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

Abstract

The present invention belongs to the field of magnetic resonance imaging technology, and discloses a magnetic resonance image reconstruction method, device, equipment and storage medium, the method comprising: generating a spatial encoding gradient field based on a multi-channel array gradient coil; in the spatial encoding gradient field, exciting an imaging object with a radio frequency signal so that the imaging object generates a magnetic resonance signal; using the spatial encoding gradient field to spatially encode the magnetic resonance signal, and collecting the magnetic resonance signal generated by the imaging object after spatial encoding to obtain an under-sampled signal; generating a coding matrix according to the spatial encoding gradient field, performing imaging reconstruction according to the coding matrix and the under-sampled signal to obtain a reconstructed image; inputting the reconstructed image into a pre-trained neural network model to obtain a predicted image, and using the predicted image as a magnetic resonance image. The present invention can quickly reconstruct a more accurate result in the case of high under-sampling of data, and the method can also effectively suppress noise interference and improve the signal-to-noise ratio.

Description

Technical Field

[0001] The present invention belongs to the technical field of magnetic resonance imaging, and in particular relates to a magnetic resonance image reconstruction method, device, equipment and storage medium. Background Art

[0002] With the emergence and development of emerging magnetic resonance imaging technology, nonlinear gradient magnetic field imaging technology has been proposed. This technology is an emerging technology that uses nonlinear gradient magnetic fields to accelerate and improve magnetic resonance imaging. This technology can improve imaging encoding efficiency by generating linear and nonlinear mixed gradient magnetic fields.

[0003] O-space imaging technology is a type of nonlinear gradient coded imaging. This technology combined with parallel imaging technology can improve the imaging speed. The spatial coding gradient field of O-space imaging is composed of x, y and z 2 The gradient in three directions (consisting of three x, y, z 2 The gradient coils realize three gradient fields respectively), where z 2 The shape of the field is concentric circles. Each encoding changes the size of the x and y gradients, thereby achieving the offset of the center of the concentric circle field (z 2 is the equation of the center coordinates at the origin. Changing x and y is equivalent to moving the center of the circle. Therefore, the shape of the synthesized gradient field is still concentric circles, but the center coordinates of the circles have moved). The multi-channel receiving coil receives the encoded signal (S), calculates the encoding matrix E according to the encoding gradient, and obtains the final image M according to the formula S = E*M.

[0004] In the prior art, the Kaczmarz iterative algorithm is used to solve the image M according to S=E*M.

[0005] This iterative reconstruction method has some disadvantages:

[0006] 1. Due to complex iterative calculations, the reconstruction time is long;

[0007] 2. The reconstruction accuracy is not high enough and there are artifacts in the reconstruction results;

[0008] 3. When the signal is interfered by noise, the noise will be further amplified, resulting in a low signal-to-noise ratio;

[0009] 4. Some hyperparameters in the iteration process (such as relaxation coefficient, number of iteration steps, etc.) will also greatly affect the reconstruction results.

[0010] Secondly, the existing technology combines compressed sensing (CS) with O-space spatial coding (consisting of three x, y, z 2 The gradient coils realize three gradient fields respectively), and each encoding changes z 2The field size causes random disturbances in the center of the synthetic gradient field, thereby optimizing the incoherence of the O-space sparse basis. The multi-channel array receiving coil receives the signal, and then according to the formula The final image is obtained, where is the encoding matrix, s is the signal, λ is the regularization parameter, and Ψ is the sparse transformation. By changing the second-order gradient field, the signal is made more incoherent in the sparse domain, and sparse constraints are introduced during reconstruction. This method can partially remove artifacts in the reconstructed image.

[0011] However, this method has the following defects:

[0012] 1. To optimize incoherence, each encoding changes z 2 The field size causes random perturbations in the center of the synthetic gradient field, but the best perturbation strategy is unknown and requires multiple experiments to determine the perturbation method based on empirical values;

[0013] 2. At the same time, each encoding changes z 2 Field size also increases the complexity of the sequence;

[0014] 3. Reconstruction uses the Kaczmarz iterative algorithm, which has a slow reconstruction speed and is affected by hyperparameters, and also amplifies random noise. Summary of the invention

[0015] The purpose of the present invention is to provide a magnetic resonance image reconstruction method to solve the problems in the prior art that the reconstruction accuracy is not high enough and the reconstruction result has artifacts.

[0016] In order to achieve the above object, the present invention adopts the following technical solutions:

[0017] In a first aspect, the present invention provides a method for reconstructing a magnetic resonance image, the method comprising:

[0018] Generate spatial encoding gradient fields based on multi-channel array gradient coils;

[0019] In the spatial encoding gradient field, the imaging object is excited by the radio frequency signal so that the imaging object generates a magnetic resonance signal; the magnetic resonance signal is spatially encoded by the spatial encoding gradient field, and the magnetic resonance signal generated by the imaging object after the spatial encoding is collected to obtain a sampling signal;

[0020] Generate a coding matrix according to the spatial coding gradient field, perform imaging reconstruction according to the coding matrix and the under-sampling signal, and obtain a reconstructed image;

[0021] The reconstructed image is input into a pre-trained neural network model to obtain a predicted image, and the predicted image is used as the magnetic resonance image.

[0022] Preferably, the multi-channel array gradient coil comprises: 40-channel gradient coil units, and the 40-channel gradient coil units are longitudinally arranged in a circular shape.

[0023] Preferably, the spatial encoding gradient field is determined by the position of each gradient coil unit and the current value applied to each gradient coil unit, and the spatial encoding gradient field includes: the shape and magnetic field strength of the gradient field and the spatial position of the gradient field.

[0024] Preferably, the magnetic resonance signal is received by a plurality of receiving coils, and a coding matrix is ​​generated according to a spatial coding gradient field, including:

[0025] Obtain sensitivity maps of several receiving coils;

[0026] The encoding matrix is ​​determined based on the sensitivity maps of several receiving coils and the shape of the spatial encoding gradient field.

[0027] Preferably, the function expression of the encoding matrix is:

[0028]

[0029] In the formula, E j (j, t, r) is the encoding matrix of the magnetic resonance signal collected by the jth receiving coil, c j (r) is the sensitivity map of the jth receiving coil, k T (t) is the k-space wave vector at time t, ψ(r) is the shape of the spatial encoding gradient field, and the relationship between the shape of the spatial encoding gradient field is: k(t)ψ(r)=γ∫G(t)*ψ(r)dt=γB(t,r), wherein G(t) is the gradient of the spatial encoding gradient field, γ is the gyromagnetic ratio, B(t,r) is the spatial encoding gradient field at time t, e is the natural base, r is the spatial position, T is the matrix transpose, and i is an imaginary number.

[0030] Preferably, the method further comprises: constructing a pre-trained neural network model, comprising:

[0031] Perform simulation encoding on the original sample image according to the spatial encoding gradient field to obtain an under-sampled simulation signal;

[0032] Generate a simulation coding matrix according to the spatial coding gradient field, perform simulation imaging reconstruction according to the simulation coding matrix and the under-sampled simulation signal, and obtain a reconstructed simulation image;

[0033] The reconstructed simulation image is used as the network input of the neural network model, and the original sample image is used as the label of the neural network model to train the neural network model;

[0034] Calculating the loss value of the neural network model during the training process according to the loss function; wherein the loss function is constructed based on the reconstructed simulation image and the original sample image;

[0035] The parameters of the neural network model are updated based on the loss value until the neural network model converges to obtain a trained neural network model.

[0036] Preferably, the loss function includes: MES function, perceptual loss function and frequency loss function;

[0037] The functional expression of the loss function is:

[0038]

[0039] Where LOSS is the loss function, MES For MES function, LOSS Perceptual is the perceptual loss function, LOSS focal-frequency is the frequency loss function, λ is the first regularization coefficient, α is the second regularization coefficient, is the nth original sample image, is the nth reconstructed simulation image, N is the total number of original sample images, φ() is the loss network function, and F() is the Fourier transform function.

[0040] In a second aspect, the present invention provides a magnetic resonance image reconstruction device for implementing the above-mentioned magnetic resonance image reconstruction method, the device comprising:

[0041] A gradient field generation module, used for generating a spatial encoding gradient field based on a multi-channel array gradient coil;

[0042] A spatial encoding module is used to excite an imaging object using a radio frequency signal in a spatial encoding gradient field so that the imaging object generates a magnetic resonance signal; spatially encode the magnetic resonance signal using the spatial encoding gradient field, and collect the magnetic resonance signal generated by the imaging object after spatial encoding to obtain an under-sampled signal;

[0043] An image reconstruction module is used to generate a coding matrix according to a spatial coding gradient field, and to perform imaging reconstruction according to the coding matrix and the under-sampling signal to obtain a reconstructed image;

[0044] The image prediction module is used to input the reconstructed image into a pre-trained neural network model to obtain a predicted image, and use the predicted image as a magnetic resonance image.

[0045] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned magnetic resonance image reconstruction method when executing the computer program.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned magnetic resonance image reconstruction method when executed by a processor.

[0047] Beneficial effects:

[0048] 1. With the help of the powerful ability of neural networks, the present invention can quickly reconstruct more accurate results in the case of high data under-collection. At the same time, the method can also effectively suppress noise interference and improve the signal-to-noise ratio;

[0049] 2. The multi-channel array gradient coil of the present invention can not only generate O-space gradient coding fields, but also can be flexibly combined to obtain a variety of different spatial gradient coding fields. It has great scalability and flexibility and can meet the requirements of various non-linear gradient coding imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0051] Figure 1 is a flow chart of a magnetic resonance image reconstruction method provided by one embodiment of the present invention;

[0052] Figure 2 It is a structural schematic diagram of a gradient coil unit provided in one embodiment of the present invention;

[0053] Figure 3 It is a schematic diagram of a partial arrangement structure of a multi-channel array gradient coil provided in one embodiment of the present invention;

[0054] Figure 4 It is a schematic diagram of the overall layout structure of a multi-channel array gradient coil provided by an embodiment of the present invention;

[0055] Figure 5 is a block diagram of a magnetic resonance image reconstruction device provided by one embodiment of the present invention;

[0056] Figure 6 The comparison effect diagrams are respectively of the predicted image of the 4-fold accelerated sampling of the 8-channel receiving coil of the present invention and the reconstructed image of the existing Kaczmarz iterative algorithm and the real data;

[0057] Figure 7 It is a comparison effect diagram of the predicted image of the 3-fold accelerated sampling of the single-channel receiving coil of the present invention and the reconstructed image of the existing Kaczmarz iterative algorithm with the real data. DETAILED DESCRIPTION

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0059] Embodiment 1

[0060] Figure 1 is a flow chart of a magnetic resonance image reconstruction method provided by one embodiment of the present invention, such as Figure 1 As shown, this embodiment provides a magnetic resonance image reconstruction method, the method comprising:

[0061] Step S10: Generate a spatial encoding gradient field based on a multi-channel array gradient coil.

[0062] In this embodiment, the multi-channel array gradient coil has 40 channels of gradient coil units, and the 40 channels of gradient coil units are arranged longitudinally in a circular shape. The structural schematic diagram of the channel gradient coil unit is as follows: Figure 2 As shown, each gradient coil unit is connected to a power supply separately, and current can be provided to the coil of each gradient coil unit; the local layout structure schematic diagram and the overall layout structure schematic diagram of the multi-channel array gradient coil are respectively shown in Figure 3 and Figure 4 shown.

[0063] The spatial encoding gradient field is determined by the position of each gradient coil unit and the current value applied to each gradient coil unit. The spatial encoding gradient field includes: the shape and magnetic field strength of the gradient field and the spatial position of the gradient field.

[0064] Therefore, different currents are applied to the 40 gradient coil units respectively, and each gradient coil unit generates a gradient field of different size and shape in a specified space. The current size and direction applied to each gradient coil unit are adjusted so that the final spatial encoding gradient field can meet the gradient field required for O-space encoding. In addition, the 40-channel array gradient coil has high flexibility and scalability, and can be used not only for this O-space encoding, but also for other gradient field encodings, which is of great help for sequence development, especially non-linear gradient imaging.

[0065] Step S20: in the spatial encoding gradient field, using a radio frequency signal to excite the imaging object so that the imaging object generates a magnetic resonance signal; using the spatial encoding gradient field to spatially encode the magnetic resonance signal, and collecting the magnetic resonance signal generated by the imaging object after the spatial encoding to obtain an under-sampled signal.

[0066] In this embodiment, a magnetic resonance instrument is used to change the magnetic field strength and magnetic field direction of the spatial encoding gradient field, so that protons of imaging objects at different positions can be spatially encoded (for example, a human body is placed in a spatial encoding gradient field, and radio frequency pulses are emitted through a coil. These pulses can excite protons, causing the protons to transition from a low energy level to a high energy level. When the radio frequency pulses stop, the protons return from a high energy level to a low energy level. In this process, magnetic resonance signals are released. These magnetic resonance signals are collected by a receiving coil and converted into electrical signals. By processing the electrical signals, an image of the inside of the human body can be reconstructed). When the protons return from a high energy level to a low energy level, the phase and amplitude of the signals they emit will vary depending on the position.

[0067] In this embodiment, a plurality of receiving coils are used to collect the magnetic resonance signals, and the spatial position of the protons can be determined by analyzing the phases and amplitudes of the magnetic resonance signals.

[0068] In the acquisition process, before each acquisition of magnetic resonance signals, the spatial encoding gradient field needs to be changed, that is, the spatial encoding gradient field is used to perform spatial encoding on the magnetic resonance signals. The spatial encoding is performed once and the magnetic resonance signals are acquired once. If the number of spatial encodings is sufficient and the number of acquired magnetic resonance signals is large, full sampling is preferred. This embodiment reduces the number of encodings of spatial encoding to achieve undersampling of magnetic resonance signals. This embodiment can speed up imaging and improve imaging efficiency by undersampling magnetic resonance signals. If full sampling requires 128 spatial encodings, then 64 spatial encodings, at this time the number of magnetic resonance signals is reduced by one half, that is, 2 times undersampling, and so on. The higher the undersampling multiple, the fewer the number of magnetic resonance signals.

[0069] The receiving coil may be a single-channel receiving coil or a multi-channel receiving coil, and the type of the receiving coil may be flexibly selected according to actual needs.

[0070] In this embodiment, the function expression of the signal model of the magnetic resonance signal received by each receiving coil is:

[0071]

[0072] In formula (1), s j (t) is the magnetic resonance signal received by the jth receiving coil, V is the volume of the imaging area of ​​the imaging object, m(r) is the image to be obtained, c j(r) is the sensitivity map of the jth receiving coil, k T (t) is the k-space wave vector at time t, ψ(r) is the shape of the spatial encoding gradient field, e is the natural base, r is the spatial position, T is the matrix transpose, and i is an imaginary number.

[0073] The signal model of the magnetic resonance signal collected by each receiving coil is discretized to obtain the magnetic resonance signal received by each receiving coil, so that the received magnetic resonance signal after reducing the number of spatial encoding times is used as an under-sampling signal.

[0074] Step S30: generating a coding matrix according to the spatial coding gradient field, and performing imaging reconstruction according to the coding matrix and the under-sampling signal to obtain a reconstructed image;

[0075] Specifically, the encoding matrix is ​​generated according to the spatial encoding gradient field, including:

[0076] Step S301: obtaining sensitivity maps of several receiving coils;

[0077] Step S302: determining a coding matrix according to the sensitivity maps of a plurality of receiving coils and the shapes of the spatial coding gradient fields.

[0078] In this embodiment, the function expression of the encoding matrix is:

[0079]

[0080] In formula (2), E j (j, t, r) is the encoding matrix of the magnetic resonance signal collected by the jth receiving coil, c j (r) is the sensitivity map of the jth receiving coil (the sensitivity map is the spatial sensitivity distribution map of the receiving coil, which describes the sensitivity of the coil to different positions. These sensitivity maps reflect the change in the receiving efficiency of the coil in the imaging area), k T (t) is the k-space wave vector at time t (the k-space wave vector is the k-space frequency), ψ(r) is the shape of the spatial encoding gradient field, and the relationship between the shape of the spatial encoding gradient field is: k(t)ψ(r)=γ∫G(t)*ψ(r)dt=γB(t,r), wherein G(t) is the gradient of the spatial encoding gradient field, γ is the gyromagnetic ratio, B(t,r) is the spatial encoding gradient field at time t, e is the natural base, r is the spatial position, T is the matrix transpose, and i is an imaginary number.

[0081] After obtaining the encoding matrix of the magnetic resonance signal collected by the jth receiving coil and the under-sampling signal collected by the jth receiving coil, imaging reconstruction is performed according to the following formula:

[0082] s j =E j *M (3);

[0083] In formula (3), E j is E in formula (2) j (j,t,r),s j is s in formula (1) j (t), M is the reconstructed image.

[0084] Step S40: input the reconstructed image into a pre-trained neural network model to obtain a predicted image, and use the predicted image as the magnetic resonance image.

[0085] In this embodiment, the image reconstructed by step S30 will have artifacts and the reconstruction quality is poor. Therefore, the present invention can quickly reconstruct a more accurate result in the case of high data under-collection by leveraging the powerful capabilities of the neural network.

[0086] As a further optimization of this embodiment, the method further includes: constructing a pre-trained neural network model, including:

[0087] Step a10: Perform simulation encoding on the original sample image according to the spatial encoding gradient field to obtain an under-sampled simulation signal.

[0088] In this embodiment, the original sample image is an original high-resolution image. By simulating the O-space encoding process on a computer, the original sample image can be simulated and encoded to obtain an under-sampled simulation signal.

[0089] Step a20: Generate a simulated coding matrix according to the spatial coding gradient field, and perform simulated imaging reconstruction according to the simulated coding matrix and the under-sampled simulated signal to obtain a reconstructed simulated image.

[0090] In this embodiment, the formula (1) to the formula (3) are also used to reconstruct a simulated image, that is, to obtain a reconstructed simulated image. In this case, the reconstructed simulated image also has artifacts.

[0091] Step a30: Use the reconstructed simulation image as the network input of the neural network model, use the original sample image as the label of the neural network model, and train the neural network model. The neural network model of this embodiment adopts the U-net++ network.

[0092] Step a40: Calculate the loss value of the neural network model during the training process according to the loss function, wherein the loss function is constructed based on the reconstructed simulation image and the original sample image.

[0093] In this embodiment, the loss function includes: MES (mean square error) function, perceptual loss function and frequency loss function;

[0094] The functional expression of the loss function is:

[0095]

[0096] Where LOSS is the loss function, MES For MES function, LOSS Perceptual is the perceptual loss function, LOSS focal-frequency is the frequency loss function, λ is the first regular coefficient, α is the second regular coefficient, y n B is the nth original sample image, y n C is the nth reconstructed simulation image, N is the total number of original sample images, φ() is the loss network function, and F() is the Fourier transform function.

[0097] Step a50: Update the parameters of the neural network model based on the loss value until the neural network model converges to obtain a trained neural network model.

[0098] In this embodiment, since the simulated image reconstructed in step a20 also has artifacts, deep learning is introduced into the field of nonlinear gradient imaging. The original high-resolution image is used as the original sample image, and the mapping relationship between the simulated image with artifacts and the high-resolution image is learned through a neural network. When the image reconstructed in step S30 is used as the input of the trained neural network model, a magnetic resonance image that is closer to the real one can be predicted, thereby achieving high-quality image reconstruction.

[0099] With the help of the powerful ability of neural networks, the present invention can quickly reconstruct more accurate results in the case of high data under-sampling. This method can not only be used for multi-channel data reconstruction, but also explores the data reconstruction combining O-space with a single receiving channel to obtain better results. At the same time, this method can also effectively suppress noise interference and improve the signal-to-noise ratio; and the multi-channel array gradient coil can not only generate O-space gradient coding fields, but also can be flexibly combined to obtain a variety of different gradient coding fields, which has great scalability and flexibility and can meet the requirements of various non-linear gradient coding imaging.

[0100] Embodiment 2

[0101] Figure 5 is a block diagram of a magnetic resonance image reconstruction device provided by one embodiment of the present invention, such as Figure 5 As shown, this embodiment provides a magnetic resonance image reconstruction device, which is used to implement the magnetic resonance image reconstruction method of embodiment 1, and the device includes:

[0102] A gradient field generation module, used for generating a spatial encoding gradient field based on a multi-channel array gradient coil;

[0103] A spatial encoding module is used to excite an imaging object using a radio frequency signal in a spatial encoding gradient field so that the imaging object generates a magnetic resonance signal; spatially encode the magnetic resonance signal using the spatial encoding gradient field, and collect the magnetic resonance signal generated by the imaging object after spatial encoding to obtain an under-sampled signal;

[0104] An image reconstruction module is used to generate a coding matrix according to a spatial coding gradient field, and to perform imaging reconstruction according to the coding matrix and the under-sampling signal to obtain a reconstructed image;

[0105] The image prediction module is used to input the reconstructed image into a pre-trained neural network model to obtain a predicted image, and use the predicted image as a magnetic resonance image.

[0106] This embodiment further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the magnetic resonance image reconstruction method of the first embodiment when executing the computer program.

[0107] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the magnetic resonance image reconstruction method of the first embodiment is implemented.

[0108] With the help of the powerful ability of neural networks, the present invention can quickly reconstruct more accurate results when the data is undersampled. At the same time, the method can also effectively suppress noise interference and improve the signal-to-noise ratio; and the multi-channel array gradient coil can not only generate O-space gradient coding fields, but also can be flexibly combined to obtain a variety of different gradient coding fields. It has great scalability and flexibility and can meet the requirements of various non-linear gradient coding imaging.

[0109] Embodiment 3

[0110] Figure 6 The comparison effect diagram of the predicted image of the 8-channel receiving coil of the present invention with 4 times accelerated sampling and the reconstructed image of the existing Kaczmarz iterative algorithm with the real data is shown as follows: Figure 6 As shown, the image reconstructed by the network is the predicted image of the present invention, the image of the Kaczmarz iteration is the reconstructed image of the existing Kaczmarz iteration algorithm, and the image of the real data is the real image.

[0111] Using 8 receiving coils and 4 times faster sampling, we compared the results of Kaczmarz and network reconstruction, as shown in Figure 6 As shown, from the SSIM (structural similarity), NMSE (standard mean square error), PSNR (peak signal-to-noise ratio) indicators and the difference graph, it can be seen that the network reconstruction result of the present invention has better quality and is closer to the real data.

[0112] Figure 7 The comparison effect diagram of the predicted image of the 3-fold accelerated sampling of the single-channel receiving coil of the present invention and the reconstructed image of the existing Kaczmarz iterative algorithm with the real data is shown as follows: Figure 7 As shown, the image reconstructed by the network is the predicted image of the present invention, the image of the Kaczmarz iteration is the reconstructed image of the existing Kaczmarz iteration algorithm, and the image of the real data is the real image.

[0113] Using a single receiving coil, 3 times faster sampling was performed to compare the results of Kaczmarz and network reconstruction, as shown in Figure 7 As shown, it can be seen from the SSIM, NMSE, PSNR indicators and the difference graph that the network reconstruction result of the present invention has better quality and is closer to the real data.

[0114] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0115] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0116] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A magnetic resonance image reconstruction method, characterized in that: The method comprises: Generate spatial encoding gradient fields based on multi-channel array gradient coils; In a spatial encoding gradient field, an imaging object is excited by a radio frequency signal so that the imaging object generates a magnetic resonance signal; the magnetic resonance signal is spatially encoded by the spatial encoding gradient field, and the magnetic resonance signal generated by the imaging object after spatial encoding is collected to obtain an under-sampled signal; wherein the magnetic resonance signal is received by a plurality of receiving coils; Generate a coding matrix according to the spatial coding gradient field, and perform imaging reconstruction according to the coding matrix and the under-sampling signal to obtain a reconstructed image; wherein generating the coding matrix according to the spatial coding gradient field includes: obtaining sensitivity maps of a plurality of receiving coils; determining the coding matrix according to the sensitivity maps of the plurality of receiving coils and the shape of the spatial coding gradient field; The reconstructed image is input into a pre-trained neural network model to obtain a predicted image, and the predicted image is used as the magnetic resonance image.

2. The magnetic resonance image reconstruction method according to claim 1, characterized in that: The multi-channel array gradient coil comprises: 40-channel gradient coil units, and the 40-channel gradient coil units are longitudinally arranged in a circular shape.

3. The magnetic resonance image reconstruction method according to claim 2, characterized in that: The spatial encoding gradient field is determined by the position of each gradient coil unit and the current value applied to each gradient coil unit. The spatial encoding gradient field includes: the shape and magnetic field strength of the gradient field and the spatial position of the gradient field.

4. The magnetic resonance image reconstruction method according to claim 1, characterized in that: The function expression of the encoding matrix is: ； In the formula, For the j The encoding matrix of the magnetic resonance signals collected by the receiving coils, For the j The sensitivity map of the receiving coils, for t Moment k The spatial wave vector, is the shape of the spatial encoding gradient field, and the relationship of the shape of the spatial encoding gradient field is: ,in, G ( t ) is the gradient of the spatial encoding gradient field, γ is the gyromagnetic ratio, B ( t , r )for t The spatial encoding gradient field at time instant, e is the natural base, r is the spatial position, T is the matrix transpose, i Is an imaginary number.

5. The magnetic resonance image reconstruction method according to claim 1, characterized in that: The method further comprises: constructing a pre-trained neural network model, comprising: Perform simulation encoding on the original sample image according to the spatial encoding gradient field to obtain an under-sampled simulation signal; Generate a simulation coding matrix according to the spatial coding gradient field, perform simulation imaging reconstruction according to the simulation coding matrix and the under-sampled simulation signal, and obtain a reconstructed simulation image; The reconstructed simulation image is used as the network input of the neural network model, and the original sample image is used as the label of the neural network model to train the neural network model; Calculating the loss value of the neural network model during the training process according to the loss function; wherein the loss function is constructed based on the reconstructed simulation image and the original sample image; The parameters of the neural network model are updated based on the loss value until the neural network model converges to obtain a trained neural network model.

6. The magnetic resonance image reconstruction method according to claim 5, characterized in that: The loss functions include: MES function, perception loss function and frequency loss function; The functional expression of the loss function is: ； In the formula, is the loss function, is the MES function, is the perceptual loss function, is the frequency loss function, is the first regular coefficient, is the second regularization coefficient, For the n The original sample image, For the n The reconstructed simulation image, N is the total number of original sample images, is the loss network function, is the Fourier transform function.

7. A magnetic resonance image reconstruction device, used to implement the magnetic resonance image reconstruction method according to any one of claims 1 to 6, characterized in that: The device comprises: A gradient field generation module, used for generating a spatial encoding gradient field based on a multi-channel array gradient coil; A spatial encoding module is used to excite an imaging object using a radio frequency signal in a spatial encoding gradient field so that the imaging object generates a magnetic resonance signal; spatially encode the magnetic resonance signal using the spatial encoding gradient field, and collect the magnetic resonance signal generated by the imaging object after spatial encoding to obtain an under-sampled signal; An image reconstruction module is used to generate a coding matrix according to a spatial coding gradient field, and to perform imaging reconstruction according to the coding matrix and the under-sampling signal to obtain a reconstructed image; The image prediction module is used to input the reconstructed image into a pre-trained neural network model to obtain a predicted image, and use the predicted image as a magnetic resonance image.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the magnetic resonance image reconstruction method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the magnetic resonance image reconstruction method according to any one of claims 1 to 6 is implemented.

Images