阅读说明：本技术 磁共振成像方法、磁共振成像方法及装置 (Magnetic resonance imaging method, magnetic resonance imaging method and magnetic resonance imaging device ) 是由 梁栋 程静 王海峰 刘新 郑海荣 于 2019-04-24 设计创作，主要内容包括：本发明公开了一种磁共振成像方法、装置、系统及存储介质。其中方法包括：根据磁共振成像的原始模型和用于求解原始模型的迭代算法建立初始网络模型,其中,迭代算法中包括待定求解算子和待定参数；将样本的欠采样K空间数据输入至初始网络模型中,得到网络模型的输出磁共振图像,根据输出磁共振图像和样本的标准磁共振图像确定损失函数；根据损失函数调节初始网络模型中网络参数和待定参数,生成用于磁共振成像的网络模型,其中,初始网络模型中网络参数用于替代迭代算法中的待定求解算子；获取待处理的欠采样K空间数据,将欠采样K空间数据输入至用于磁共振成像的网络模型,生成磁共振图像,提高了磁共振图像的质量。(The invention discloses a magnetic resonance imaging method, a magnetic resonance imaging device, a magnetic resonance imaging system and a storage medium. The method comprises the following steps: establishing an initial network model according to an original model of magnetic resonance imaging and an iterative algorithm for solving the original model, wherein the iterative algorithm comprises an undetermined solving operator and undetermined parameters; inputting the undersampled K space data of the sample into an initial network model to obtain an output magnetic resonance image of the network model, and determining a loss function according to the output magnetic resonance image and a standard magnetic resonance image of the sample; adjusting network parameters and undetermined parameters in the initial network model according to the loss function to generate a network model for magnetic resonance imaging, wherein the network parameters in the initial network model are used for replacing the undetermined solution operator in the iterative algorithm; the method comprises the steps of obtaining under-sampled K space data to be processed, inputting the under-sampled K space data into a network model for magnetic resonance imaging, generating a magnetic resonance image, and improving the quality of the magnetic resonance image.)

1. A magnetic resonance imaging method, comprising:

establishing an initial network model according to an original model of magnetic resonance imaging and an iterative algorithm for solving the original model, wherein the iterative algorithm comprises an undetermined solving operator and undetermined parameters;

inputting undersampled K-space data of a sample into the initial network model to obtain an output magnetic resonance image of the network model, and determining a loss function according to the output magnetic resonance image and a standard magnetic resonance image generated by fully-sampled K-space data of the sample;

adjusting network parameters and undetermined parameters in the initial network model according to the loss function to generate a network model for magnetic resonance imaging, wherein the network parameters in the initial network model are used for replacing the undetermined solver in the iterative algorithm;

acquiring undersampled K space data to be processed, and inputting the undersampled K space data into the network model for magnetic resonance imaging to generate a magnetic resonance image.

2. The method of claim 1, wherein the iterations for solving the original model comprise a dual iteration, a base iteration, and an association of the dual iteration with the base iteration;

Correspondingly, an initial network model is established according to an original model of magnetic resonance imaging and iteration for solving the original model, and the method comprises the following steps:

establishing at least one first subnetwork model for performing the dual iteration;

establishing at least one second sub-network model for performing the base iteration;

determining an association module of the first sub-network model and the second sub-network model according to the association relation of the dual iteration and the basic iteration;

determining a connection relationship between the first sub-network model, the second sub-network model and the association module according to an iterative relationship between the dual iteration and the base iteration;

and connecting the at least one first sub-network model, the at least one second sub-network model and the association module according to the connection relation to generate the initial network model.

3. Method according to claim 2, characterized in that the initial network model comprises a first sub-network model and a second sub-network model of a preset hierarchy, the outputs of the first sub-network model being connected to the inputs of the second sub-network model in the same hierarchy and the first sub-network model in the next hierarchy, respectively, the outputs of the second sub-network model being connected to the inputs of the second sub-network model in the next hierarchy, and being connected to the inputs of the first sub-network model in the next hierarchy on the basis of the association module.

4. The method of claim 2, wherein the first and second sub-network models are both residual networks;

the first sub-network model comprises a first preprocessing layer, a convolutional layer and an activation layer, and is used for preprocessing different types of input parameters received by the first sub-network model according to preset rules and sending generated first multi-dimensional matrix data to the convolutional layer connected with the first preprocessing layer;

the second sub-network model further comprises a second preprocessing layer, a convolutional layer and an activation layer, and the second preprocessing layer is used for preprocessing different types of input parameters received by the second sub-network model according to preset rules and sending generated second multi-dimensional matrix data to the convolutional layer connected with the second preprocessing layer.

5. The method of claim 4, wherein each convolutional layer in the first and second sub-network models comprises a real component channel for convolving real data of input information of the convolutional layer and an imaginary component channel for convolving imaginary data of input information of the convolutional layer.

6. The method of claim 2, wherein inputting undersampled K-space data of samples into the initial network model, resulting in an output magnetic resonance image of the network model, comprises:

inputting the undersampled K-space data, initial dual parameters, and initial connection information for the sample to a first sub-network model of a first level of the initial network model;

inputting an initial image to a second sub-network model of a first level of the initial network model;

and determining a feature map output by a second sub-network model in an output hierarchy in the initial network model as a magnetic resonance image generated by the initial network model.

7. The method of claim 6, wherein determining a feature map output by a second sub-network model in an output hierarchy in the initial network model as the magnetic resonance image generated by the initial network model comprises:

combining the output data of the real part channel and the output data of the imaginary part channel which have corresponding relation with the second sub-network model in the output level in the initial network model to generate reconstructed magnetic resonance data;

and generating a magnetic resonance image according to the reconstructed magnetic resonance data.

8. The method of claim 1, wherein determining a loss function from the output magnetic resonance image and a standard magnetic resonance image generated from fully-sampled K-space data of the sample comprises:

the loss function loss is determined according to the following formula:

wherein, the

9. A magnetic resonance imaging apparatus, characterized by comprising:

the initial network model establishing module is used for establishing an initial network model according to an original model of magnetic resonance imaging and an iterative algorithm used for solving the original model, wherein the iterative algorithm comprises an undetermined solving operator and undetermined parameters;

the loss function determining module is used for inputting the undersampled K space data of the sample into the initial network model to obtain an output magnetic resonance image of the network model, and determining a loss function according to the output magnetic resonance image and a standard magnetic resonance image generated by the fully-sampled K space data of the sample;

the network model training module is used for adjusting network parameters and undetermined parameters in the initial network model according to the loss function and generating a network model for magnetic resonance imaging, wherein the network parameters in the initial network model are used for replacing the undetermined solution operator in the iterative algorithm;

And the magnetic resonance imaging module is used for acquiring the undersampled K space data to be processed, inputting the undersampled K space data into the network model for magnetic resonance imaging and generating a magnetic resonance image.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the magnetic resonance imaging method as set forth in any one of claims 1-8.

11. A magnetic resonance imaging system comprising a magnetic resonance apparatus and a computer apparatus, wherein the computer apparatus comprises a memory, one or more processors and a computer program stored on the memory and executable on the processors, characterized in that the processor when executing the program is operable to perform the magnetic resonance imaging method of any one of claims 1-8.

Technical Field

The embodiments of the present invention relate to deep learning technologies, and in particular, to a magnetic resonance imaging method, and a magnetic resonance imaging apparatus.

Background

Magnetic resonance images human tissue using static and radio frequency magnetic fields, which not only provides rich tissue contrast, but also is harmless to the human body, thus becoming a powerful tool for medical clinical diagnosis. However, the low imaging speed has been a big bottleneck limiting the rapid development.

In terms of fast imaging, the currently common techniques are parallel imaging and compressed sensing. Parallel imaging uses the correlation between multi-channel coils to accelerate acquisition, and compressed sensing uses the prior information of sparsity of an imaged object to reduce k-space sampling points. However, the parallel imaging acceleration times are limited due to the conditions such as hardware and the like, and the phenomenon of noise amplification of the image can occur along with the increase of the acceleration times; the compressed sensing technology has very long reconstruction time due to the adoption of iterative reconstruction, and is difficult to select sparse transformation and reconstruction parameters.

Disclosure of Invention

The invention provides a magnetic resonance imaging method, a magnetic resonance imaging method and a magnetic resonance imaging device, which are used for improving the quality of a magnetic resonance image.

In a first aspect, an embodiment of the present invention provides a magnetic resonance imaging method, including:

establishing an initial network model according to an original model of magnetic resonance imaging and an iterative algorithm for solving the original model, wherein the iterative algorithm comprises an undetermined solving operator and undetermined parameters;

Inputting undersampled K-space data of a sample into the initial network model to obtain an output magnetic resonance image of the network model, and determining a loss function according to the output magnetic resonance image and a standard magnetic resonance image generated by fully-sampled K-space data of the sample;

adjusting network parameters and undetermined parameters in the initial network model according to the loss function to generate a network model for magnetic resonance imaging, wherein the network parameters in the initial network model are used for replacing the undetermined solver in the iterative algorithm;

acquiring undersampled K space data to be processed, and inputting the undersampled K space data into the network model for magnetic resonance imaging to generate a magnetic resonance image.

In a second aspect, an embodiment of the present invention further provides a magnetic resonance imaging apparatus, including:

the initial network model establishing module is used for establishing an initial network model according to an original model of magnetic resonance imaging and an iterative algorithm used for solving the original model, wherein the iterative algorithm comprises an undetermined solving operator and undetermined parameters;

the loss function determining module is used for inputting the undersampled K space data of the sample into the initial network model to obtain an output magnetic resonance image of the network model, and determining a loss function according to the output magnetic resonance image and a standard magnetic resonance image generated by the fully-sampled K space data of the sample;

The network model training module is used for adjusting network parameters and undetermined parameters in the initial network model according to the loss function and generating a network model for magnetic resonance imaging, wherein the network parameters in the initial network model are used for replacing the undetermined solution operator in the iterative algorithm;

and the magnetic resonance imaging module is used for acquiring the undersampled K space data to be processed, inputting the undersampled K space data into the network model for magnetic resonance imaging and generating a magnetic resonance image.

In a third aspect, the embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements a magnetic resonance imaging method as provided in any of the embodiments of the present invention.

In a fourth aspect, embodiments of the present invention further provide a magnetic resonance imaging system, including a magnetic resonance apparatus and a computer apparatus, where the computer apparatus includes a memory, one or more processors, and a computer program stored in the memory and executable on the processors, and when the computer program is executed by the processors, the magnetic resonance imaging method as provided in any of the embodiments of the present invention is implemented.

According to the technical scheme provided by the embodiment of the invention, the initial network model is established according to the iterative algorithm based on the undetermined solving operator and the undetermined parameter, the initial network model is trained according to the sample, and the solving operator and the undetermined parameter in the iterative algorithm are learned by adjusting the network parameter in the initial network model, so that the degree of freedom of the network model is improved. Furthermore, the acquired undersampled K space data is reconstructed based on the trained network model, so that a high-quality magnetic resonance image is obtained, and the quality of the magnetic resonance image is improved.

Drawings

FIG. 1 is a schematic diagram of a magnetic resonance imaging system;

fig. 2 is a schematic flow chart of a magnetic resonance imaging method according to an embodiment of the present invention;

FIG. 3 is a diagram of providing an initial network model according to an embodiment of the invention;

FIG. 4 is a diagram illustrating a first sub-network model in an initial network model according to an embodiment of the present invention;

FIG. 5 is a diagram illustrating a second sub-network model and an association module in the initial network model according to an embodiment of the invention;

FIG. 6 is a schematic diagram of a comparison of magnetic resonance images generated by different algorithms according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of a magnetic resonance imaging apparatus according to a second embodiment of the present invention;

Fig. 8 is a schematic structural diagram of a magnetic resonance imaging system according to a second embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.

A magnetic resonance imaging system typically comprises a magnet having an aperture, a transmit coil for transmitting radio frequency signals and a receive coil for receiving magnetic resonance signals, gradient coils for spatially localizing the magnetic resonance signals, a pulse generator for generating a scan sequence, and a control system. The magnetic resonance imaging system is operated by an operator (clinician) controlling a console connected to the control system, which may include a keyboard or other input device, a control panel and a display to input commands and display the generated images.

Fig. 1 is a schematic structural diagram of a magnetic resonance imaging system in which a clinician first places a subject 3 on a bed 1 and places a local coil for receiving magnetic resonance signals on the body surface of the subject 3 when performing a magnetic resonance examination; then the clinician controls the scanning bed 1 to move towards the aperture formed by the magnet 2 by operating the console connected with the control system 5, after the magnetic resonance imaging system monitors that the clinician sends out the instruction of the movement of the scanning bed 1, the control system 5 monitors the movement range of the scanning bed 1 immediately, when the scanning bed 1 enters the edge of the scanning imaging area 4, the control system 5 controls the pulse sequence generator to generate a corresponding sequence for scanning, the sequence can control the excitation to generate the radio frequency pulse, and the radio frequency pulse can excite the body area of the examinee 3 to generate the precession nuclear spin. In the moving process of the scanning bed 1, the gradient magnetic field generated by the gradient coil can carry out phase encoding, frequency encoding or slice selection encoding on the precession nuclear spin, the receiving coil placed on the surface of the body of the detected object can move in the inner space of the magnet space along with the scanning bed 1, and the receiving coils at different positions are in an open state or a closed state under the action of the control system so as to receive corresponding magnetic resonance signals.