanning imaging systems 131 and / or databases 133 via a network 130. The network 130 includes, for example, a wireless local area network (WLAN) connection, a WAN (wide area network) connection, a LAN (local area network) connection, or a combination thereof. The database 133 can include information related to the patient, the scanning imaging system, the anatomy, the scan geometry, the scan parameters, the scan, etc. The database 133 can include, for example, an electronic medical record (EMR) database including the patient's EMR, a radiology information system database, a medical image database, a PACS, a hospital information system database, and / or other databases including data for planning scan geometries. The database 133 can include, for example, a training data set used for training performed by the present subject matter.

[0072] Memory 107 may also include an artificial intelligence (AI) component 150. AI component 150 may or may not be part of software component 121. AI component 150 may, for example, include a trained machine learning model 160. Trained machine learning model 160 may be configured to receive a set of X-dimensional matrices and provide an image reconstruction output representing one of the matrices in the set. For example, the set of X-dimensional matrices may include three matrices: Arr1, Arr2, and Arr3. The trained model may reconstruct the image output of matrix Arr1, while the other matrices, Arr2 and Arr3, may be used to account for their correlation with Arr1 during reconstruction. Thus, the three matrices may be input to the trained model to receive the image reconstruction output of matrix Arr1. The image reconstruction output may be a subset or portion of the final output. For example, the image reconstruction output may be an image of a range within the total volume. There may be a mapping of a subrange of the M-dimensional input to a subrange of the P-dimensional output.

[0073] The machine learning model 160 can be trained, for example, by providing multiple sets of XD matrices and associating each set of X-dimensional matrices with a matrix of a fully sampled k-space reconstructed image, wherein the fully sampled k-space reconstructed image corresponds to a matrix (Arr1) in the set of X-dimensional matrices.

[0074] Figure 2 is a flow chart of a method for reconstructing a magnetic resonance image of a target volume, such as a brain of a subject. Figure 4 ) can be configured to acquire data by scanning or imaging a target volume of an object.

[0075] In step 201, an M-dimensional matrix including acquired data (by an MRI system) may be received. The acquired data may, for example, be k-space data or other image data, such as aliased image data. The acquired data may, for example, include at least one dimension representing one of three directions of the target volume. The M dimensions may include additional dimensions, such as a time dimension. In one example, M >= 2.

[0076] Assume for simplicity that M = 3, which corresponds to the spatial dimensions, for example, kx, ky, and kz. In this case, the M-dimensional matrix can be defined as mr(x, y, z), where x has multiple Sx values, y has Sy values, and z has Sz values, i.e., the size of the matrix mr is Sx*Sy*Sz.

[0077] In step 203, at least one dimension of the M dimensions may be selected or identified (i.e., K dimensions may be selected, where K>=1). In this example, K=1, but is not limited thereto. For example, the selected dimension may be a dimension selected by a user, for example, user input may be received. The user input indicates the selected dimension. In another example, automatic selection of the dimension may be performed, for example, random selection may be performed. For simplicity of description, the selected dimension may be dimension x of the matrix mr. The selected dimension has a corresponding set of values ​​in the M-dimensional matrix. The set of values ​​includes Sx values, where Sx is, for example, the number of phase encodings in the case where the selected dimension is a dimension corresponding to phase encoding.

[0078] The dimensions may be selected such that the M-1 dimensional matrix has dimension M-1=X, where X is the dimension of the set of matrices used as input to the trained machine learning model 160. Each matrix in the set of M-1 dimensional matrices may include at least one dimension representing spatial frequency information in one of three directions of the target volume.

[0079] A subset of values ​​in the set of values ​​can be determined in step 205. Following the above example, the subset of values ​​can be, for example, a subset of Sx values. The subset of values ​​can be determined as follows: the first value (or current value) of the subset is selected from the set of Sx values, and the remaining values ​​of the subset are also selected. The remaining values ​​and the current value may form a range (the remaining values ​​and the current value are consecutive values ​​of the matrix), or they may not form a range. In one example, the remaining values ​​can be selected so that the current value is the center value. For example, assume that a set of values ​​is provided in the following order: v1, v2, v3, ... vSx. If the current value is v4, the remaining values ​​can be v3 and v5, or can be v2, v3, v5, and v6, where v4 is the center value of the subset defined by v3, v4, and v5, or by v2, v3, v4, v5, and v6. If, in one example, the current value being processed is the first value v1 (or the last value vSx), the determined subset can include values ​​other than v1 and v1, such as v2 and v3.

[0080] Each value vx in the set of Sx values ​​can define a two-dimensional matrix mrx(y,z) associated with the value vx. The two-dimensional matrix mrx(y,z) can represent a two-dimensional slice at the value vx. For example, the value v4 can be associated with the two-dimensional matrix mr4(y,z), v3 can be associated with the two-dimensional matrix mr3(y,z), and v2 can be associated with the 2D matrix mr2(y,z), and so on. In one example, the remaining values ​​associated with each current value can be determined by determining the correlation between the two-dimensional matrix of each value in the remaining values ​​and the two-dimensional matrix of the current value. According to the above example, if the current center value is v4, the method can determine the correlation value between the two-dimensional 2D matrix associated with each value and the 2D matrix mr4(y,z) for each value in v1 to vSx (excluding v4). Only values ​​vx associated with correlation values ​​above a predetermined threshold can be selected. Those selected values ​​and the current value can form a subset.

[0081] In one example, the subset of values ​​may include a predefined number of values, such as 3. That is, the subset may include the current center value and two surrounding values. This may be advantageous because it may implement an optimal selection method while still considering the correlation between the remaining values ​​and the center value.

[0082] For each value in the subset, an M-1 dimensional matrix including the collected data corresponding to the value can be determined in step 207. This may result in a set of M-1 dimensional matrices. Following the above example, if the subset determined in step 205 is [v3, v4, v5], then the three matrices mr3(y, z), mr4(y, z), and mr5(y, z) can be the determined M-1 dimensional matrices for the three values ​​of the subset [v3, v4, v5].

[0083] This set of M-1 dimensional matrices may be input to the trained machine learning model 160 to receive an output of a reconstructed image from the trained machine learning model in step 209. Following the above example, a 2D image of the 2D slice mr4(y,z) associated with the current value v4 may be reconstructed.

[0084] In one example, steps 205-209 can be repeated for additional subsets of values ​​for the selected dimension. Following the above example, a first run of the method can be performed for value v1 in the set of values ​​v1 to vSx. Steps 205-209 can be repeated for each value in the set of values ​​v2 to Sx. This may result in a 2D slice reconstructed from Sx.

[0085] Figure 3 Depicted is a block diagram illustrating the expected gain of increases in memory cost 301 and computational load 303 relative to the size of input data provided to a trained machine learning model. Figure 3Three different input sizes are shown, namely one 2D slice, five 2D slices, and a full 3D volume. Figure 3 The expected gains, the "cost" in terms of memory, and the computational load versus input data size are explained.

[0086] Figure 4 A magnetic resonance imaging system 700 is shown as an example of medical system 100. The magnetic resonance imaging system 700 includes a magnet 704. Magnet 704 is a superconducting cylindrical magnet having a bore 706 therein. It is also possible to use different types of magnets; for example, split cylindrical magnets and so-called open or sealed magnets can also be used. Split cylindrical magnets are similar to standard cylindrical magnets, except that the cryostat is divided into two sections to allow access to the magnet's isoplane. Such magnets can be used, for example, in conjunction with charged particle beam therapy. An open magnet has two magnet sections, one above the other, with a space in between large enough to accommodate an object 718 to be imaged. The arrangement of these two sections is similar to that of Helmholtz coils. Inside the cylindrical magnet's cryostat is a collection of superconducting coils. Within the bore 706 of the cylindrical magnet 704 is an imaging region or volume or anatomical structure 708, where the magnetic field is sufficiently strong and uniform to perform magnetic resonance imaging.

[0087] Also within the bore 706 of the magnet is a magnetic field gradient coil assembly 710 that is used during the acquisition of magnetic resonance data to spatially encode magnetic spins of a target volume within an imaging volume or examination volume 708 of the magnet 704. The magnetic field gradient coils 710 are connected to a magnetic field gradient coil power supply 712. The magnetic field gradient coils 710 are intended to be representative. Typically, the magnetic field gradient coils 710 comprise a set of three separate coils for encoding in three orthogonal spatial directions. A magnetic field gradient power supply supplies current to the magnetic field gradient coils. The current supplied to the magnetic field gradient coils 710 is controlled over time and can be ramped or pulsed.

[0088] The MRI system 700 also includes an RF transmit coil 714 positioned at the subject 718 and adjacent to the examination region 708 for generating RF excitation pulses. The RF coil 714 may include, for example, a set of surface coils or other specialized RF coils. The RF coil 714 may be used alternately for transmitting RF pulses and for receiving magnetic resonance signals. For example, the RF coil 714 may be implemented as a matrix transmit coil comprising multiple RF transmit coils. The RF coil 714 is connected to one or more RF amplifiers 715.

[0089] The magnetic field gradient coil power supply 712 and the RF amplifier 715 are connected to the hardware interface of the control system 111. The memory 107 of the control system 111 may, for example, include a control module. The control module contains computer-executable code that enables the processor 103 to control the operation and functions of the magnetic resonance imaging system 700. The computer-executable code also enables basic operations of the magnetic resonance imaging system 700, such as the acquisition of magnetic resonance data.

[0090] As will be appreciated by those skilled in the art, several aspects of the present invention may be implemented as apparatus, methods, or computer program products. Thus, aspects of the present invention may take the form of entirely hardware embodiments, entirely software embodiments (including firmware, resident software, microcode, etc.), or embodiments combining software and hardware aspects, which may be collectively referred to herein as "circuits," "modules," or "systems." Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer-readable media having computer executable code embodied thereon.

[0091] Any combination of one or more computer-readable media can be used. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. As used herein, "computer-readable storage medium" includes any tangible storage medium that can store instructions that can be executed by a processor of a computing device. The computer-readable storage medium can be referred to as a "computer-readable non-transient storage medium." The computer-readable storage medium can also be referred to as a tangible computer-readable medium. In some embodiments, a computer-readable storage medium can also store data that can be accessed by the processor of the computing device. Examples of computer-readable storage media include, but are not limited to, floppy disks, magnetic hard disk drives, solid-state hard disks, flash memory, USB thumb drives, random access memories (RAM), read-only memories (ROM), optical disks, magneto-optical disks, and register files of processors. Examples of optical disks include compact discs (CDs) and digital versatile discs (DVDs), such as CD-ROMs, CD-RWs, CD-Rs, DVD-ROMs, DVD-RWs, or DVD-R disks. The term computer-readable storage medium also refers to various types of recording media that can be accessed by the computer device via a network or communication link. For example, data can be retrieved via a modem, via the Internet, or via a local area network. Computer executable code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0092] A computer-readable signal medium may include a propagated data signal having computer-executable code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that is not a computer-readable storage medium and that is capable of conveying, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0093] "Computer memory" or "memory" is an example of a computer-readable storage medium. Computer memory is any memory that is directly accessible to a processor. "Computer storage" or "storage" is another example of a computer-readable storage medium. Computer storage is any non-volatile computer-readable storage medium. In some embodiments, computer storage may also be computer memory, or vice versa.

[0094] As used herein, "processor" encompasses an electronic component capable of executing a program or machine-executable instructions or computer-executable code. References to a computing device comprising a "processor" should be interpreted as being capable of comprising more than one processor or processing core. The processor may, for example, be a multi-core processor. A processor may also refer to a collection of processors within a single computer system or distributed across multiple computer systems. The term computing device should also be interpreted as possibly referring to a collection or network of computing devices, each of which includes a processor or multiple processors. The computer-executable code may be run by multiple processors, which may be within the same computing device or may even be distributed across multiple computing devices.

[0095] The computer executable code may include machine executable instructions or programs that cause a processor to perform various aspects of the present invention. The computer executable code for performing operations for various aspects of the present invention may be written in any combination of one or more programming languages ​​(including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages) and compiled into machine executable instructions. In some cases, the computer executable code may be used in the form of a high-level language or in precompiled form and in conjunction with an interpreter that generates machine executable instructions on the fly.

[0096] The computer executable code may be run entirely on the user's computer, partially on the user's computer, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server as a stand-alone software package. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or a connection that can be made to an external computer (e.g., over the Internet using an Internet service provider).

[0097] Aspects of the present invention are described with reference to the flow diagram and / or block diagram according to the method, device (system) and computer program product of an embodiment of the present invention.It will be understood that each frame of flow diagram, diagram and / or block diagram or the part of frame can be implemented by the computer program instruction in the form of computer executable code when applicable.It should also be understood that, when not mutually exclusive, in different flow diagrams, the combination of blocks in the diagram and / or block diagram can be combined.These computer program instructions can be provided to the processor of general-purpose computer, special-purpose computer or other programmable data processing devices to produce machine, so that the instruction running via the processor of computer or other programmable data processing devices creates the unit for implementing the function / action specified in one or more frames of flow diagram and / or block diagram.

[0098] These computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing apparatus, or other device to work in a specific manner, so that the instructions stored in the computer-readable medium produce an article of manufacture including instructions for implementing the functions / actions specified in the flowchart and / or one or more block diagram blocks.

[0099] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operating steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions running on the computer or other programmable apparatus provide a process for implementing the functions / actions specified in the flowchart and / or one or more block diagram blocks.

[0100] As used in this article, "user interface" is an interface that allows a user or operator to interact with a computer or computer system. "User interface" can also be referred to as a "human-machine interface device". The user interface can provide information or data to the operator and / or receive information or data from the operator. The user interface can enable the input from the operator to be received by the computer, and output can be provided to the user from the computer. In other words, the user interface can allow the operator to control or manipulate the computer, and the interface can allow the computer to indicate the effect of the operator's control or manipulation. The display of data or information on a display or a graphical user interface is an example of providing information to the operator. Receiving data by a keyboard, mouse, trackball, touchpad, pointing stick, graphic input board, joystick, game pad, webcam, helmet, gear lever, steering wheel, pedals, wired gloves, dance board, remote controller and accelerometer is an example of a user interface component that receives information or data from the operator.

[0101] As used herein, "hardware interface" encompasses an interface that enables a processor of a computer system to interact with or control an external computing device and / or apparatus. A hardware interface allows a processor to send control signals or instructions to an external computing device and / or apparatus. A hardware interface also enables a processor to exchange data with an external computing device and / or apparatus. Examples of hardware interfaces include, but are not limited to, a universal serial bus, an IEEE 1394 port, a parallel port, an IEEE 1284 port, a serial port, an RS-232 port, an IEEE-488 port, a Bluetooth connection, a wireless LAN connection, a TCP / IP connection, an Ethernet connection, a control voltage interface, a MIDI interface, an analog input interface, and a digital input interface.

[0102] As used herein, a "display" or "display device" encompasses an output device or user interface suitable for displaying images or data. A display can output visual, audio, and tactile data. Examples of displays include, but are not limited to, computer monitors, television screens, touch screens, tactile electronic displays, Braille screens, cathode ray tubes (CRTs), memory tubes, bi-stable displays, electronic paper, vectorscopes, flat panel displays, vacuum fluorescent displays (VFs), light emitting diode (LED) displays, electroluminescent displays (ELDs), plasma display panels (PDPs), liquid crystal displays (LCDs), organic light emitting diode displays (OLEDs), projectors, and head-mounted displays.

[0103] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The invention is not limited to the disclosed embodiments.

[0104] Those skilled in the art will be able to understand and implement other variations to the disclosed embodiments when practicing the claimed invention by studying the drawings, the disclosure and the claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "one" or "an" does not exclude a plurality. A single processor or other unit can implement the functions of several items recited in the claims. Although specific measures are recited in mutually different dependent claims, this does not indicate that a combination of these measures cannot be used to advantage. The computer program can be stored / distributed on a suitable medium such as an optical storage medium or a solid-state medium provided together with other hardware 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. Any figure marks in the claims should not be interpreted as limiting the scope.

Claims

1. A medical analysis system (100) for reconstructing a magnetic resonance image, the medical analysis system comprising a processor (103) and at least one memory (107) storing machine-executable instructions, the processor being configured to control the medical analysis system (100), wherein: Execution of the machine executable instructions causes the processor (103): providing a trained machine learning model configured to reconstruct an image from input data; receiving (201) an M-dimensional multidimensional matrix containing M-dimensional collected data; determining (205) at least a subset of values ​​of at least one selected dimension of the M-dimensional matrix; For each value of each subset in the at least one subset, determining (207) an MK-dimensional matrix including the collected data corresponding to the value, to obtain a set of MK-dimensional matrices, where K≥1; The set of MK-dimensional matrices is input (209) to the trained machine learning model and an image reconstruction output is received from the trained machine learning model, and wherein execution of the machine-executable instructions further causes the processor to determine the subset of values ​​such that the correlation between the MK-dimensional matrix associated with the center value of the subset and the remaining MK-dimensional matrices in the set is above a predefined threshold.

2. The system according to claim 1, wherein: Execution of the machine executable instructions further causes the processor to determine additional subsets of values ​​and to repeat steps d) and e) for each additional subset of the additional subsets. The system of claim 2 , wherein the subsets are non-overlapping subsets.

4. The system according to any one of claims 1 to 3, wherein the M-dimensional acquired data is 3D k-space data, M=3, wherein The selected dimension K is the readout direction, K=1, wherein the set of M-1 dimensional matrices represents a corresponding set of 2D slices.

5. The system of any one of claims 1-3, the subset of values ​​comprising a predefined odd number of values.

6. The system of any one of claims 1-3, determining the subset of values ​​comprises identifying a central value along the selected dimension and two or more surrounding values ​​of the central value, the subset comprising the central value and the surrounding values.

7. The system according to any one of claims 1-3, wherein the trained machine learning model is a deep neural network (DNN) model.

8. A method for reconstructing a magnetic resonance image, comprising: providing a trained machine learning model configured to reconstruct an image from input k-space data; receiving (201) an M-dimensional multidimensional matrix containing M-dimensional collected data; determining (205) at least a subset of values ​​of at least one selected dimension of the M-dimensional matrix; For each value of each subset in the at least one subset, determining (207) an MK-dimensional matrix comprising the collected data corresponding to the value, thereby generating a set of MK-dimensional matrices, where K≥1; inputting (209) the set of MK-dimensional matrices to the trained machine learning model and receiving a reconstructed image from the trained machine learning model, and The subset of values ​​is determined such that a correlation between an MK-dimensional matrix associated with a center value of the subset and the remaining MK-dimensional matrices in the set is above a predefined threshold.

9. A computer program product comprising machine-executable instructions for execution by a processor, wherein: Execution of the machine-executable instructions causes the processor to perform at least part of the method of claim 8.

10. A magnetic resonance imaging (MRI) system (700), comprising the system according to any one of the preceding claims 1-7, wherein the MRI system is configured to acquire M-dimensional data.

Images