Magnetic resonance imaging method, system, terminal and storage medium based on deep learning

By undersampling and SPIRiT convolution kernel estimation of magnetic resonance imaging data, combined with convolutional neural network training, image reconstruction without estimating coil sensitivity is achieved, solving the problem of imaging speed and quality, significantly reducing reconstruction time and pseudo-sham Improve imaging effect.

CN113933773BActive Publication Date: 2024-11-15SHENZHEN INST OF ADVANCED TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111158049.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-30
Publication Date
2024-11-15
Estimated Expiration
2041-09-30

AI Technical Summary

Technical Problem

The existing magnetic resonance imaging technology is difficult to ensure imaging quality while improving the imaging speed, especially due to the aliasing artifacts and the long reconstruction time caused by inaccurate estimation of coil sensitivity information.

Method used

By undersampling the full-sampling k-space data of magnetic resonance, the SPIRiT convolution kernel is estimated to be filled, and the undersampling k-space data and SPIRiT convolution kernel are input to the convolution neural network for training, a k-space-based image reconstruction model is obtained, and a k-space-based image reconstruction model is realized. Image reconstruction avoids estimation of coil sensitivity information and reduces reconstruction time and memory usage.

Benefits of technology

This method can remove image artifacts caused by undersampling, significantly reduce reconstruction time, improve image reconstruction quality and robustness, and obtain better imaging effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113933773B_ABST
    Figure CN113933773B_ABST
Patent Text Reader

Abstract

The present application relates to a magnetic resonance imaging method, system, terminal and storage medium based on deep learning. The method includes: undersampling the full k-space data of magnetic resonance to generate undersampled k-space data; estimating the SPIRiT convolution kernel that fills the k-space based on the undersampled k-space data; inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained image reconstruction model based on k-space; and reconstructing magnetic resonance images using the trained image reconstruction model based on k-space. The embodiment of the present application adopts an image reconstruction model based on k-space for image reconstruction, and uses a convolutional neural network to learn the prior information of the image, without estimating the coil sensitivity information, which can greatly reduce the reconstruction time and obtain better image reconstruction effects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of magnetic resonance imaging technology, and in particular relates to a magnetic resonance imaging method, system, terminal and storage medium based on deep learning. Background Art

[0002] Magnetic resonance imaging uses static magnetic fields and radio frequency magnetic fields to image human tissues. It not only provides rich tissue contrast but is also harmless to the human body. It has become a powerful tool in clinical medicine. However, the slow imaging speed has always been a major bottleneck restricting the rapid development of magnetic resonance imaging. It is particularly important to increase the scanning speed and reduce the scanning time while ensuring the imaging quality.

[0003] In terms of fast imaging, traditional technologies are parallel imaging and compressed sensing. Parallel imaging uses the correlation between multi-channel coils to accelerate acquisition, but due to hardware and other conditions, the acceleration factor is limited, and as the acceleration factor increases, the image will show noise amplification. Compressed sensing uses the prior information of the sparsity of the imaged object to reduce the k-space sampling points. Due to the use of iterative reconstruction, the reconstruction time is very long, and it is difficult to select sparse transformation and reconstruction parameters.

[0004] Deep learning methods use neural networks to learn the optimal parameters required for reconstruction from a large amount of training data or directly learn the mapping relationship between undersampled data and fully sampled images, thereby achieving better imaging quality and higher acceleration than parallel imaging or compressed sensing methods. The current deep learning reconstruction method mainly uses the image-based SENSE reconstruction model, which requires the coil sensitivity information to be estimated in advance. However, the coil sensitivity information is often inaccurate in places where the phase changes sharply on the image, making it difficult to remove aliasing artifacts or residual phase singularities; in addition, for the case of image curling, the conventional coil sensitivity distribution map cannot fully characterize the coil distribution, and multiple coil sensitivity distribution maps need to be estimated, which greatly increases the reconstruction time and calculation and storage pressure. Summary of the invention

[0005] The present application provides a deep learning-based magnetic resonance imaging method, system, terminal and storage medium, aiming to solve at least one of the above-mentioned technical problems in the prior art to a certain extent.

[0006] In order to solve the above problems, this application provides the following technical solutions:

[0007] A magnetic resonance imaging method based on deep learning, comprising:

[0008] Under-sampling the full-sampling k-space data of magnetic resonance to generate under-sampling k-space data;

[0009] estimating a SPIRiT convolution kernel that fills the k-space according to the undersampled k-space data;

[0010] Inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model;

[0011] Magnetic resonance image reconstruction is performed using the trained k-space-based image reconstruction model.

[0012] The technical solution adopted by the embodiment of the present application also includes: the undersampling of the full-sampling k-space data of magnetic resonance also includes:

[0013] Acquire magnetic resonance full-sampling k-space data from a magnetic resonance scanner;

[0014] The magnetic resonance full-sampling k-space data is multiplied by a mask matrix to generate multi-channel under-sampling k-space data.

[0015] The technical solution adopted by the embodiment of the present application also includes: the SPIRiT convolution kernel for estimating the filling k-space according to the under-sampled k-space data includes;

[0016] Performing a SPIRiT operation on the undersampled K-space data to estimate a SPIRiT convolution kernel that fills the k-space;

[0017] The SPIRiT operation is to reconstruct unsampled data points in the undersampled K-space data using linear weighting factors calculated using calibration data, wherein the linear weighting factors are SPIRiT convolution kernels, and the calibration data are the fully sampled low-frequency parts of the center of the undersampled K-space data;

[0018] The SPIRiT convolution kernel calculation formula is:

[0019]

[0020] Where m represents the mth coil, x is the k-space data point, r represents the position, and R r is the operation of selecting all k-space points in a neighborhood around position r, g jm represents the weight vector obtained from all points in the neighborhood around position r, Yes jm The conjugate transpose of .

[0021] The technical solution adopted by the embodiment of the present application also includes: the image reconstruction model based on k-space is:

[0022]

[0023] Where x is the k-space to be reconstructed, D is the under-sampling mode, y represents the under-sampling k-space data, G represents the SPIRiT operation, λ1 and λ2 are penalty parameters, and R(x) is the penalty function related to the prior information.

[0024] The technical solution adopted by the embodiment of the present application also includes: the under-sampled k-space data and the SPIRiT convolution kernel are input into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model, which includes:

[0025] The optimization algorithm is used to iteratively solve the image reconstruction model based on k-space, and the solution process is:

[0026]

[0027] Where n represents the number of iterations, F represents Fourier transform; prOx R,τ represents the approximation operator, which is defined as:

[0028]

[0029] The technical solution adopted by the embodiment of the present application also includes: the under-sampled k-space data and the SPIRiT convolution kernel are input into a convolutional neural network for training, and the trained k-space-based image reconstruction model is obtained, which also includes:

[0030] Use convolutional neural network to fit the penalty function R(x) related approximation operator prOx R,τ , the reconstruction process is expressed as:

[0031]

[0032] Where Λ is a convolutional neural network.

[0033] The technical solution adopted by the embodiment of the present application also includes: the loss function of the convolutional neural network is:

[0034]

[0035] Where m rec is the image output by the convolutional neural network, m Ref is the gold standard image generated by the full k-space data.

[0036] Another technical solution adopted by the embodiment of the present application is: a magnetic resonance imaging system based on deep learning, comprising:

[0037] Undersampling module: used to undersample the full-sampling k-space data of magnetic resonance to generate undersampling k-space data;

[0038] SPIRiT processing module: used for estimating the SPIRiT convolution kernel for filling the k-space according to the undersampled k-space data;

[0039] Model training module: used for inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model;

[0040] Image reconstruction module: used for reconstructing magnetic resonance images using the trained k-space-based image reconstruction model.

[0041] Another technical solution adopted by the embodiment of the present application is: a terminal, the terminal includes a processor and a memory coupled to the processor, wherein:

[0042] The memory stores program instructions for implementing the deep learning-based magnetic resonance imaging method;

[0043] The processor is used to execute the program instructions stored in the memory to control deep learning-based magnetic resonance imaging.

[0044] Another technical solution adopted by the embodiment of the present application is: a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute the deep learning-based magnetic resonance imaging method.

[0045] Compared with the prior art, the beneficial effects produced by the embodiments of the present application are as follows: the magnetic resonance imaging method, system, terminal and storage medium based on deep learning of the embodiments of the present application estimate the convolution kernel for filling the under-sampled k-space from the calibration data, and use the k-space-based image reconstruction model to reconstruct the image, without estimating the coil sensitivity information, reducing the memory space occupied, and making the reconstruction model more robust; and using the convolutional neural network to learn the prior information of the image and assist in imaging, thereby removing the image artifacts caused by under-sampled, which can greatly reduce the reconstruction time and obtain better image reconstruction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of a magnetic resonance imaging method based on deep learning according to an embodiment of the present application;

[0047] Figure 2 This is a schematic diagram of the SPIRiT operation of an embodiment of the present application;

[0048] Figure 3 A schematic diagram of the iterative training process of a convolutional neural network according to an embodiment of the present application;

[0049] Figure 4 A schematic diagram of the structure of a magnetic resonance imaging system based on deep learning according to an embodiment of the present application;

[0050] Figure 5 This is a schematic diagram of the terminal structure of an embodiment of the present application;

[0051] Figure 6 A schematic diagram of the structure of a storage medium according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] See also Figure 1 , is a flow chart of a magnetic resonance imaging method based on deep learning according to an embodiment of the present application. The magnetic resonance imaging method based on deep learning according to an embodiment of the present application comprises the following steps:

[0054] S1: Acquire full-sampling k-space data from the magnetic resonance scanner;

[0055] S2: under-sampling the full-sampling k-space data of the magnetic resonance to generate multi-channel under-sampling k-space data;

[0056] In this step, undersampling specifically involves multiplying the magnetic resonance full-sampled k-space data by a mask matrix to generate multi-channel under-sampled k-space data.

[0057] S3: Perform SPIRiT operation on the undersampled K-space data to estimate the SPIRiT convolution kernel that fills the k-space;

[0058] In this step, if Figure 2 As shown, it is a schematic diagram of the SPIRiT operation of an embodiment of the present application. The SPIRiT operation is to reconstruct the unsampled data points in the undersampled K-space data using the linear weighting factor calculated by the self-calibration data. The linear weighting factor is the SPIRiT convolution kernel, where the calibration data is the central low-frequency full-sampled part of the undersampled K-space data. Specifically, the present application obtains the weight coefficient by solving the following formula:

[0059]

[0060] Where m represents the mth coil, x is the k-space data point, r represents the position, and R r is an operation that selects all k-space points in a neighborhood around position r, g jm is the weight vector obtained from all points in the neighborhood around position r, Yes jm The conjugate transpose of .

[0061] S4: Input the undersampled k-space data and the estimated SPIRiT convolution kernel into the built convolutional neural network for iterative training to obtain a trained k-space-based image reconstruction model;

[0062] In this step, the convolutional neural network is a three-layer convolutional neural network with a convolution kernel size of 3 and a feature map of 32, 32, 2, where the 2 channel is used to represent the real and imaginary parts of the image.

[0063] This application adopts an image reconstruction model based on k-space, and the model is expressed as follows:

[0064]

[0065] Where x is the k-space to be reconstructed, D is the under-sampled mode, y represents the under-sampled k-space data, G represents the SPIRiT operation, that is, estimating the relationship between the under-sampled k-space data and the surrounding channel data from the corrected data (expressed by kernel), and then applying the kernel to the entire k-space, λ1 and λ2 are penalty parameters, R(x) is a penalty function related to prior information, and the penalty function includes but is not limited to Tikhonov regularization and image sparse regularization.

[0066] Since the image reconstruction model (2) based on k-space requires iterative solution, it usually requires a large number of iterations to converge, resulting in a long image reconstruction time. In addition, the selection of reconstruction parameters such as penalty function and penalty parameter is relatively difficult, and improper selection may lead to image reconstruction failure. Therefore, the present application expands the solution process of the image reconstruction model based on k-space to a trained convolutional neural network, and finally obtains an image reconstruction model based on k-space through network training.

[0067] First, a suitable optimization algorithm is used to solve the k-space-based image reconstruction model (2). In the embodiment of the present application, the projections onto convex sets (POCS) algorithm is used as an example. The minimization problem (2) can be solved by the following iterative process:

[0068]

[0069] Where n is the number of iterations, F is the Fourier transform; prox R,τ represents the approximation operator, which is defined as follows:

[0070]

[0071] The third step in the solution process (3) is to transform to the image domain because the prior information based on the image is easier to express. Although the prior information based on the image domain is easier to express, it is still difficult to select appropriate prior information. Therefore, the embodiment of the present application uses a convolutional neural network to learn the prior information from a large amount of training data. In combination with the solution process (3), a convolutional neural network is used to fit the approximation operator prOx related to the penalty function R(x) R,τ , that is, the reconstruction process can be expressed as the following iterative process:

[0072]

[0073] Where Λ is a convolutional neural network.

[0074] Specifically, Figure 3 As shown, it is a schematic diagram of the iterative training process of the convolutional neural network of the embodiment of the present application. Among them, the SPIRiT operation, data fidelity DC and convolutional neural network correspond to the three steps in the iterative process (5), IFT (Inverse Fourier Transform) and FT (Fourier Transform, IFT) represent inverse Fourier transform and Fourier transform, respectively, and RSS represents coil channel merging. The input of the convolutional neural network is the under-sampled k-space data and the SPIRiT convolution kernel. The image output by the network is compared with the gold standard image generated by the full-sampled k-space data. The loss function is calculated based on the difference between the image output by the network and the gold standard image generated by the full-sampled k-space data, that is:

[0075] J(θ)=min Θ ‖m reC -m ref ‖1 (6)

[0076] Where m rec is the image output by the network, m ref The gold standard image is generated by the full k-space data. The loss function is minimized through the optimization algorithm and the network parameters are continuously updated until the training converges to obtain the final k-space-based image reconstruction model.

[0077] S5: Reconstruct the magnetic resonance image using the trained k-space-based image reconstruction model.

[0078] Based on the above, the deep learning-based magnetic resonance imaging method of the embodiment of the present application estimates the convolution kernel that fills the under-sampled k-space from the calibration data, and uses the k-space-based image reconstruction model to reconstruct the image. There is no need to estimate the coil sensitivity information, which reduces the memory space occupied and makes the reconstruction model more robust. It also uses a convolutional neural network to learn the prior information of the image and assist in imaging, thereby removing image artifacts caused by under-sampling, which can greatly reduce the reconstruction time and obtain better image reconstruction effects.

[0079] See also Figure 4 , is a schematic diagram of the structure of a magnetic resonance imaging system based on deep learning according to an embodiment of the present application. The magnetic resonance imaging system 40 based on deep learning according to an embodiment of the present application includes:

[0080] The undersampling module 41 is used to undersample the full-sampling k-space data of the magnetic resonance to generate undersampling k-space data;

[0081] SPIRiT processing module 42: used for estimating a SPIRiT convolution kernel for filling k-space according to the undersampled k-space data;

[0082] Model training module 43: used for inputting undersampled k-space data and SPIRiT convolution kernel into convolutional neural network for training, so as to obtain a trained k-space-based image reconstruction model;

[0083] Image reconstruction module 44: used for reconstructing magnetic resonance images using a trained k-space-based image reconstruction model.

[0084] See also Figure 5 , is a schematic diagram of the terminal structure of an embodiment of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0085] The memory 52 stores program instructions for implementing the above-mentioned deep learning-based magnetic resonance imaging method.

[0086] The processor 51 is used to execute program instructions stored in the memory 52 to control deep learning-based magnetic resonance imaging.

[0087] The processor 51 may also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip having the ability to process signals. The processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0088] See also Figure 6 , which is a schematic diagram of the structure of the storage medium of the embodiment of the present application. The storage medium of the embodiment of the present application stores a program file 61 that can implement all the above methods, wherein the program file 61 can be stored in the above storage medium in the form of a software product, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0089] The above description of the disclosed embodiments enables professionals and technicians in the field to implement or use the present application. Various modifications to these embodiments will be apparent to professionals and technicians in the field, and the general principles defined in this application can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown in the present application, but will conform to the widest range consistent with the principles and novel features disclosed in the present application.

Claims

1. A magnetic resonance imaging method based on deep learning, characterized in that: include: Under-sampling the full-sampling k-space data of magnetic resonance to generate under-sampling k-space data; estimating a SPIRiT convolution kernel that fills the k-space based on the undersampled k-space data; Inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model; Reconstructing a magnetic resonance image using the trained k-space-based image reconstruction model; The step of inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model comprises: The optimization algorithm is used to iteratively solve the image reconstruction model based on k-space. The solution process is: Where x is k-space data, D is undersampling mode, y is undersampling k-space data, G is SPIRiT operation, n is the number of iterations, and F is Fourier transform; prox R,τ represents the approximation operator, which is defined as: Use convolutional neural network to fit the penalty function R(x) related approximation operator prox R,τ , the reconstruction process is expressed as: Where Λ is a convolutional neural network.

2. The deep learning-based magnetic resonance imaging method according to claim 1, characterized in that: The undersampling of the full-sampling k-space data of magnetic resonance includes: Acquire magnetic resonance full-sampling k-space data from a magnetic resonance scanner; The magnetic resonance full-sampling k-space data is multiplied by a mask matrix to generate multi-channel under-sampling k-space data.

3. The deep learning-based magnetic resonance imaging method according to claim 2, characterized in that: The estimating the SPIRiT convolution kernel for filling the k-space according to the undersampled k-space data comprises: Performing a SPIRiT operation on the undersampled k-space data to estimate a SPIRiT convolution kernel that fills the k-space; The SPIRiT operation is to reconstruct unsampled data points in the undersampled k-space data using linear weighting factors calculated using calibration data, wherein the linear weighting factors are the SPIRiT convolution kernels, and the calibration data are the fully sampled low-frequency parts of the center of the undersampled k-space data.

4. The deep learning-based magnetic resonance imaging method according to any one of claims 1 to 3, characterized in that: The image reconstruction model based on k-space is: Where x is k-space data, D is the under-sampling mode, y represents under-sampling k-space data, G represents the SPIRiT operation, λ1 and λ2 are penalty parameters, and R(x) is a penalty function related to prior information.

5. The deep learning-based magnetic resonance imaging method according to claim 4, characterized in that: The loss function of the convolutional neural network is: Where m rec is the image output by the convolutional neural network, m ref is the gold standard image generated by the full k-space data.

6. A deep learning-based magnetic resonance imaging system using the deep learning-based magnetic resonance imaging method according to claim 1, characterized in that: include: Undersampling module: used to undersample the full-sampling k-space data of magnetic resonance to generate undersampling k-space data; SPIRiT processing module: used for estimating the SPIRiT convolution kernel for filling the k-space according to the undersampled k-space data; Model training module: used for inputting the undersampled k-space data and the SPIRiT convolution kernel into a convolutional neural network for training to obtain a trained k-space-based image reconstruction model; Image reconstruction module: used for reconstructing magnetic resonance images using the trained k-space-based image reconstruction model.

7. A terminal, characterized in that: The terminal includes a processor and a memory coupled to the processor, wherein: The memory stores program instructions for implementing the deep learning-based magnetic resonance imaging method according to any one of claims 1 to 5; The processor is used to execute the program instructions stored in the memory to control deep learning-based magnetic resonance imaging.

8. A storage medium, characterized in that: Program instructions executable by a processor are stored, and the program instructions are used to execute the deep learning-based magnetic resonance imaging method according to any one of claims 1 to 5.