Bipolar parallel imaging reconstruction method and device

Abstract

The invention provides a bipolar parallel imaging reconstruction method and device, and relates to the technical field of magnetic resonance imaging. The bipolar parallel imaging reconstruction method comprises the following steps: acquiring magnetic resonance imaging data and reference data; bipolar convolution kernel design parameters are determined according to actual scanning parameters of the acquired magnetic resonance imaging data, and a bipolar convolution kernel is constructed according to the bipolar convolution kernel design parameters; training the bipolar convolution kernel based on the reference data; processing the magnetic resonance imaging data by using the trained bipolar convolution kernel to obtain a processing result; and converting a processing result to an image domain to obtain a reconstructed magnetic resonance image. According to the technical scheme of the embodiment of the invention, the design parameters of the bipolar convolution kernel are determined through the actual scanning parameters, the reference data are utilized to train the bipolar convolution kernel, and the obtained bipolar convolution kernel can correct ghost artifacts while the image acquisition efficiency is not affected.

Description

Technical Field

[0001] The present application relates to the technical field of magnetic resonance imaging, and in particular to a bipolar parallel imaging reconstruction method and device. Background Art

[0002] At present, the Nyquist ghost artifact (or N / 2 ghost artifact), hereinafter referred to as ghost artifact, in magnetic resonance imaging (MRI) is a common artifact in echo planar imaging (EPI) sequences. It usually appears as a series of repeated images at different positions in the phase encoding direction. This type of artifact seriously affects the image quality of diffusion MRI (dMRI) and functional MRI (fMRI) based on EPI.

[0003] After acquiring the signal, ghost artifacts can be corrected through a series of algorithms to improve the fidelity and usability of the image. The traditional correction method is linear phase correction (LPC), which extracts the linear phase difference between odd and even echoes by additionally collecting reference data in the positive and negative readout directions and uses it for correction. However, linear phase correction can usually only correct simpler phase errors, and has poor correction effects on complex, nonlinear or quadratic phase errors. At an ultra-high field of 7 Tesla, the image after linear phase correction still has obvious artifacts.

[0004] In order to solve this problem, dual polarity acquisition was proposed for correction. This method uses the positive and negative readout directions alternately for data acquisition. By performing complex averaging on the echoes of two different polarities, the artifacts caused by phase errors can be effectively offset. This method can avoid the influence of first-order and above phase differences, and the correction effect is significant. It can correct ghost artifacts and eliminate ripple artifacts caused by low-frequency phases. However, averaging different echoes will introduce a smoothing effect in the time dimension and reduce the effective time resolution, which is detrimental to the image quality of functional magnetic resonance imaging.

[0005] Based on this, some studies use dual-polarity GRAPPA to correct ghost artifacts. Dual-polarity GRAPPA (DPG) is a technology that can take into account both ghost artifact correction and acquisition efficiency. Bipolar GRAPPA has a good performance on traditional single-shot 2D EPI data, but cannot be directly applied to EPI acquired by multiple excitations. Summary of the invention

[0006] Based on this, the present application provides a bipolar parallel imaging reconstruction method and device to achieve magnetic resonance imaging correction without affecting image acquisition efficiency.

[0007] According to one aspect of the present application, a bipolar parallel imaging reconstruction method is proposed, comprising: acquiring magnetic resonance imaging data and reference data; determining bipolar convolution kernel design parameters according to actual scanning parameters of the acquired magnetic resonance imaging data, and constructing a bipolar convolution kernel according to the bipolar convolution kernel design parameters; training the bipolar convolution kernel based on the reference data; processing the magnetic resonance imaging data using the trained bipolar convolution kernel to obtain a processing result; and transforming the processing result to an image domain to obtain a reconstructed magnetic resonance image.

[0008] According to some embodiments, before the processing result is transformed into the image domain to obtain a reconstructed magnetic resonance image, it also includes: determining the monopolar convolution kernel design parameters according to the actual scanning parameters of the acquired magnetic resonance imaging data, and constructing the monopolar convolution kernel according to the monopolar convolution kernel design parameters; training the monopolar convolution kernel based on the phase difference-free reference data in the reference data; and using the trained monopolar convolution kernel to restore the downsampled data in the magnetic resonance imaging data.

[0009] According to some embodiments, bipolar convolution kernel design parameters are determined according to actual scanning parameters for acquiring magnetic resonance imaging data, and a bipolar convolution kernel is constructed according to the bipolar convolution kernel design parameters, including: determining the number of bipolar convolution kernels according to the number of excitations in the actual scanning parameters; determining the interval inside the bipolar convolution kernel according to the downsampling multiple in the actual scanning parameters; setting the first convolution kernel size of the first encoding dimension to a first preset value; setting the second convolution kernel size of the second encoding dimension to a second preset value; and constructing a bipolar convolution kernel according to the first convolution kernel size, the second convolution kernel size, the number of bipolar convolution kernels, and the interval inside the bipolar convolution kernel.

[0010] According to some embodiments, training a bipolar convolution kernel based on reference data includes: translating a target bipolar convolution kernel in k-space to extract data conforming to the shape of the target bipolar convolution kernel to obtain first target data, wherein the target bipolar convolution kernel is one of the bipolar convolution kernels, and the first target data includes target forward readout reference data and target reverse readout reference data; extracting target zero-phase difference reference data according to the position of the target forward readout reference data and / or the target reverse readout reference data; linearly combining the target forward readout reference data, the target reverse readout reference data, and the forward weight and the reverse weight, and constructing a first weight equation using the result of the linear combination and the target zero-phase difference reference data; solving the forward weight and the reverse weight according to the first weight equation to complete the training.

[0011] According to some embodiments, the monopolar convolution kernel design parameters are determined according to the actual scanning parameters for acquiring magnetic resonance imaging data, and the monopolar convolution kernel is constructed according to the monopolar convolution kernel design parameters, including: determining the third convolution kernel size of the first coding dimension according to the requirements for computational efficiency and signal-to-noise ratio in the actual scanning parameters; setting the fourth convolution kernel size of the second coding dimension to a third preset value; and constructing the monopolar convolution kernel according to the third convolution kernel size and the fourth convolution kernel size.

[0012] According to some embodiments, a monopolar convolution kernel is trained based on phase-difference-free reference data in the reference data, including: translating a target monopolar convolution kernel in k-space to extract data that conforms to the shape of the target monopolar convolution kernel to obtain second target data, wherein the target monopolar convolution kernel is one of the monopolar convolution kernels, and the second target data is phase-difference-free reference data; extracting an uncollected data point within the range of the target monopolar convolution kernel as third target data; linearly combining the second target data and the target weight, and constructing a second weight equation using the result of the linear combination and the third target data; solving the target weight according to the second weight equation to complete the training.

[0013] According to some embodiments, collecting reference data includes: collecting a set of forward encoding reference data and a set of reverse encoding reference data based on full sampling; reorganizing the forward encoding reference data and the reverse encoding reference data to obtain forward readout reference data and reverse readout reference data; merging the forward readout reference data and the reverse readout reference data to obtain a set of reference data without phase difference.

[0014] According to one aspect of the present application, a bipolar parallel imaging reconstruction device includes: a data acquisition module for acquiring magnetic resonance imaging data and reference data; a convolution kernel construction module for determining bipolar convolution kernel design parameters based on actual scanning parameters for acquiring magnetic resonance imaging data, and constructing a bipolar convolution kernel based on the bipolar convolution kernel design parameters; a convolution kernel training module for training the bipolar convolution kernel based on reference data; a data processing module for processing the magnetic resonance imaging data using the trained bipolar convolution kernel to obtain a processing result; and a reconstruction result module for transforming the processing result to the image domain to obtain a reconstructed magnetic resonance image.

[0015] According to one aspect of the present application, an electronic device is proposed, which includes: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0016] According to one aspect of the present application, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.

[0017] Through the above-mentioned embodiments provided by the present application, in the process of bipolar convolution kernel design, the actual scanning parameters of the magnetic resonance imaging data are used to determine the bipolar convolution kernel design parameters, and in the process of bipolar convolution kernel training, the bipolar convolution kernel is trained using reference data. The obtained bipolar convolution kernel can correct the phase difference between the positive and negative readout data of multiple excitation planar echo imaging, thereby correcting ghost artifacts without affecting the image acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without exceeding the scope of protection required by the present application.

[0020] Figure 1 A flow chart of a bipolar parallel imaging reconstruction method provided in an embodiment of the present application;

[0021] Figure 2 A flowchart of collecting reference data provided in an embodiment of the present application;

[0022] Figure 3 A flowchart of reconstructing a magnetic resonance image in the presence of downsampling provided in an embodiment of the present application;

[0023] Figure 4 A flowchart of determining bipolar convolution kernel design parameters according to actual scanning parameters for acquiring magnetic resonance imaging data, and constructing a bipolar convolution kernel according to the bipolar convolution kernel design parameters, provided in an embodiment of the present application;

[0024] Figure 5 A schematic diagram of a bipolar convolution kernel design provided in an embodiment of the present application;

[0025] Figure 6 A flowchart of training a bipolar convolution kernel based on reference data provided in an embodiment of the present application.

[0026] Figure 7 A schematic diagram of fitting the weights of a bipolar convolution kernel through reference data provided in an embodiment of the present application;

[0027] Figure 8A flowchart of determining the design parameters of a monopolar convolution kernel according to actual scanning parameters for acquiring magnetic resonance imaging data, and constructing a monopolar convolution kernel according to the design parameters of the monopolar convolution kernel provided in an embodiment of the present application;

[0028] Fig. 9 A flowchart of training a unipolar convolution kernel based on phase-difference-free reference data in reference data provided in an embodiment of the present application;

[0029] Fig.10 A schematic diagram of the reorganization and synthesis of reference data provided in the embodiments of the present application;

[0030] Fig.11 A block diagram of an open set fine-grained recognition device provided in an embodiment of the present application;

[0031] Fig.12 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0033] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0034] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0035] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0036] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another component. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the concepts of the present application. As used herein, the term "and / or" includes any one of the associated listed items and all combinations of one or more.

[0037] Magnetic resonance imaging is the acquisition of encoded radio frequency signals and the acquisition of images after inverse Fourier transformation. The frequency space where the data is located before the inverse Fourier transform is called k-space, and the data is called k-space data. The k-space trajectory of echoplanar imaging (EPI) is a sawtooth or square wave. In this trajectory, each k-space line corresponds to an echo signal, and the odd and even echoes of the echo sequence are usually collected in opposite directions. During imaging reconstruction, the even echo must be time-reversed to align with the odd echo for inverse Fourier transform. When the forward and reverse echoes are not perfectly symmetrical, artifacts are introduced. For example, if the start time of the first echo is slightly delayed, this delay will be propagated to all subsequent echoes, resulting in a slight time difference between the peaks of the odd and even echoes. When the inverse Fourier transform is performed, this phase error will cause the signal intensity to shift in the phase encoding direction, resulting in an artifact that is N / 2 pixels different from the main image position (where N is the number of pixels in the field of view), which is the ghost artifact. Ghost artifacts are mainly caused by the following factors:

[0038] Inconsistent gradient waveforms: In EPI, frequency-encoded gradients of alternating polarity are used. Due to imperfections in the gradient hardware or eddy current effects, the sampling of positive and negative polarity gradients is not completely symmetrical, resulting in artifacts in the image.

[0039] Magnetic field inhomogeneity: Local magnetic field inhomogeneities exacerbate this artifact, especially at the interface between air and tissue (such as sinuses and ear canals). Magnetic field inhomogeneities increase with increasing field strength, so ghost artifacts are more obvious in ultra-high field strength (7 Tesla and above) MRI systems.

[0040] Differences in spatial sensitivity of multi-channel receive arrays: Differences in spatial sensitivity of different receive coils can also cause artifacts.

[0041] Bipolar GRAPPA combines ghost correction with the GeneRalized Autocalibrating Partially Parallel Acquisitions (GRAPPA) process to estimate the missing k-space data while correcting the first-order and above phase errors between the forward and reverse readout data in the EPI data, thereby effectively removing these artifacts. This method only requires an additional set of reference data to be collected, and does not affect the temporal resolution of imaging, but bipolar GRAPPA is not suitable for EPI data with multiple excitations.

[0042] Based on this, the present application proposes a bipolar parallel imaging reconstruction method and device, and develops a ghost artifact correction method for EPI data of multiple excitations without reducing image acquisition efficiency.

[0043] For specific implementation methods, please refer to the following embodiments.

[0044] Figure 1 Flow chart of the bipolar parallel imaging reconstruction method provided in the embodiment of the present application. Figure 1 As shown, the method includes steps S110 to S150.

[0045] In step S110 , magnetic resonance imaging data and reference data are acquired.

[0046] First of all, it should be explained that the multiple encoding dimensions of k-space are defined as kx (readout dimension or frequency encoding dimension), ky (phase encoding dimension) and kz (layer phase encoding dimension, which only exists in three-dimensional data).

[0047] This application can actually be understood as dual polarity GRAPPA for multi-shot EPI, which is hereinafter referred to as msh-DPG for the convenience of explanation. msh-DPG is a method of reconstructing GRAPPA data of multiple shots of EPI while correcting the phase difference caused by the imperfection of the system, thereby correcting the ghost artifacts of the image.

[0048] In step S110 , similar to the simple bipolar GRAPPA, in order to restore the downsampled data, it is necessary to first collect low-resolution full-sample data as calibration or reference data (calibration / reference data), which are hereinafter referred to as reference data for the convenience of description.

[0049] The reference data can provide sensitivity information of the multi-channel coil, which is used to restore the downsampled signal in the imaging data or solve its aliasing. In the traditional technology, only one set of reference data needs to be collected, while the present application requires repeated collection of two sets of low-resolution full-sampled data as reference data.

[0050] According to the exemplary embodiment, the reference data includes at least two sets of data with opposite readout polarities. In the actual implementation process, it is necessary to repeatedly collect two sets of reference data and use opposite readout polarities for the two sets, that is, take opposite kx encoding directions, such as Figure 2 The two sets of data with opposite readout polarities are recorded as forward readout reference data and reverse readout reference data.

[0051] Further, based on the above embodiment, the reference data also includes reference data without phase difference. It should be noted that the reference data without phase difference is data with almost no phase difference between forward and reverse readout. So far, a total of three sets of reference data are obtained, one set for forward readout, one set for reverse readout, and one set for data without phase difference.

[0052] According to an example embodiment, the reference data without phase difference is obtained by reorganizing and synthesizing the forward readout reference data and the reverse readout reference data.

[0053] In step S110, magnetic resonance imaging data also needs to be collected. That is, in the present application, the collection of magnetic resonance signals is divided into two parts, the first part is reference data, and the second part is image data used for imaging, which is recorded as magnetic resonance imaging data.

[0054] It should be noted that, according to the exemplary embodiment, the reference data may be collected before the formal image scanning, that is, firstly collect the low-resolution full-sampled reference data, and then collect the magnetic resonance imaging data that may be downsampled.

[0055] Further, according to the example embodiment, the reference data is collected using multiple excitations, and the number of excitations is equal to the number of excitations of the imaging data multiplied by the downsampling multiple in the ky direction (denoted as Ry). This ensures that the effective echo intervals of the reference data and the imaging data are consistent and the degree of deformation is consistent. For the imaging data, only one kx encoding direction needs to be fixed and collected in a conventional multiple excitation EPI manner.

[0056] In step S120, bipolar convolution kernel design parameters are determined according to actual scanning parameters for acquiring magnetic resonance imaging data, and a bipolar convolution kernel is constructed according to the bipolar convolution kernel design parameters.

[0057] After obtaining the reference data and the magnetic resonance imaging data, bipolar GRAPPA reconstruction is used to replace the acquired data with the phase-corrected data. Specifically, first, in step S120, a bipolar convolution kernel is designed according to the data acquisition method.

[0058] In step S130, the bipolar convolution kernel is trained based on the reference data.

[0059] Afterwards, the weights of the convolution kernel are obtained by fitting using the three sets of reference data prepared in step S110, thereby completing the training of the bipolar convolution kernel.

[0060] In step S140, the magnetic resonance imaging data is processed using the trained bipolar convolution kernel to obtain a processing result.

[0061] Finally, in step S140, all sampled points in the magnetic resonance imaging data are replaced with new data without phase difference using the trained bipolar convolution kernel to complete the correction of artifacts.

[0062] In step S150, the processing result is transformed into the image domain to obtain a reconstructed magnetic resonance image.

[0063] After the correction in step S140 is completed, the data is still in the k-space and can be transformed into the image domain through inverse Fourier transform. After necessary processing such as channel merging, the final magnetic resonance image can be obtained.

[0064] In the process of bipolar convolution kernel design, the present application uses the actual scanning parameters of magnetic resonance imaging data to determine the bipolar convolution kernel design parameters. In the process of bipolar convolution kernel training, the bipolar convolution kernel is trained using reference data. The obtained bipolar convolution kernel can correct the phase difference between the positive and negative readout data of multiple-excitation planar echo imaging, thereby correcting ghost artifacts without affecting the image acquisition efficiency. This application breaks through the limitation that the original DPG algorithm can only be used for single-excitation EPI data, and can be used for submillimeter high-resolution imaging. This solution does not require high computing resources and is easy to implement parallel computing and memory optimization.

[0065] According to some embodiments, before step S150, the method further includes:

[0066] In step S210, the monopolar convolution kernel design parameters are determined according to the actual scanning parameters for acquiring the magnetic resonance imaging data, and the monopolar convolution kernel is constructed according to the monopolar convolution kernel design parameters.

[0067] In the specific implementation process, the convolution kernel size is designed according to the actual scanning parameters of the magnetic resonance data.

[0068] According to an example embodiment, the convolution kernel size is designed according to the downsampling factor of the magnetic resonance data.

[0069] In step S220, the unipolar convolution kernel is trained based on the reference data without phase difference in the reference data.

[0070] The weight of the convolution kernel is obtained by fitting the reference data without phase difference prepared in step S110, thereby completing the training of the unipolar convolution kernel.

[0071] In step S230, the downsampled data in the magnetic resonance imaging data is restored using the trained unipolar convolution kernel.

[0072] After the unipolar convolution kernel training is completed, the trained convolution kernel is used to fill all unsampled data in the imaging data.

[0073] The phase difference between the forward and reverse readout data has been removed in step S110 to step S140 , so if step S210 to step S230 occur after step S140 , there is no need to distinguish the forward and reverse readout data, and the conventional GRAPPA can be used directly to reconstruct the downsampled data points.

[0074] It should be emphasized that if there is no downsampling of the data, there is no need to perform steps S210 to S230.

[0075] Further, after step S230, step S150 is performed, that is, the processing result of the unipolar convolution kernel is transformed into the image domain to obtain a reconstructed magnetic resonance image, such as Figure 3 shown.

[0076] According to some embodiments, reference Figure 4 In step S120, the bipolar convolution kernel design parameters are determined according to the actual scanning parameters of the acquired magnetic resonance imaging data, and the bipolar convolution kernel is constructed according to the bipolar convolution kernel design parameters, which can be specifically implemented through steps S410 to S450.

[0077] In step S410, the number of bipolar convolution kernels is determined according to the number of excitations in the actual scanning parameters.

[0078] In this application, the size of the convolution kernel in each dimension is defined as the number of sampled points included in the dimension, and unsampled or spaced points are not included in the size.

[0079] If the magnetic resonance imaging data is two-dimensional EPI data, the reconstruction only considers the two dimensions of kx coding and ky coding, and the convolution kernel is two-dimensional; if it is three-dimensional EPI data, the kz coding dimension needs to be considered on this basis, and the convolution kernel size of this dimension is at least 2, that is, it includes two collected kz coding data, and the convolution kernel is three-dimensional.

[0080] In both the kx coding and ky coding dimensions, the two-dimensional and three-dimensional convolution kernels comply with the design of steps S410 to S450.

[0081] Specifically, the number of excitations determines the total number of convolution kernels required.

[0082] In step S420, the interval inside the bipolar convolution kernel is determined according to the downsampling multiple in the actual scanning parameters.

[0083] It should be emphasized that the downsampling factor only determines the spacing inside the convolution kernel and does not affect the size of the convolution kernel.

[0084] In step S430, the first convolution kernel size of the first encoding dimension is set to a first preset value.

[0085] The kx encoding direction is recorded as the first encoding dimension. That is, the size of the bipolar convolution kernel in the kx direction can be set to a first preset value.

[0086] According to an example embodiment, the first preset value can be defined based on the signal center offset caused by the phase difference and the requirements for calculation efficiency and signal-to-noise ratio. Further, the first preset value can also be set based on experience.

[0087] In step S440, the second convolution kernel size of the second encoding dimension is set to a second preset value.

[0088] The ky encoding direction is recorded as the second encoding dimension. The size of the bipolar convolution kernel in the ky direction is usually fixed to a second preset value, which is recorded as the second convolution kernel size.

[0089] According to an example embodiment, the second preset value is 2. That is, in the ky coding direction, the convolution kernel includes a row (ie, one ky coded data) of data k read out in the forward direction. + And a row of reverse read data k closest to it - There may be unsampled rows or data collected under other excitations between these two rows of data, and these data are not included in the range of data taken by the convolution kernel.

[0090] In step S450, a bipolar convolution kernel is constructed according to the first convolution kernel size, the second convolution kernel size, the number of bipolar convolution kernels, and the interval inside the bipolar convolution kernel.

[0091] In order to explain step S410 to step S440 in more detail, Figure 5 As shown, the following specific embodiments are given. In this embodiment, the convolution kernel size used in the kx direction is 3 for simplicity.

[0092] If two excitations are used for data acquisition in the kx-ky plane (Nshots=2), and the downsampling factor in the phase encoding direction is 2 (Ry=2), then a total of two convolution kernels are required. The convolution kernel includes two rows of data in the phase encoding direction, with one row of unsampled data between them. The phase-corrected data output by one convolution kernel is used to replace the forward readout encoded data, and the phase-corrected data output by another convolution kernel is used to replace the reverse readout encoded data. If four excitations are used for data acquisition in the kx-ky plane (Nshots=4), and the downsampling factor in the phase encoding direction is 1 (Ry=1), then a total of four convolution kernels are required. The convolution kernel includes two rows of data in the phase encoding direction. Two of the convolution kernels have no interval between the two rows of sampled data, and the phase-corrected data output by one convolution kernel is used to replace the forward readout encoded data, and the phase-corrected data output by the other convolution kernel is used to replace the reverse readout encoded data. The other two types of convolution kernels have a row of intervals between two rows of sampled data. There are sampled data in this interval but it is not included in the range of the convolution kernel. This is mainly to correct the data in the multi-excitation data that is not adjacent to the reverse readout (kx) encoded data. The phase-corrected data output by one of the convolution kernels is used to replace the forward readout encoded data, and the phase-corrected data output by the other convolution kernel is used to replace the reverse readout encoded data. Similarly, if three excitations are used for data acquisition in the kx-ky plane (Nshots=3), and the downsampling factor in the phase encoding direction is 2 (Ry=2), then a total of three types of convolution kernels are required. Two of the convolution kernels include two rows of data in the phase encoding direction, separated by a row of unsampled data. The third type of convolution kernel includes two rows of data in the phase encoding direction, separated by two rows and a row of sampled data.

[0093] According to some embodiments, reference Figure 6 In step S130, the bipolar convolution kernel is trained based on the reference data, which can be specifically implemented through steps S610 to S640.

[0094] In step S610, the target bipolar convolution kernel is used to translate in k-space to extract data that conforms to the shape of the target bipolar convolution kernel to obtain first target data, wherein the target bipolar convolution kernel is one of the bipolar convolution kernels, and the first target data includes target forward readout reference data and target reverse readout reference data;

[0095] After designing the bipolar convolution kernel, it needs to be trained, that is, the weights (w + and w - ),like Figure 7 This process uses the three sets of reference data in step S110 for fitting.

[0096] In actual operation, the same convolution kernel is used to translate in k-space, and all data that conform to the shape of the convolution kernel are extracted to obtain the first target data.

[0097] In step S620, target non-phase difference reference data is extracted according to the position of the target forward readout reference data and / or the target reverse readout reference data.

[0098] It should be noted that the forward problem in the bipolar convolution kernel is to multiply the forward readout data and the reverse readout data by a set of weights (w + and w - ) and then add them together, that is, perform linear combination, and then we can get a data point k without phase difference phase_corr The position of this data point is consistent with the original forward read data k + overlaps with one of the data points in .

[0099]

[0100] That is, the target non-phase difference reference data can be extracted according to the position of the target forward read reference data. The same is true for the target reverse read reference data, which will not be elaborated in this application.

[0101] In step S630, a linear combination is performed using the target forward readout reference data, the target reverse readout reference data, the forward weight and the reverse weight, and a first weight equation is constructed using the result of the linear combination and the target zero-phase-difference reference data.

[0102] The data of each position extracted in step S610-step S620 can constitute an equation in equation group (1).

[0103] In the same way, by changing the readout polarity, other convolution kernels with opposite polarities can be obtained to correct the phase difference of the reverse readout data. That is, the convolution kernel includes a row of reverse readout data k - and a row of forward read data k closest to it + The two can be linearly combined to obtain a phase difference-free data point k for replacing the reverse readout data k- phase_corr .

[0104] In step S640, the forward weight and the reverse weight are solved according to the first weight equation to complete the training.

[0105] For the trained bipolar convolution kernel, the convolution kernel is used to translate in the imaging data, and the data within the convolution kernel is weighted linearly combined, and the output k is used. phase_corr Replace the original sampled data points of the imaging data. Use multiple convolution kernels to slide in the k-space, and the output k phase_corrAll sampled data points in the imaging data can be replaced.

[0106] It should be noted that the same data point is reconstructed using only one convolution kernel, in which the distance between the forward readout and the reverse readout data is the shortest. Based on the above requirements, multiple convolution kernels can be designed and used to move in k-space. phase_corr Replace all sampled data points in the imaging data.

[0107] According to some embodiments, reference Figure 8 In step S210, the monopolar convolution kernel design parameters are determined according to the actual scanning parameters of the acquired magnetic resonance imaging data, and the monopolar convolution kernel is constructed according to the monopolar convolution kernel design parameters, which can be specifically implemented through steps S810 to S830.

[0108] In step S810, the size of the third convolution kernel of the first encoding dimension is determined according to the requirements of the computational efficiency and the signal-to-noise ratio in the actual scanning parameters.

[0109] It should be emphasized that if the magnetic resonance imaging data is two-dimensional EPI data, the reconstruction only considers the two dimensions of kx coding and ky coding, and the convolution kernel is two-dimensional; if it is three-dimensional EPI data, the kz coding dimension needs to be considered on this basis, and the convolution kernel size of this dimension is at least 2, that is, it includes two acquired kz coding data, and the convolution kernel is three-dimensional. In the other two dimensions, the two-dimensional and three-dimensional convolution kernels comply with the design of steps S810-S830.

[0110] The kx encoding direction is recorded as the first encoding dimension. In other words, the size in the kx encoding direction can be defined according to the requirements of computational efficiency and signal-to-noise ratio, and is recorded as the third convolution kernel size.

[0111] In some embodiments, the third convolution kernel size may also be set empirically.

[0112] In step S820, the fourth convolution kernel size of the second encoding dimension is set to a third preset value.

[0113] The ky coding direction is recorded as the second coding dimension. That is, the size of the unipolar convolution kernel in the ky direction is usually fixed to the third preset value, which is recorded as the fourth convolution kernel size.

[0114] According to an example embodiment, the third preset value is 2. In the ky encoding direction, the convolution kernel includes two or more rows of acquired readout data.

[0115] In step S830, a unipolar convolution kernel is constructed according to the third convolution kernel size and the fourth convolution kernel size.

[0116] According to some embodiments, reference Fig. 9 In step S220, the unipolar convolution kernel is trained based on the reference data without phase difference in the reference data, which can be specifically implemented through steps S910 to S940.

[0117] In step S910, the target unipolar convolution kernel is translated in k-space to extract data that conforms to the shape of the target unipolar convolution kernel to obtain second target data, wherein the target unipolar convolution kernel is one of the unipolar convolution kernels, and the second target data is reference data without phase difference.

[0118] The process of training the convolution kernel is to use the reference data without phase difference to fit the weight w of the convolution kernel. Specifically, the convolution kernel is translated in k-space to extract all data that conform to the shape of the convolution kernel to obtain the second target data.

[0119] In step S920, an uncollected data point within the target unipolar convolution kernel range is extracted as the third target data.

[0120] An uncollected data point within the convolution kernel range is extracted as the third target data.

[0121] It should be noted that the forward problem in the unipolar convolution kernel is to multiply the collected data k by a set of weights w and then add them, that is, perform linear combination, and then obtain an uncollected data point k within the convolution kernel range. target .

[0122] k target =wk (2)

[0123] In step S930, a linear combination is performed using the second target data and the target weight, and a second weight equation is constructed using the result of the linear combination and the third target data.

[0124] The data of each position extracted in step S910-step S920 can constitute an equation in equation group (2).

[0125] In step S940, the target weight is solved according to the second weight equation to complete the training.

[0126] Solving the set of equations can obtain the weight w of the convolution kernel, and the training is completed.

[0127] After completing the training of the convolution kernel, the imaging data can be reconstructed. Use the convolution kernel to translate in the imaging data and perform a weighted linear combination of the data collected in the convolution kernel. target All uncollected downsampled data can be supplemented.

[0128] According to some embodiments, in step S110, reference data is collected, which can be specifically implemented through steps S111 to S113.

[0129] In step 111, a set of forward coding reference data and a set of reverse coding reference data are collected based on full sampling. Fig.10 As shown, each coded reference data has a forward readout part and a reverse readout part. If the acquisition method of the forward coded reference data is "forward-reverse-forward", the acquisition method of the reverse coded reference data is "reverse-forward-reverse". In step 112, the forward coded reference data and the reverse coded reference data are reorganized to obtain forward readout reference data and reverse readout reference data. By reorganizing the forward coded reference data and the reverse coded reference data, a group of data with a kx coding direction that is completely forward (recorded as forward readout reference data) and a group of data with a completely kx reverse coding direction (recorded as reverse coding reference data) can be obtained, as shown in FIG. Fig.10 It should be emphasized that recombination means exchanging the positive and negative parts to form "positive-positive-positive" and "negative-negative-negative" data.

[0130] In step 113, the forward readout reference data and the reverse readout reference data are combined to obtain a set of reference data without phase difference.

[0131] After correcting the phase difference between the two sets of data, they are merged to synthesize a set of images with almost no phase difference between the forward and reverse readouts, such as Fig.10 shown.

[0132] The above data can be used as reference data after being transformed into k-space. There are many merging methods, including PLACE (phase labeling for additional coordinate encoding) technology (Xiang, Magn Reson Med, 2007) and GESTE (ghost elimination via spatial and temporal encoding) technology (Hoge WS, Magn Reson Med, 2010). So far, a total of three sets of reference data have been obtained, one set for forward readout, one set for reverse readout, and one set for data without phase difference.

[0133] The reconstruction algorithm used in this application is similar to the GRAPPA algorithm, which uses a convolution kernel to translate in k-space to replace or fill in data.

[0134] The following describes an apparatus embodiment of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, reference can be made to the method embodiment of the present application.

[0135] Fig.11 A block diagram of a bipolar parallel imaging reconstruction apparatus according to an exemplary embodiment is shown.

[0136] Fig.11 The device shown can execute the aforementioned bipolar parallel imaging reconstruction method according to the embodiment of the present application.

[0137] like Fig.11 As shown, the bipolar parallel imaging reconstruction device may include:

[0138] See also Fig.11 And referring to the previous description, the data acquisition module 1110 is used to collect magnetic resonance imaging data and reference data.

[0139] The convolution kernel construction module 1120 is used to determine the bipolar convolution kernel design parameters according to the actual scanning parameters of the acquired magnetic resonance imaging data, and to construct the bipolar convolution kernel according to the bipolar convolution kernel design parameters.

[0140] The convolution kernel training module 1130 is used to train the bipolar convolution kernel based on reference data.

[0141] The data processing module 1140 is used to process the magnetic resonance imaging data using the trained bipolar convolution kernel to obtain a processing result.

[0142] The reconstruction result module 1150 is used to transform the processing result into the image domain to obtain a reconstructed magnetic resonance image.

[0143] The device performs functions similar to the method provided above. For other functions, please refer to the previous description and will not be repeated here.

[0144] Fig.12 An electronic device according to an exemplary embodiment of the present application is shown. Fig.12 1200 according to this embodiment of the present application is described. Fig.12 The electronic device 1200 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0145] like Fig.12 As shown, the electronic device 1200 is in the form of a general computing device. The components of the electronic device 1200 may include, but are not limited to: at least one processing unit 1210, at least one storage unit 1220, a bus 1230 connecting different system components (including the storage unit 1220 and the processing unit 1210), a display unit 1240, etc.

[0146] The storage unit stores program codes, which can be executed by the processing unit 1210, so that the processing unit 1210 executes the methods described in this specification according to various exemplary embodiments of the present application. For example, the processing unit 1210 can execute the method described above.

[0147] The storage unit 1220 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 12201 and / or a cache storage unit 12202 , and may further include a read-only storage unit (ROM) 12203 .

[0148] The storage unit 1220 may also include a program / utility 12204 having a set (at least one) of program modules 12205, such program modules 12205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0149] Bus 1230 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0150] The electronic device 1200 may also communicate with one or more external devices 300 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1200, and / or communicate with any device that enables the electronic device 1200 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1250. In addition, the electronic device 1200 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 1260. The network adapter 1260 may communicate with other modules of the electronic device 1200 via the bus 1230. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1200, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0151] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. The technical solution according to the implementation method of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the above method according to the implementation method of the present application.

[0152] The software product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0153] Computer readable storage media may include data signals propagated in baseband or as part of a carrier wave, wherein readable program codes are carried. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program codes contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination thereof.

[0154] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0155] The computer-readable medium carries one or more programs. When the one or more programs are executed by a device, the computer-readable medium implements the aforementioned functions.

[0156] Those skilled in the art will appreciate that the above modules can be distributed in the device according to the description of the embodiment, or can be changed accordingly and only used in one or more devices different from the embodiment. The modules of the above embodiments can be combined into one module, or further divided into multiple sub-modules.

[0157] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiment of the present application.

[0158] The exemplary embodiments of the present application are specifically shown and described above. It should be understood that the present application is not limited to the detailed structures, configurations or implementations described herein; on the contrary, the present application is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended claims.

Claims

1. A bipolar parallel imaging reconstruction method, characterized in that: include: Acquiring magnetic resonance imaging data and reference data; Determining bipolar convolution kernel design parameters according to actual scanning parameters for acquiring the magnetic resonance imaging data, and constructing a bipolar convolution kernel according to the bipolar convolution kernel design parameters; Training the bipolar convolution kernel based on the reference data; Processing the magnetic resonance imaging data using the trained bipolar convolution kernel to obtain a processing result; The processing result is transformed into the image domain to obtain a reconstructed magnetic resonance image.

2. The method according to claim 1, characterized in that Before transforming the processing result into the image domain to obtain the reconstructed magnetic resonance image, the method further includes: Determining a monopolar convolution kernel design parameter according to actual scanning parameters for acquiring the magnetic resonance imaging data, and constructing a monopolar convolution kernel according to the monopolar convolution kernel design parameter; Training the unipolar convolution kernel based on the phase difference-free reference data in the reference data; The trained unipolar convolution kernel is used to restore the downsampled data in the magnetic resonance imaging data.

3. The method according to claim 1, characterized in that The step of determining bipolar convolution kernel design parameters according to actual scanning parameters for acquiring the magnetic resonance imaging data, and constructing a bipolar convolution kernel according to the bipolar convolution kernel design parameters, comprises: Determining the number of the bipolar convolution kernels according to the number of excitations in the actual scanning parameters; Determining the interval inside the bipolar convolution kernel according to the downsampling multiple in the actual scanning parameter; Setting a first convolution kernel size of a first encoding dimension to a first preset value; Setting a second convolution kernel size of a second encoding dimension to a second preset value; A bipolar convolution kernel is constructed according to the first convolution kernel size, the second convolution kernel size, the number of the bipolar convolution kernels, and the interval inside the bipolar convolution kernel.

4. The method according to claim 1, characterized in that: The training of the bipolar convolution kernel based on the reference data comprises: Utilizing a target bipolar convolution kernel to translate in k-space to extract data conforming to the shape of the target bipolar convolution kernel, to obtain first target data, wherein the target bipolar convolution kernel is one of the bipolar convolution kernels, and the first target data includes target forward readout reference data and target reverse readout reference data; Extracting target phase difference-free reference data according to the position of the target forward readout reference data and / or the target reverse readout reference data; Performing a linear combination using the target forward readout reference data, the target reverse readout reference data, the forward weight, and the reverse weight, and constructing a first weight equation using the result of the linear combination and the target phase difference-free reference data; The forward weight and the reverse weight are solved according to the first weight equation to complete the training.

5. The method according to claim 2, characterized in that: The step of determining the monopolar convolution kernel design parameters according to the actual scanning parameters for acquiring the magnetic resonance imaging data, and constructing the monopolar convolution kernel according to the monopolar convolution kernel design parameters, comprises: Determining a third convolution kernel size of the first encoding dimension according to requirements for computational efficiency and signal-to-noise ratio in the actual scanning parameters; Setting the fourth convolution kernel size of the second encoding dimension to a third preset value; A unipolar convolution kernel is constructed according to the third convolution kernel size and the fourth convolution kernel size.

6. The method according to claim 2, characterized in that The training of the unipolar convolution kernel based on the reference data without phase difference in the reference data comprises: Utilizing a target monopolar convolution kernel to translate in k-space to extract data conforming to the shape of the target monopolar convolution kernel, to obtain second target data, wherein the target monopolar convolution kernel is one of the monopolar convolution kernels, and the second target data is reference data without phase difference; Extracting an uncollected data point within the target unipolar convolution kernel range as the third target data; Performing a linear combination using the second target data and the target weight, and constructing a second weight equation using the result of the linear combination and the third target data; The target weight is solved according to the second weight equation to complete the training.

7. The method according to claim 1, characterized in that Collect reference data, including: A set of forward coding reference data and a set of reverse coding reference data are collected based on full sampling; Recombining the forward encoding reference data and the reverse encoding reference data to obtain forward readout reference data and reverse readout reference data; The forward readout reference data and the reverse readout reference data are combined to obtain a set of reference data without phase difference.

8. A bipolar parallel imaging reconstruction device, characterized in that: include: A data acquisition module, used for acquiring magnetic resonance imaging data and reference data; A convolution kernel construction module, used to determine bipolar convolution kernel design parameters according to actual scanning parameters for acquiring the magnetic resonance imaging data, and to construct a bipolar convolution kernel according to the bipolar convolution kernel design parameters; A convolution kernel training module, used for training the bipolar convolution kernel based on the reference data; A data processing module, used to process the magnetic resonance imaging data using the trained bipolar convolution kernel to obtain a processing result; The reconstruction result module is used to transform the processing result into the image domain to obtain a reconstructed magnetic resonance image.

9. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images