A compressed sensing magnetic resonance image reconstruction method, device, storage medium and electronic equipment

Abstract

The application discloses a compressed sensing magnetic resonance image reconstruction method and device, a storage medium and an electronic device. The method comprises the following steps: acquiring initial undersampling data; performing image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifacts; and performing compressed sensing image reconstruction processing based on at least the initial undersampling data, taking the first image as a constraint term, to obtain a target reconstructed image. The compressed sensing magnetic resonance image reconstruction method can avoid the roll-off artifacts generated in the conventional compressed sensing reconstruction method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical imaging, and in particular to a compressed sensing magnetic resonance image reconstruction method and device, a storage medium and an electronic device. BACKGROUND

[0002] Compressed sensing (CS) is a leading signal acquisition and processing innovation technology based on applied mathematics, which can greatly improve the magnetic resonance scanning speed and time and spatial resolution. Compressed sensing is a mathematical method for realizing image reconstruction by acquiring a small amount of signals at a rate lower than the Nyquist sampling theorem for signals with sparse characteristics.

[0003] However, in the existing Cartesian magnetic resonance acquisition process mode, the readout direction is full-signal sampling. After image reconstruction by the compressed sensing method, the readout direction signal does not meet the compressed sensing theory, resulting in a severe folding artifact in the reconstructed image. SUMMARY

[0004] Therefore, the present application provides a compressed sensing magnetic resonance image reconstruction method, device, storage medium and electronic device, which mainly aims to solve the problem of severe folding artifact in traditional compressed sensing reconstruction.

[0005] To solve the above problems, the present application provides a compressed sensing magnetic resonance image reconstruction method, comprising:

[0006] acquiring initial undersampling data;

[0007] performing image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifact;

[0008] performing compressed sensing image reconstruction processing based on at least the initial undersampling data to obtain a target reconstructed image, taking the first image as a constraint term.

[0009] Optionally, the image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifact comprises:

[0010] performing data processing on the initial undersampling data by using a GROWL reconstruction method to obtain target data;

[0011] performing image reconstruction processing on the target data to obtain the first image.

[0012] Optionally, the image reconstruction processing on the target data to obtain the first image comprises:

[0013] perform inverse Fourier transform on the first data corresponding to each acquisition channel in the target data to obtain a first sub-image corresponding to each acquisition channel;

[0014] determine a second sub-image corresponding to each acquisition channel based on the conjugate value of the sensitivity spectrum and each first sub-image;

[0015] combine each second sub-image to obtain the first image.

[0016] Optionally, the GROWL reconstruction method is used to process the initial undersampling data to obtain the target data, and the method specifically includes:

[0017] perform data extraction processing on the initial undersampling data to obtain target region sampling data;

[0018] perform first data processing on the target region sampling data to obtain a first target data matrix along the phase encoding direction;

[0019] perform second data processing on the target region sampling data to obtain a second target data matrix along the slice encoding direction;

[0020] perform data padding processing on the initial undersampling data based on the first target data matrix, the second target data matrix and the initial undersampling data to obtain first sampling data;

[0021] combine the first sampling data and the initial undersampling data to obtain the target data.

[0022] Optionally, the method further includes:

[0023] obtain an initial image based on any one of the first image, a null image and a prior image;

[0024] perform compressed sensing image reconstruction processing based on the initial undersampling data and the initial image to obtain a target reconstruction image, with the first image as a constraint term.

[0025] perform compressed sensing image reconstruction processing based on the initial undersampling data and the initial image to obtain a target reconstruction image, with the first image as a constraint term.

[0026] Optionally, the constraint term adopts L2 norm or L p norm.

[0027] To solve the above problems, the present application provides a compressed sensing magnetic resonance image reconstruction device, which includes:

[0028] an acquisition module configured to acquire initial undersampling data;

[0029] a first image reconstruction module configured to perform image reconstruction based on the initial undersampled data to obtain a first image without artifacts;

[0030] a target image reconstruction module configured to perform compressed sensing image reconstruction based on the initial undersampled data and the first image to obtain a target reconstructed image.

[0031] Optionally, the first image reconstruction module is specifically configured to:

[0032] perform data processing on the initial undersampled data by using a GROWL reconstruction method to obtain target data;

[0033] perform image reconstruction on the target data to obtain the first image.

[0034] To solve the above problems, the present application provides a storage medium, the storage medium stores a computer program, the computer program is executed by a processor to realize the steps of the compressed sensing magnetic resonance image reconstruction method of any one of the above.

[0035] To solve the above problems, the present application provides an electronic device, at least comprising a memory and a processor, the memory stores a computer program, and the processor realizes the steps of the compressed sensing magnetic resonance image reconstruction method of any one of the above when executing the computer program on the memory.

[0036] The present application obtains a first image without artifacts from initial undersampled data, and uses the first image without artifacts as a constraint term for compressed sensing reconstruction to obtain a target reconstructed image, which can remove the folding artifacts in the readout direction caused by not meeting the compressed sensing theory in the signal acquisition process of the existing compressed sensing method, and can promote more accurate diagnosis and treatment results.

[0037] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0038] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered as limiting the present application. Moreover, the same reference symbols are used to represent the same components throughout the drawings. In the drawings:

[0039] Figure 1A flow chart of a compressed sensing magnetic resonance image reconstruction method according to an embodiment of the application;

[0040] Figure 2 A flow chart of a compressed sensing magnetic resonance image reconstruction method according to another embodiment of the application;

[0041] Figure 3 A block diagram of a compressed sensing magnetic resonance image reconstruction apparatus according to another embodiment of the application. DETAILED DESCRIPTION

[0042] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0043] It is to be understood that various alterations, modifications and improvements can be made to the embodiments of the application herein disclosed. Accordingly, it is intended to embrace all such alterations, modifications and improvements as fall within the scope and spirit of the application.

[0044] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments of the present application and, together with the description given above and the claims, serve to explain the principles of the present application.

[0045] These and other characteristics of the present application will become apparent from the following description and appended claims, taken in conjunction with the accompanying drawings.

[0046] It should also be understood that, although the terms "first" and "second" can be used herein to describe various elements, these elements should not be limited by these terms. These terms are generally only used to distinguish one element from another.

[0047] The above and other aspects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0048] Specific embodiments of the present application are described hereinbelow, reference being made to the drawings; however, it should be understood that the embodiments described are merely examples of the present application, which can be embodied in various forms. Well-known and / or redundant functions and structures are not described in detail to avoid obscuring the present application unnecessarily. Therefore, the specific structural and functional details disclosed herein are not intended to limit the present application, but merely as a basis for the claims and a representative basis for teaching one of ordinary skill in the art to variously employ the present application in virtually any appropriate detailed structure.

[0049] The specification can use phrases such as "in one embodiment", "in another embodiment", "in yet another embodiment", or "in other embodiments", which can refer to one or more of the same or different embodiments of the application.

[0050] The embodiment of the present application provides a compressed sensing magnetic resonance image reconstruction method, which comprises the following steps as shown in the figure: Figure 1

[0051] Step S101: acquiring initial undersampling data;

[0052] In the specific implementation process of the present step, the initial undersampling data is first randomly sampled from the sample to be tested, and the initial undersampling data is data collected at a frequency lower than that required by the Nyquist sampling theorem. In an example, the signal in the central region of the K space can be fully sampled, the signal in the peripheral region outside the central region of the K space can be undersampled, and the initial undersampling data can be fully sampled in the frequency encoding direction.

[0053] Step S102: performing image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifacts;

[0054] In the specific implementation process of the present step, the k-space data can be first filled by using the GROWL algorithm, the first sampling data of the k-space vacancy is obtained after the data filling processing, the target data is obtained by combining the first sampling data of the k-space and the initial undersampling data, and the first image is obtained by performing image reconstruction processing on the target data. A preliminary result, that is, the first image, is reconstructed by using the GROWL algorithm. Although the image signal-to-noise ratio is low and the image details are missing due to the fact that the sampling mode violates the Nyquist sampling theorem, there is no roll-off artifact after the CS reconstruction. In the present example, the first image without artifacts mainly refers to the image without roll-off artifact. The reconstruction method of the first image is not limited to the GROWL algorithm, and as long as the method for obtaining the image without roll-off artifact by using the initial undersampling data can be applied to the embodiment of the present application, the specific method is not particularly limited.

[0055] Step S103: performing compressed sensing image reconstruction processing based on at least the initial undersampling data by taking the first image as a constraint term to obtain a target reconstruction image.

[0056] In the specific implementation process of the present step, the image reconstruction is realized by constructing a compressed sensing image solving model, and the compressed sensing image solving model of the present embodiment is shown in formula (1):

[0057]

[0058] In formula (1), u is an image variable, the optimal solution of u is the target reconstruction image in the present embodiment; f is the initial undersampling data; Ψ T represents a sparse domain transformation; argmin represents taking the minimum value; ‖‖ represents a norm, ‖‖1 represents an L1 norm; ‖‖2 represents an L2 norm; ‖‖​TV denotes total variation; μ, λ, α are regularization coefficients, which are generally empirical values, for example: μ can be 0.01, λ can be 0.01, and α can be 10000, μ, λ, α values in practical application are related to the scanning site and the scanning sequence, and the values can be adjusted according to actual needs. Wherein A p and f are defined as:

[0059]

[0060] S represents the sensitivity spectrum, F p denotes downsampling and Fourier transform. 1 to J represent 1 to J acquisition channels for collecting sample data; f J denotes the magnetic resonance compressed sensing sampling data corresponding to J channels; in formula (1), as the L2 norm of the first image, the constraint term is combined with the traditional compressed sensing reconstruction model to obtain the compressed sensing reconstruction model shown in formula (1), wherein in formula (1),

[0061]

[0062] wherein, is the partial derivative of u in the x direction, is the partial derivative of u in the y direction, and the partial derivative is also a finite difference. The target reconstruction image is obtained by inputting the initial undersampling data, the initial image and the first image into the compressed sensing image solving model formula (1), and when the loop iteration reaches the preset iteration stop condition, the target reconstruction image is obtained.

[0063] The present application obtains a first image without artifacts through initial undersampling data, and uses the first image without artifacts as a constraint term of compressed sensing reconstruction to perform reconstruction processing to obtain a target reconstruction image. For example, a norm constraint term based on the GROWL operator is added to the traditional compressed sensing solving model to obtain a preset compressed sensing model, and the target reconstruction image obtained by image reconstruction based on the preset compressed sensing model can be more accurate due to the constraint of the prior information of the first image without artifacts. The global ringing effect caused by the folding artifacts in the readout direction of the reconstructed image due to the fact that the acquired signal does not meet the compressed sensing theory can be removed, and the diagnosis and treatment result can be more accurate.

[0064] On the basis of the above-mentioned embodiments, in order to make the compressed sensing magnetic resonance image reconstruction method more accurate, the present application further provides a compressed sensing magnetic resonance image reconstruction method, comprising the following steps:

[0065] Step S201: acquiring initial undersampling data;

[0066] In the implementation, the initial undersampling data is collected at a frequency lower than that required by the Nyquist sampling theorem. In an example, the initial undersampling data in the center region of K-space can be fully sampled, and the initial undersampling data in the surrounding region of K-space can be undersampled.

[0067] Step S202: performing data extraction processing on the initial undersampling data to obtain target region sampling data.

[0068] In the implementation, a K-space center region of a full sampling region is determined, and the target region sampling data is sampling data corresponding to the K-space center region. The application analyzes the linear relationship of the target region sampling data along the phase encoding direction and along the slice encoding direction, and fills the K-space data based on the linear relationship. The initial undersampling data is distributed at different positions in the K-space, and each position corresponds to different K-space coordinate values. The sampling data corresponds to four dimensions in the K-space, i.e., x dimension, y dimension, z dimension, and sampling channel dimension. The coordinate values of the initial undersampling data in the K-space are denoted as K(k x ,k y ,k z ), and the initial undersampling data corresponds to a vector with a channel length in the K-space.

[0069] Step S203: performing first data processing on the target region sampling data to obtain a first target data matrix.

[0070] In the implementation, each of the sampling data in the target region is sequentially taken as reference sampling data. Then, based on each of the reference sampling data, a sampling data pair corresponding to each reference sampling data is obtained along the phase encoding direction. Finally, based on each of the sampling data pairs and the coordinate values of each of the sampling data in the K-space, a first target data matrix is calculated by using the least square method. For example, when there are 16*16 sampling data in the K-space center region, the 16*16 sampling data is taken as reference sampling data. Along the phase encoding direction, the sampling data corresponding to each reference sampling data is matched to obtain a sampling data pair corresponding to each reference sampling data. Specifically, the adjacent two data along the phase encoding direction can be matched. When the 16*16 sampling data is matched, 15*16 data pairs are obtained. Based on each of the sampling data pairs and the coordinate values of each of the sampling data in the K-space, the following formula (2) is used:

[0071]

[0072] The linear relationship between the reference sampled data in the central region of K space along the phase encoding direction is calculated using the least squares method, which is the first target data matrix. Here, H represents the Hermite transpose; -1 represents the inverse operation of the matrix; m is the displacement of one point along the phase encoding direction (y-direction); Δk y K(k) represents the span of a point in the phase encoding direction (y-direction) of K-space; x ,k y ,k z ) represents the vector of acquisition channel lengths; · represents matrix multiplication. For n c ×n c The convolution kernel matrix, where n c This represents the number of data acquisition channels. The matrix values ​​change with the value of m, resulting in different numerical matrices. For example, when m = 1, a linear relationship matrix between the reference sampling point and its upper adjacent sampling point along the phase encoding direction can be obtained; when m = -1, a linear relationship matrix between the reference sampling point and its lower adjacent sampling point along the phase encoding direction can be obtained. Based on the linear relationship between the reference sampling data in the central region of K-space along the phase encoding direction, i.e., the first target data matrix, the foundation is laid for subsequently recovering the missing sampling data along the phase encoding direction of the initial undersampled data.

[0073] Step S204: Perform a second data processing on the sampled data of the target area to obtain a second target data matrix;

[0074] In the specific implementation process of this step, firstly, each of the sampled data in the target area is taken as the reference sampled data in turn; secondly, based on each of the reference sampled data along the layer selection coding direction, the sampled data pairs corresponding to each reference sampled data are obtained; finally, based on each of the sampled data pairs and the coordinate values ​​of each sampled data in the K space, the second target data matrix is ​​calculated using the least squares method. For example: As mentioned above, there are 16*16 sampled data in the central area of ​​the K space. These 16*16 sampled data are taken as the reference sampled data. Along the layer selection coding direction, the sampled data corresponding to the reference sampled data is matched to obtain the sampled data pairs corresponding to each reference sampled data. Specifically, the two adjacent data can be matched along the layer selection coding direction. After the 16*16 sampled data are matched, 16*15 data pairs are obtained. Based on each of the sampled data pairs and the coordinate values ​​of each sampled data in the K space, the following formula (3) is used:

[0075]

[0076] The least square method is used to calculate the linear relationship of each reference sampling data in the center region of K-space along the layer encoding direction, i.e. the second target data matrix. Wherein H represents the Hermitian transpose; -1 represents the inverse operation of the matrix; n is the displacement of one point along the layer encoding direction (z direction); Δk z is the span of K-space layer encoding direction (z direction) one point; K(k x ,k y ,k z is the vector of the length of the acquisition channel; · is the matrix multiplication operation; is the convolution kernel matrix of n c × n c , wherein n c represents the number of acquisition channels. The matrix value changes with the value of n to obtain different value matrices, for example, when n = 1, the linear relationship matrix of the reference sampling point and the right adjacent sampling point along the layer encoding direction can be obtained, and when n = -1, the linear relationship matrix of the reference sampling point and the left adjacent sampling point along the phase encoding direction can be obtained. The linear relationship of the reference sampling data in the center region of K-space along the phase encoding direction, i.e. the second target data matrix, lays the foundation for subsequent recovery of the missing sampling data of the initial undersampling data along the layer encoding direction.

[0077] Step S205: performing data filling processing based on the first target data matrix, the second target data matrix and the initial undersampling data to obtain first sampling data;

[0078] In the specific implementation process of this step, first, the sampling data with the sampling mask being 1 in the initial undersampling data is obtained, and the following formula (4) is applied based on each sampling data:

[0079]

[0080] The initial undersampling data is filled with data, and specifically, the coordinate values of the sampling data with the sampling mask being 1 in the K-space are multiplied with the first target data matrix and the second target data matrix respectively, so that the missing data values along the layer encoding direction and the phase encoding direction on both sides of the sampling data can be obtained, for example, when n = 0, m = 1, the formula can be converted to K(k x ,k y + Δky, kz = Gy, m·Kkx, ky, kz, at this time, one sampling point adjacent to the current sampling point along the phase encoding direction can be obtained according to the current sampling point; and for example, when n = 0, m = -1, the formula can be converted to At this time, one sampling point adjacent to the current sampling point in the phase encoding direction can be obtained according to the current sampling point; for example, when n = 1 and m = 0, the formula can be converted into At this time, one sampling point adjacent to the current sampling point on the right side in the layer selection encoding direction can be obtained according to the current sampling point; for example, when n = -1 and m = 0, the formula can be converted into At this time, one sampling point adjacent to the current sampling point on the left side in the layer selection encoding direction can be obtained according to the current sampling point. Based on this method, the sampling data above, below, left and right of the sampling data with the sampling mask being 1 are filled with data, and the first sampling data is obtained.

[0081] Step S206: The target data is obtained by combining the first sampling data and the initial undersampling data.

[0082] In the implementation process, the first sampling data of k-space recovery and the initial undersampling data are combined to obtain relatively complete sampling data, that is, the target data, which lays a foundation for subsequent image reconstruction based on the target data.

[0083] Step S207: Image reconstruction processing is performed on the target data to obtain the first image. In the implementation process, the first image is obtained according to the following formula (5)

[0084]

[0085] In the formula, F * represents inverse Fourier transform, S H represents the conjugate transform of the sensitivity spectrum. First, the first data K i corresponding to each acquisition channel in the target data is subjected to inverse Fourier transform processing to obtain a first sub-image corresponding to each acquisition channel. At this time, the first sub-image obtained has a non-uniform gray scale distribution. Second, based on each first sub-image and the conjugate value of the sensitivity spectrum, a second sub-image corresponding to each acquisition channel is determined. Specifically, the first sub-image corresponding to each acquisition channel is multiplied by the conjugate value of the sensitivity spectrum corresponding to the channel to obtain a second image with a relatively uniform gray scale. Finally, the second sub-images are synthesized to obtain the first image Growl(f). Through the GROWL algorithm, a pre-CS result, that is, the first image, is reconstructed. Although the sampling mode violates the Nyquist sampling theorem, resulting in a low signal-to-noise ratio of the image reconstructed by the GROWL algorithm and a lack of image details, there is no artifact after CS reconstruction.

[0086] Step S208: An initial image is obtained.

[0087] The initial image can be an empty image, which is mathematically expressed as u0=0, or a non-zero prior image, which is mathematically expressed as Or directly using the reconstruction result of Growl as the first image as the initial image, which is mathematically expressed as u0=Growl(f).

[0088] Wherein, the initial image u0=0 is an empty image, which is a full black image; the prior image Can be an image obtained by processing by a magnetic resonance algorithm. For example, a magnetic resonance compressed sensing sampling is performed on a to-be-detected part of a subject to obtain data D1, the data D1 being initial undersampling data; a magnetic resonance algorithm is performed on the to-be-detected part of the subject to obtain an image I1, the image I1 being a prior image Generally, the prior image contains a fold artifact; a GROWL algorithm is used to perform image reconstruction processing based on the initial undersampling data to obtain an artifact-free image I2; the empty image, the image I1 and the image I2 can be used as initial images of a compressed sensing reconstruction model in the present application.

[0089] Step S209: performing compressed sensing image reconstruction processing based on the Lp norm or L2 norm of the first image as a constraint term, and at least based on the initial undersampling data and the initial image, to obtain a current reconstructed image;

[0090] In the specific implementation process, the compressed sensing image reconstruction model used for image reconstruction is as shown in the following formula (1):

[0091]

[0092] In formula (1), u is an image variable, and the optimal solution of u is a target reconstructed image in the present embodiment; f is initial undersampling data; Ψ T represents a sparse domain transformation; argmin represents taking a minimum value; ‖‖ represents a norm, ‖‖1 represents an L1 norm; ‖‖2 represents an L2 norm; ‖‖ TV represents a total variation; μ, λ and α are regularization coefficients, which generally adopt empirical values, for example, when a double echo scan of an abdomen is performed, μ can take a value of 0.01, λ can take a value of 0.01, and α can take a value of 10000, the values of μ, λ and α are related to a scanning part and a scanning sequence in actual application, and the values can be adjusted according to actual needs. Wherein A p and f are defined as:

[0093]

[0094] S represents a sensitivity spectrum, and Fp represents the reduction and Fourier transform. 1 to J represent 1 to J acquisition channels for acquiring sample data; f J represents the J-channel corresponding magnetic resonance compressed sensing sample data; in formula (1), is a constraint term of L2 norm of the first image, and the constraint term is combined with a traditional compressed sensing reconstruction model to obtain the compressed sensing reconstruction model shown in formula (1), wherein in formula (1),

[0095]

[0096] wherein, is a partial derivative of u in the x direction, is a partial derivative of u in the y direction, and the partial derivative is also a finite difference. Specifically, the initial undersampling data, the initial image and the first image are substituted into the compressed sensing image reconstruction model formula (1) to obtain a current reconstructed image.

[0097] The compressed sensing reconstruction model can also be a formula shown in formula (6) as follows:

[0098]

[0099] In the formula, ‖·‖ p represents an Lp norm, and α‖u-Growl(f)‖ p is a constraint term of Lp norm of the first image, and the constraint term is combined with a traditional compressed sensing reconstruction model to obtain the compressed sensing reconstruction model shown in formula (6).

[0100] The compressed sensing reconstruction model can also be a formula shown in formula (7) as follows:

[0101]

[0102] The compressed sensing reconstruction model can also be a formula shown in formula (8) as follows:

[0103]

[0104] The compressed sensing reconstruction model can also be a formula shown in formula (9) as follows:

[0105]

[0106] The compressed sensing reconstruction model can also be a formula shown in formula (10) as follows:

[0107]

[0108] The compressed sensing reconstruction model can also be a formula shown in formula (11) as follows:

[0109]

[0110] The compressed sensing reconstruction model can also be a formula as shown in the following formula (12):

[0111]

[0112] The compressed sensing reconstruction model can also be a formula as shown in the following formula (13):

[0113]

[0114] The compressed sensing reconstruction model can also be a formula as shown in the following formula (14):

[0115]

[0116] The compressed sensing reconstruction model can also be a formula as shown in the following formula (15):

[0117]

[0118] The compressed sensing reconstruction model can also be a formula as shown in the following formula (16):

[0119]

[0120] When the first image is obtained by using the GROWL algorithm, the SNR (Signal-Noise Ratio) of the result after the GROWL algorithm is reconstructed is low, but the tissue structure is relatively complete, and therefore the L2 norm has a good effect.

[0121] Step S210: determining whether a preset iteration stopping condition is met based on the current reconstructed image and / or the number of image reconstructions, and executing step S211 when the preset iteration stopping condition is met, or executing step S212 when the preset iteration stopping condition is not met.

[0122] The preset iteration stopping condition can be that the number of iterations reaches a maximum image reconstruction number threshold, and the maximum image reconstruction number threshold can be set according to actual needs.

[0123] The preset iteration stopping condition can also be that the norm of the difference between the value u i and the value u i-1 of the image variable obtained after the last reconstruction is less than or equal to a preset difference threshold ε. The value of ε can be set according to actual needs.

[0124] Step S211: obtaining the target reconstructed image based on the current reconstructed image.

[0125] The current reconstructed image obtained in the step is the optimal solution of the compressed sensing image reconstruction model formula (1), that is, the target reconstructed image.

[0126] Step S212: image update the initial image based on the current reconstructed image and return to step S208 to obtain an updated initial image as the initial image of the next image reconstruction process.

[0127] In the implementation process, the initial image is updated based on the current reconstructed image to obtain the initial image of the next image reconstruction process, and image reconstruction processing is performed based on the obtained initial image of the next image reconstruction process and the initial undersampling data to obtain the current reconstructed image.

[0128] The present application obtains a first image without artifacts through initial undersampling data, and performs reconstruction processing to obtain a target reconstructed image by taking the first image without artifacts as a constraint term of compressed sensing reconstruction. For example, a preset compressed sensing model can be obtained by adding a norm constraint term based on the GROWL operator in the traditional compressed sensing solving model. The target reconstructed image obtained by image reconstruction based on the preset compressed sensing model can remove the global ringing effect caused by the folding artifacts of the reconstructed image in the readout direction due to the fact that the signal acquisition process of the existing compressed sensing method does not meet the compressed sensing theory, thereby facilitating more accurate diagnosis and treatment results.

[0129] Another embodiment of the present application provides a compressed sensing magnetic resonance image reconstruction device, as shown in the accompanying drawings, comprising: Figure 3

[0130] The acquisition module 1 is configured to acquire initial undersampling data.

[0131] The first image reconstruction module 2 is configured to perform image reconstruction processing based at least on the initial undersampling data to obtain a first image without artifacts.

[0132] The target image reconstruction module 3 is configured to take the first image as a constraint term and perform compressed sensing image reconstruction processing based at least on the initial undersampling data and the first image to obtain a target reconstructed image.

[0133] In the implementation process, the first image reconstruction module is specifically configured to perform data processing on the initial undersampling data by using the GROWL reconstruction method to obtain target data, and perform image reconstruction processing on the target data to obtain the first image.

[0134] ​Specifically, the first image reconstruction module is further configured to: perform data processing on the initial undersampling data by using a GROWL reconstruction method to obtain target data; and perform image reconstruction processing on the target data to obtain the first image; wherein the data processing on the initial undersampling data by using the GROWL reconstruction method to obtain the target data specifically comprises: performing data extraction processing on the initial undersampling data to obtain target region sampling data; performing first data processing on the target region sampling data to obtain a first target data matrix along a phase encoding direction; performing second data processing on the target region sampling data to obtain a second target data matrix along a slice encoding direction; performing data padding processing on the initial undersampling data based on the first target data matrix, the second target data matrix, and the initial undersampling data to obtain first sampling data; and combining the first sampling data and the initial undersampling data to obtain the target data.

[0135] Specifically, the first image reconstruction module is further configured to: perform image reconstruction processing on the target data to obtain the first image, specifically comprising: performing inverse Fourier transform processing on first data corresponding to each acquisition channel in the target data to obtain first sub-images corresponding to each acquisition channel; determining second sub-images corresponding to each acquisition channel based on the first sub-images and conjugate values of sensitivity spectra; and synthesizing the second sub-images to obtain the first image.

[0136] Specifically, the first image reconstruction module is further configured to: the constraint term adopts an L2 norm or an L p norm.

[0137] The compressed magnetic resonance image reconstruction apparatus further comprises a second acquisition module configured to: acquire an initial image based on any one of the first image, a null image, and a prior image.

[0138] The target image reconstruction module is specifically configured to: perform compressed sensing image reconstruction processing on the initial undersampling data based on the first image as a constraint term to obtain a target reconstruction image, specifically comprising: performing compressed sensing image reconstruction processing on the initial undersampling data based on the first image as a constraint term and the initial image to obtain a target reconstruction image.

[0139] The preset compressed sensing model is obtained by adding a norm constraint term based on the GROWL operator in a traditional compressed sensing solving model, and the target reconstructed image obtained by image reconstruction based on the preset compressed sensing model can remove the global ringing effect caused by the folding artifacts of the reconstructed image in the readout direction due to the fact that the existing compressed sensing method does not meet the compressed sensing theory in the signal acquisition process, thereby promoting more accurate diagnosis and treatment results.

[0140] Another embodiment of the present application provides a storage medium storing a computer program, which, when executed by a processor, implements the following method steps:

[0141] Step one: obtaining initial undersampling data;

[0142] Step two: performing image reconstruction processing based at least on the initial undersampling data to obtain a first image without artifacts;

[0143] Step three: taking the first image as a constraint term, performing compressed sensing image reconstruction processing based at least on the initial undersampling data to obtain a target reconstructed image.

[0144] The present application obtains a first image without artifacts through initial undersampling data, and takes the first image without artifacts as a constraint term for compressed sensing reconstruction to obtain a target reconstructed image. For example, a preset compressed sensing model can be obtained by adding a norm constraint term based on the GROWL operator in a traditional compressed sensing solving model, and the target reconstructed image obtained by image reconstruction based on the preset compressed sensing model can remove the global ringing effect caused by the folding artifacts of the reconstructed image in the readout direction due to the fact that the existing compressed sensing method does not meet the compressed sensing theory in the signal acquisition process, thereby promoting more accurate diagnosis and treatment results.

[0145] Another embodiment of the present application provides an electronic device comprising at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the following method steps when executing the computer program stored in the memory:

[0146] Step one: obtaining initial undersampling data;

[0147] Step two: performing image reconstruction processing based at least on the initial undersampling data to obtain a first image without artifacts;

[0148] Step three: taking the first image as a constraint term, performing compressed sensing image reconstruction processing based at least on the initial undersampling data to obtain a target reconstructed image.

[0149] The present application obtains a first image without artifacts by initially undersampling data, and uses the first image without artifacts as a constraint term for compressed sensing reconstruction, and performs reconstruction processing to obtain a target reconstruction image. For example, a preset compressed sensing model can be obtained by adding a constraint term of a norm based on a GROWL operator in a traditional compressed sensing solving model. The target reconstruction image obtained by performing image reconstruction based on the preset compressed sensing model can remove the global ringing effect caused by the fact that the signal acquisition process of the existing compressed sensing method does not meet the compressed sensing theory, and the folding artifact of the reconstructed image in the readout direction, thereby enabling more accurate diagnosis and treatment results.

[0150] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements shall also be considered to fall within the protection scope of the present application.

Claims

1. A compressed sensing magnetic resonance image reconstruction method, characterized in that, The method comprises: acquiring initial undersampling data; performing image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifacts; The first image is taken as a constraint term, and a compressive sensing image reconstruction processing is performed based on at least the initial undersampling data, to obtain a target reconstructed image, wherein the constraint term adopts an L2 norm or an L p norm. the image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifacts comprises: performing data processing on the initial undersampling data by using a GROWL reconstruction method to obtain target data; performing image reconstruction processing on the target data to obtain the first image.

2. The method of claim 1, wherein, The image reconstruction processing on the target data to obtain the first image specifically comprises: performing inverse Fourier transform processing on first data corresponding to each acquisition channel in the target data to obtain first sub-images corresponding to each acquisition channel; determining second sub-images corresponding to each acquisition channel based on the conjugate values of the first sub-images and a sensitivity spectrum; combining the second sub-images to obtain the first image.

3. The method of claim 1, wherein, The data processing on the initial undersampling data by using the GROWL reconstruction method to obtain target data specifically comprises: performing data extraction processing on the initial undersampling data to obtain target region sampling data; performing first data processing on the target region sampling data to obtain a first target data matrix along a phase encoding direction; performing second data processing on the target region sampling data to obtain a second target data matrix along a slice selection encoding direction; performing data padding processing on the initial undersampling data based on the first target data matrix, the second target data matrix, and the initial undersampling data to obtain first sampling data; combining the first sampling data and the initial undersampling data to obtain the target data.

4. The method of claim 1, wherein, The method further comprises: acquiring an initial image based on any one of the first image, a null image, and a prior image; the compressed sensing image reconstruction processing based on at least the initial undersampling data with the first image as a constraint term to obtain a target reconstruction image specifically comprises: the compressed sensing image reconstruction processing based on the initial undersampling data and the initial image with the first image as a constraint term to obtain a target reconstruction image.

5. A compressed sensing magnetic resonance image reconstruction apparatus, characterized in that The method comprises: an acquisition module configured to acquire initial undersampling data; a first image reconstruction module configured to perform image reconstruction processing based on at least the initial undersampling data to obtain a first image without artifacts; a target image reconstruction module configured to perform a compressed sensing image reconstruction process based on at least the initial under-sampled data and the first image, with the first image as a constraint term, to obtain a target reconstructed image, wherein the constraint term adopts an L2 norm or an L p norm. the first image reconstruction module is specifically configured to: perform data processing on the initial undersampling data by using a GROWL reconstruction method to obtain target data; perform image reconstruction processing on the target data to obtain the first image.

6. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the compressed sensing magnetic resonance image reconstruction method in any one of claims 1-4.

7. An electronic device, comprising: At least comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the compressed sensing magnetic resonance image reconstruction method in any one of claims 1-4 when executing the computer program on the memory.

Images