Magnetic Resonance Imaging Processing Apparatus and Method

Abstract

An embodiment of the present invention provides a magnetic resonance image processing method based on a magnetic resonance imaging processing device, including the following steps: acquiring subsampled magnetic resonance signals; obtaining first k-space data from the subsampled magnetic resonance signals by using a first parallel imaging technique; obtaining a first magnetic resonance image from the first k-space data by using an inverse Fourier transform; preprocessing the first magnetic resonance image to generate first input image data; and obtaining a first output magnetic resonance image from the first input image data by using a first artificial neural network model.

Description

Technical Field

[0001] The present invention relates to a magnetic resonance imaging processing apparatus and method, and more particularly, to a magnetic resonance imaging processing apparatus and method for accelerating the acquisition of magnetic resonance images from magnetic resonance signals using an artificial neural network. Background Art

[0002] Generally, a medical imaging device is a device that acquires a patient's body information and provides an image. Medical imaging devices include an X-ray imaging device, an ultrasonic diagnostic device, a computed tomography (CT) device, a magnetic resonance imaging (MRI) device, and the like.

[0003] Herein, magnetic resonance imaging uses a magnetic field harmless to the human body and non-ionizing radiation to cause nuclear magnetic resonance in hydrogen nuclei in the body, thereby imaging the density and physical / chemical properties of the nuclei. The imaging conditions of a magnetic resonance imaging device are relatively free and contain various diagnostic information in soft tissues, and can also provide images with excellent contrast. Therefore, it occupies an important position in the field of diagnosis using medical images.

[0004] Specifically, a magnetic resonance imaging device is an imaging diagnostic device that supplies energy of a certain frequency in a state where a certain magnetic field is applied to the nuclei and converts the energy released by the nuclei into a signal to diagnose the inside of the human body. Since the protons constituting the nuclei have their own spin angular momentum and magnetic dipole moment, they will align along the magnetic field direction after the magnetic field is applied, and the nuclei precess around the magnetic field direction. By virtue of this precessional motion, human body images can be obtained through nuclear magnetic resonance phenomena.

[0005] On the other hand, the imaging time using a magnetic resonance imaging device ranges from more than 20 minutes to more than 1 hour depending on factors such as the imaging site and the type of MR image. That is, the magnetic resonance imaging device has the disadvantage that the imaging time is relatively longer than that of other medical imaging devices. This disadvantage causes a burden on the patient during imaging, and patients with claustrophobia can hardly undergo this method. Therefore, hitherto, people have still been working hard to develop technologies that can shorten the imaging time, and at the same time, the image quality also needs to be improved. Summary of the Invention

[0006] Technical Problem

[0007] The magnetic resonance imaging processing apparatus and method of the present invention provide a magnetic resonance imaging processing apparatus and method that can obtain high-quality reconstructed images even when the magnetic resonance imaging time is shortened, and can easily diagnose the diseased part of a patient.

[0008] Technical Solution

[0009] An embodiment of the present invention provides a magnetic resonance imaging processing method based on a magnetic resonance imaging processing apparatus. The magnetic resonance imaging processing method includes the following steps: obtaining a subsampled magnetic resonance signal; obtaining first k-space data from the subsampled magnetic resonance signal by using a first parallel imaging technique; obtaining a first magnetic resonance image from the first k-space data by using an inverse Fourier transform; preprocessing the first magnetic resonance image to generate first input image data; and obtaining a first output magnetic resonance image from the first input image data by using a first artificial neural network model.

[0010] This embodiment can provide the following magnetic resonance imaging processing method. That is, if there are multiple subsampled magnetic resonance signals, multiple first k-space data and multiple first magnetic resonance images are respectively obtained. The step of preprocessing the first magnetic resonance image to generate first input image data includes the following steps: combining multiple first magnetic resonance images to obtain a single second magnetic resonance image; and preprocessing the second magnetic resonance image to generate first input image data.

[0011] This embodiment can provide the following magnetic resonance imaging processing method. That is, the step of combining multiple first magnetic resonance images to obtain a single second magnetic resonance image is performed by using the following mathematical formula 1.

[0012] Mathematical formula 1:

[0013] c i =S i ·m,

[0014]

[0015]

[0016] where c i =the i-th first magnetic resonance image,

[0017] S i =the sensitivity matrix of the i-th coil,

[0018] m=the second magnetic resonance image based on complex-valued information,

[0019] m real =the second magnetic resonance image based on real-valued information.

[0020] This embodiment can provide the following magnetic resonance imaging processing method. That is, the step of combining multiple first magnetic resonance images to obtain a single second magnetic resonance image is performed by using the following mathematical formula 2.

[0021] Mathematical formula 2:

[0022]

[0023] where m = the second magnetic resonance image based on complex-valued information,

[0024] S i * = the conjugate transposed matrix of the sensitivity matrix of the i-th coil,

[0025] c i = the i-th first magnetic resonance image.

[0026] This embodiment can provide the following magnetic resonance image processing method. That is, the first magnetic resonance image includes a target image and an aliased image in the phase encoding direction. The steps of preprocessing the first magnetic resonance image to generate first input image data include the following steps: shifting the first magnetic resonance image to obtain a shifted first magnetic resonance image so that the aliased image matches the target image; and stacking and grouping the first magnetic resonance image and the shifted first magnetic resonance image along a first direction perpendicular to the phase encoding direction and the readout direction respectively to generate first input image data.

[0027] This embodiment can provide the following magnetic resonance image processing method. That is, the second magnetic resonance image includes a target image and an aliased image in the phase encoding direction. The steps of preprocessing the second magnetic resonance image to generate first input image data include the following steps: shifting the second magnetic resonance image to obtain a shifted second magnetic resonance image so that the aliased image matches the target image; and stacking and grouping the second magnetic resonance image and the shifted second magnetic resonance image along a first direction perpendicular to the phase encoding direction and the readout direction respectively to generate first input image data.

[0028] This embodiment can provide the following magnetic resonance image processing method. That is, the first artificial neural network model inputs a noise pattern image generated based on the first magnetic resonance image.

[0029] This embodiment can provide the following magnetic resonance image processing method. That is, the first artificial neural network model has an artificial neural network structure including a contracting path and an expanding path. The contracting path includes a plurality of convolution layers and pooling layers, and the expanding path includes a plurality of convolution layers and unpooling layers. The noise pattern image is input into at least one of the plurality of convolution layers and unpooling layers of the expanding path.

[0030] An extended embodiment of the present invention can provide the following magnetic resonance imaging processing method, that is, it includes the step of obtaining second k-space data from a first output magnetic resonance image by using Fourier operation. The step of obtaining second k-space data includes the following steps: decomposing the first output magnetic resonance image to obtain a plurality of second output magnetic resonance images; and obtaining second k-space data from the second output magnetic resonance image by using Fourier operation.

[0031] This embodiment can provide the following magnetic resonance imaging processing method, that is, the step of decomposing the first output magnetic resonance image to obtain a plurality of second output magnetic resonance images is performed by using the following mathematical formula 3 or mathematical formula 4.

[0032] Mathematical formula 3:

[0033]

[0034] Wherein,

[0035] S i = the sensitivity matrix of the i-th coil,

[0036]

[0037] Mathematical formula 4:

[0038]

[0039] Wherein,

[0040] S i = the sensitivity matrix of the i-th coil,

[0041]

[0042]

[0043] This embodiment can provide the following magnetic resonance imaging processing method, that is, it includes the step of obtaining third k-space data from the second k-space data by using a second parallel imaging technique..

[0044] This embodiment can provide the following magnetic resonance imaging processing method, that is, it includes the following steps: obtaining a third magnetic resonance image from the third k-space data by using inverse Fourier operation; preprocessing the third magnetic resonance image to generate second input image data; and obtaining a third output magnetic resonance image from the second input image data by using a second artificial neural network model.

[0045] A magnetic resonance imaging processing apparatus for performing a magnetic resonance imaging processing method according to an embodiment of the present invention provides a magnetic resonance imaging processing apparatus, which includes: a memory storing a magnetic resonance imaging processing program; and a processor executing the program; as the program is executed, the processor acquires a subsampled magnetic resonance signal, obtains first k-space data from the subsampled magnetic resonance signal using a first parallel imaging technique, obtains a first magnetic resonance image from the first k-space data using an inverse Fourier transform, preprocesses the first magnetic resonance image to generate first input image data, and obtains a first output magnetic resonance image from the first input image data using a first artificial neural network model.

[0046] Effects of the Invention

[0047] The magnetic resonance imaging processing apparatus and method of the present invention provide a magnetic resonance imaging processing apparatus and method that can obtain high-quality reconstructed images even when the magnetic resonance imaging acquisition time is shortened, and can easily diagnose the diseased part of a patient.

[0048] Moreover, the magnetic resonance imaging processing apparatus and method according to an embodiment of the present invention acquire a plurality of magnetic resonance images and provide a high-accuracy output magnetic resonance image reconstructed from input data generated by preprocessing the plurality of magnetic resonance images using a first artificial neural network model and a second artificial neural network model, and can easily diagnose the diseased part of a patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a structural diagram showing the structure of a magnetic resonance imaging processing apparatus according to an embodiment of the present invention.

[0050] Figure 2 It is a flowchart showing the process of a magnetic resonance imaging processing method according to an embodiment of the present invention.

[0051] Figure 3 It is a schematic diagram showing a magnetic resonance imaging processing method according to an embodiment of the present invention.

[0052] Figure 4 It is a schematic diagram for explaining the difference between full sampling and subsampling according to an embodiment of the present invention.

[0053] Figure 5 It is a schematic diagram for explaining the difference between the parallel imaging techniques GRAPPA (Generalized Autocalibrating Partially Parallel Acquisition) and SPIRiT (Self-Calibrating Parallel Imaging Reconstruction) according to an embodiment of the present invention.

[0054] Figure 6 It is a flowchart showing the process of a first magnetic resonance image preprocessing method according to an embodiment of the present invention.

[0055] Figure 7It is a schematic diagram showing a first magnetic resonance imaging preprocessing method according to an embodiment of the present invention.

[0056] Figure 8 It is a flowchart showing the process of a method for preprocessing a first magnetic resonance image to generate first input image data according to a modified embodiment of the present invention.

[0057] Figure 9 It is a schematic diagram showing the steps of obtaining a first output magnetic resonance image from first input image data using a first artificial neural network model according to an embodiment of the present invention.

[0058] Figure 10 It is a schematic diagram showing the structure of a first artificial neural network model according to an embodiment of the present invention.

[0059] Figure 11 It is used to illustrate a noise pattern image that is applicable as learning data to an artificial neural network model in an embodiment of the present invention.

[0060] Figure 12 It shows the effect of applying a noise pattern image as learning data to an artificial neural network model to mitigate hallucination in an embodiment of the present invention.

[0061] Figure 13 It is a flowchart showing the process of a magnetic resonance imaging processing method including an extended embodiment of the magnetic resonance imaging processing method according to an embodiment of the present invention.

[0062] Figure 14 It is a schematic diagram showing the magnetic resonance imaging processing method according to an extended embodiment of the present invention.

[0063] Figure 15 It is a flowchart showing the steps of obtaining a second output magnetic resonance image and obtaining second k-space data according to an extended embodiment of the present invention.

[0064] Figure 16 It shows a reconstructed image with a 4-fold acceleration and a comparative example in an experimental example of the present invention.

[0065] Figure 17 It shows a reconstructed image with a 6-fold acceleration and a comparative example in another experimental example of the present invention. Detailed Description of the Invention

[0066] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art to which the present invention pertains can easily implement it. However, the present invention can be implemented in various different forms, and thus the present invention should not be limited to the embodiments described herein. Parts irrelevant to the description are omitted from the drawings for clarity, and similar elements are denoted by similar reference numerals throughout the specification.

[0067] This specification does not describe all elements of the embodiments, and general content in the technical field to which the present invention pertains or content repeated between embodiments will be omitted. The term "part" used in the specification may be implemented by software or hardware. In some embodiments, multiple "parts" are implemented by one element, or in some embodiments, one "part" includes multiple elements.

[0068] In this specification, an "image" may mean multi-dimensional data composed of discrete image elements (e.g., pixels in a two-dimensional image and voxels in a three-dimensional image). For example, an image may include medical images obtained by medical imaging devices such as magnetic resonance imaging (MRI) devices, computed tomography (CT) devices, ultrasonic imaging devices, or X-ray imaging devices.

[0069] In this specification, an "object" is the object to be imaged and may include a person, an animal, or a part thereof. For example, an object may include a part of the body (such as an organ) or a phantom. A phantom refers to a substance having a volume with a very similar biological density and effective atomic number and may include a spherical phantom having properties similar to those of the body.

[0070] A magnetic resonance imaging (MRI) system is a system that obtains an image of a tomographic region of an object by showing the intensity of a magnetic resonance (MR) signal of an RF (radio frequency) signal generated in a magnetic field with a specific intensity in a light and dark contrast manner.

[0071] Specifically, the MRI system forms a static magnetic field and aligns the direction of the magnetic dipole moment of specific atomic nuclei of an object located in the static magnetic field in the direction of the static magnetic field. The gradient magnetic field coil applies a gradient signal to the static magnetic field to form a gradient magnetic field, which can induce different resonance frequencies according to the parts of each object. The RF coil can irradiate a magnetic resonance signal at the resonance frequency of the part where the image is to be obtained. Moreover, the RF coil can receive magnetic resonance signals with different resonance frequencies radiated from each part of the object as the gradient magnetic field is formed. The MRI system applies an image reconstruction technique to the magnetic resonance signals received through this step to obtain an image. Moreover, the MRI system can perform serial or parallel signal processing on multiple magnetic resonance signals received through a multi-channel RF coil to reconstruct the multiple magnetic resonance signals into image data.

[0072] The magnetic resonance image processing apparatus 100 according to an embodiment of the present invention will be described below.

[0073] A magnetic resonance imaging processing apparatus 100 according to an embodiment of the present invention relates to a magnetic resonance imaging processing apparatus 100 that is directly / indirectly applicable to the aforementioned MRI system and accelerates the acquisition of magnetic resonance images from magnetic resonance signals.

[0074] The magnetic resonance imaging processing apparatus 100 may not only be an MRI system that can sense magnetic resonance signals by itself and acquire magnetic resonance images, but also an image processing apparatus that processes images taken from the outside, a smart phone, a tablet PC, a personal computer, a smart TV, a micro server, other household appliances, and other mobile devices or non-mobile computing devices having a processing function for magnetic resonance images. However, the present invention is not limited thereto. Moreover, the magnetic resonance imaging processing apparatus 100 may be a wearable device such as a watch, glasses, a headband, and a ring having a communication function and a data processing function.

[0075] Figure 1 It is a structural diagram showing the structure of a magnetic resonance imaging processing apparatus 100 according to an embodiment of the present invention.

[0076] Please refer to Figure 1 , the magnetic resonance imaging processing apparatus 100 may include a communication module 110, a memory 120, a processor 130, and a database (DB) 140.

[0077] The communication module 110 is linked to a communication network to provide a communication interface for the magnetic resonance imaging processing apparatus 100, and can function to transmit and receive data with an MRI imaging apparatus, a user terminal, and a management server. Here, the communication module may be a device including the following hardware and software, and the hardware and software are required to transmit and receive signals such as control signals or data signals through wired / wireless connections with other network devices.

[0078] The memory 120 may be a recording medium storing a magnetic resonance imaging processing program. Moreover, the memory 120 can function to temporarily or permanently store data processed by the processor. Here, the memory 120 may include a volatile storage medium or a non-volatile storage medium, but the scope of the present invention is not limited thereby.

[0079] The processor 130 may control the overall process of the magnetic resonance imaging processing apparatus 100 to perform a magnetic resonance imaging processing program. Each step of the process executed by the processor 130 will be described later in combination Figures 2 to 15 for explanation.

[0080] Here, the processor 130 may include all types of devices capable of processing data, such as a processor. Here, as an example, a "processor" may refer to a data processing device built into hardware, which has a circuit physically structured to execute the functions implemented by the code or instructions contained in a program. As described above, examples of data processing devices built into hardware include processing devices such as microprocessors, central processing units (CPUs), processor cores, multiprocessors, application-specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs), but the scope of the present invention is not limited thereto.

[0081] The database 140 may store various data required for the magnetic resonance imaging processing device 100 to execute a program. For example, the database may store learning data of an artificial neural network, magnetic resonance signal 310 data, k-space data, magnetic resonance image data, etc., which will be described later.

[0082] On the other hand, in the present invention, a "terminal" may be a wireless communication device that ensures portability and mobility. For example, it may be any type of handheld wireless communication device such as a smartphone, tablet computer, or laptop computer. Moreover, a "terminal" may also be a wearable device such as a watch, glasses, headband, and ring that has communication functions and data processing functions. Moreover, a "terminal" may also be a wired communication device such as a personal computer that can access other terminals or servers through a network.

[0083] Moreover, a network refers to a connection structure that enables information exchange between various nodes such as terminals and servers, including local area networks (LANs), wide area networks (WANs), the Internet (WWW), wired and wireless data communication networks, telephone networks, wired and wireless television communication networks, etc. Wireless data communication networks include, but are not limited to, 3G, 4G, 5G, the 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), mobile hotspots (Wi-Fi), Bluetooth communication, infrared communication, ultrasonic communication, visible light communication (VLC), and light fidelity (LiFi).

[0084] Next, a magnetic resonance image processing method performed by the magnetic resonance image processing apparatus 100 according to an embodiment of the present invention will be described in detail. Figure 2 FIG. is a flowchart showing a process of a magnetic resonance image processing method according to an embodiment of the present invention. Figure 3 FIG. is a schematic diagram showing a magnetic resonance image processing method according to an embodiment of the present invention.

[0085] Please refer to Figure 2 and Figure 3 , in the magnetic resonance image processing method performed by the magnetic resonance image processing apparatus 100 according to an embodiment of the present invention, a step S210 of obtaining a subsampled magnetic resonance signal 310 may be performed first.

[0086] Figure 4 FIG. is a schematic diagram for explaining the difference between full sampling and subsampling according to an embodiment of the present invention.

[0087] Please refer to Figure 4 , the subsampled magnetic resonance signal 310 may be a magnetic resonance signal 310 sampled at a sampling rate lower than the Nyquist sampling rate. Moreover, the subsampled magnetic resonance image is an image obtained by collecting the magnetic resonance signal 310 at a sampling rate lower than the Nyquist sampling rate. The subsampled magnetic resonance image may be an image including aliasing artifacts. Aliasing artifacts may be artificial images that occur in magnetic resonance images when the object being scanned is larger than the field of view (FOV). On the other hand, the fully sampled magnetic resonance image may be an image obtained by collecting k-space data at a sampling rate equal to or higher than the Nyquist sampling rate.

[0088] For example, the number of lines of the fully sampled magnetic resonance signal 310 can be n, while the number of lines of the subsampled magnetic resonance signal 310 can be n / 2. Here, if the reduction degree of the sampling lines is a factor of 1 / 2, the acceleration index of magnetic resonance imaging can be 2. If the reduction degree of the sampling lines is a factor of 1 / 3 or 1 / 4, the acceleration indices can be 3 and 4 respectively.

[0089] Moreover, subsampling methods can be classified into uniform subsampling and non-uniform subsampling. Uniform subsampling means sampling while maintaining a specified interval between sampling lines. In contrast, non-uniform subsampling generally means that the interval between sampling lines gradually narrows as it approaches the central part of the sampling data to increase sampling and gradually widens as it moves away from the central part to reduce sampling.

[0090] On the other hand, the magnetic resonance imaging processing device 100 is included in the MRI system and can obtain input data corresponding to the subsampled magnetic resonance signal 310 based on the magnetic resonance signal received from the RF coil. Moreover, the magnetic resonance imaging processing device 100 can also obtain input data corresponding to the subsampled magnetic resonance signal 310 from at least one of an external magnetic resonance imaging device, an external server, and a database.

[0091] Please refer back to Figure 2 And Figure 3 , after step S210, step S220 can be performed, that is, first k-space data 320 is obtained from the subsampled magnetic resonance signal 310 using the first parallel imaging technique.

[0092] Here, the parallel imaging technique is an image reconstruction technique that obtains fully sampled k-space data and / or high-accuracy k-space data equivalent to the magnetic resonance image and / or the magnetic resonance image from the subsampled magnetic resonance signal 310 and / or k-space data.

[0093] When performing image reconstruction based on parallel imaging technology, any parallel imaging technology that can adopt well-known technologies such as SPACE RIP (Sensitivity Profiles From an Array of Coils for Encoding and Reconstruction in Parallel), SMASH (Simultaneous acquisition of spatial harmonics), PILS (Partially Parallel Imaging With Localized Sensitivities), GRAPPA (Generalized Autocalibrating Partially Parallel Acquisitions), SPIRiT (iterative Self-consistent Parallel Imaging Reconstruction), etc. can be applied without restriction.

[0094] Figure 5 It is a schematic diagram for explaining the differences between the parallel imaging technologies GRAPPA and SPIRiT of an embodiment of the present invention.

[0095] Please refer to Figure 4 and Figure 5 , preferably, the first parallel imaging technology uses GRAPPA for image reconstruction. This GRAPPA takes the subsampled magnetic resonance signal 310 as input data and the fully sampled k-space image data as output.

[0096] That is, the first parallel imaging technology can be the GRAPPA technology. GRAPPA can correct the subsampled magnetic resonance signal 310 data into high-accuracy k-space data similar to the fully sampled k-space data. That is to say, when the number of lines of the fully sampled image data is n and the number of lines of the subsampled image data is n / 2, GRAPPA can infer and generate the remaining n / 2 unsampled lines from the n / 2 lines of the subsampled image data.

[0097] For example, although the first, third, and fourth lines are acquired from the sub-sampled magnetic resonance signal 310 of one channel of the RF coil and the second line is not acquired, a linear combination of the first and third lines closest to the second line can be performed to infer the second line. As described above, GR APPA can infer the remaining unacquired lines from the linear combination of adjacent lines and correct them into image data similar to fully sampled data. That is, image data with a high-accuracy k-space form can be generated for each channel.

[0098] Different from this GRAPPA, SPIRiT can infer the data of unsampled pixels by linearly combining the pixel data adjacent to the line including the pixel to be inferred and the data of the sampled lines adjacent to the corresponding line.

[0099] After step S220, step S230 of obtaining the first magnetic resonance image from the first k-space data 320 by performing an inverse Fourier (IFFT) operation can be performed.

[0100] GRAPPA can perform an inverse Fourier transform on the image data with a complete k-space form to generate a reconstructed image, that is, the first magnetic resonance image. Here, the first magnetic resonance image can include the target image and the image aliased in the phase-encoding direction Ky.

[0101] Please refer to Figure 4 and Figure 5 , the phase-encoding direction Ky can be a direction extending parallel to the stacking direction of the sampling lines during the sampling process of the magnetic resonance signal 310. Moreover, the readout direction Kx can be the direction in which the sampling lines extend. On the other hand, the Kz direction can be named the axial direction of the coil or the first direction Kz perpendicular to the phase-encoding direction Ky and the readout direction Kx described later, respectively.

[0102] On the other hand, if there are multiple sub-sampled magnetic resonance signals 310, multiple first k-space data 320 and multiple first magnetic resonance images can be obtained respectively. That is, if there are multiple RF coils for receiving the magnetic resonance signals 310, the multiple reconstructed images generated corresponding to the multiple magnetic resonance signals 310 received by each channel become the first magnetic resonance images.

[0103] After step S230, step S240 of preprocessing the first magnetic resonance image to generate the first input image data 340 can be performed.

[0104] Figure 6 is a flowchart showing the process of the first magnetic resonance image preprocessing method according to an embodiment of the present invention. Figure 7 is a schematic diagram showing the first magnetic resonance image preprocessing method according to an embodiment of the present invention.

[0105] Please refer to Figure 6 and Figure 7For the step S240 of preprocessing the first magnetic resonance image 720 to generate the first input image data 740, it may include step S241, that is, shifting the first magnetic resonance image 720 to obtain the shifted first magnetic resonance images 710 and 730 so that the aliased images A1 and A2 match the target image T.

[0106] Please refer to Figure 7 (a) thereof. For example, the first magnetic resonance image 720 may include the target image T of the object (brain) and the aliased images A1 and A2. Here, if the first magnetic resonance image 720 is shifted left / right along the phase encoding direction Ky, the shifted first magnetic resonance images 710 and 730 can be obtained.

[0107] Here, the shifting of the first magnetic resonance image 720 can be completed to the extent that the aliased images A1 and A2 match the target image T. For example, it can be shifted as follows, that is, the aliased image A2 of the object contained in the first magnetic resonance image 710 shifted to the left is arranged at the same position as the target image T of the object contained in the first magnetic resonance image 720 of the original image with respect to the plane formed by the phase encoding direction Ky and the readout direction Kx. Moreover, it can be shifted as follows, that is, the aliased image A1 of the object contained in the first magnetic resonance image 730 shifted to the right is arranged at the same position as the target image T of the object contained in the first magnetic resonance image 720 of the original image with respect to the plane formed by the phase encoding direction Ky and the readout direction Kx.

[0108] Moreover, please refer to Figure 7 (b) thereof. For the step S240 of preprocessing the first magnetic resonance image 720 to generate the first input image data 740, it may include step S242, that is, stacking the first magnetic resonance image 720 and the shifted first magnetic resonance images 710 and 730 along the first direction Kz perpendicular to the phase encoding direction Ky and the readout direction Kx respectively and grouping them to generate the first input image data 740.

[0109] Here, in Figure 7 (b) thereof, the first direction Kz is a direction parallel to the Kz direction. The phase encoding direction Ky is a direction parallel to the Ky direction, and the readout direction Kx is a direction parallel to the Kx direction. Moreover, the stacked first magnetic resonance image 720 and the multiple shifted first magnetic resonance images 710 and 730 can be bundled into a group to generate the first input image data 740 input to the first artificial neural network model 500 described later.

[0110] Figure 8 is a flowchart showing the process of the method for preprocessing the first magnetic resonance image to generate the first input image data 340 according to a modified embodiment of the present invention.

[0111] Please refer to Figure 8 As a variant embodiment of the present invention, the step S240' of preprocessing the first magnetic resonance image to generate the first input image data 340 may include the step S241' of combining multiple first magnetic resonance images to obtain a single second magnetic resonance image 330. The multiple first magnetic resonance images can be generated from the magnetic resonance signals 310 respectively received by multiple RF coil channels. Combining is a process of inputting multiple first magnetic resonance images received by multiple coils and outputting a single second magnetic resonance image 330.

[0112] Specifically, the step S241' of combining multiple first magnetic resonance images to obtain a single second magnetic resonance image 330 can be executed using the following mathematical formula 1:

[0113] Mathematical formula 1:

[0114] c i = S i ·m,

[0115]

[0116]

[0117] where, c i = the i-th first magnetic resonance image,

[0118] S i = the sensitivity matrix of the i-th coil,

[0119] m = the second magnetic resonance image based on complex-valued information,

[0120] m real = the second magnetic resonance image based on real-valued information.

[0121] Here, the second magnetic resonance image 330 can be obtained based on the first magnetic resonance image information and the sensitivity information of the RF coil. The first magnetic resonance image information can be the value of the second magnetic resonance image 330 information multiplied by the sensitivity information of each coil as a weighting value. On the other hand, the method of obtaining the second magnetic resonance image 330 can use the following method, that is, the method of dividing the first magnetic resonance image by the sensitivity information of each corresponding coil, but the problem of noise amplification may occur. Therefore, in order to prevent noise amplification, simplify calculation and accelerate calculation, it may be easier to obtain the second magnetic resonance image 330 based on real-valued information. In order to obtain the second magnetic resonance image 330 based on real-valued information, the sum of the sizes of the sensitivity matrices of each coil should be 1.

[0122] On the other hand, the step S241' of synthesizing a plurality of first magnetic resonance images to obtain a single second magnetic resonance image 330 can be performed using the following mathematical formula 2:

[0123] Mathematical formula 2

[0124]

[0125] where m real = the second magnetic resonance image based on real-valued information,

[0126] S i * = the conjugate transpose matrix of the sensitivity matrix of the i-th coil,

[0127] c i = the i-th first magnetic resonance image.

[0128] Mathematical formula 2 can be used to obtain the second magnetic resonance image 330 based on complex-valued information. At this time, compared with the case of using Mathematical formula 1, although the operation for obtaining the second magnetic resonance image 330 is more complex, a more accurate second magnetic resonance image 330 can be obtained.

[0129] Moreover, the step S240' of preprocessing the first magnetic resonance image to generate the first input image data 340 may include the step S242', that is, the step of preprocessing the second magnetic resonance image 330 synthesized using the Mathematical formula 1 or Mathematical formula 2 to generate the first input image data 340.

[0130] The second magnetic resonance image 330 synthesized from a plurality of first magnetic resonance images can include the target image and the image aliased in the phase encoding direction Ky. Moreover, the step S242' of preprocessing the second magnetic resonance image 330 to generate the first input image data 340 can perform the same process as the following step S240, which is the step of preprocessing the aforementioned first magnetic resonance image to generate the first input image data 340.

[0131] That is, the step S242' of preprocessing the second magnetic resonance image 330 to generate the first input image data 340 can include the following steps (not shown): moving the second magnetic resonance image 330 to obtain a shifted second magnetic resonance image 330 so that the aliased image matches the target image; stacking and grouping the second magnetic resonance image 330 and the shifted second magnetic resonance image 330 along a first direction Kz perpendicular to the phase encoding direction Ky and the readout direction Kx respectively to generate the first input image data 340.

[0132] In summary, the process of obtaining the second magnetic resonance image 330 by synthesizing multiple first magnetic resonance images and thereby generating the first input image data 340 generates highly accurate first input image data 340 based on more magnetic resonance signal 310 data, which can then be input into the artificial neural network described later.

[0133] After step S240, step S250 can be performed, that is, the first output magnetic resonance image 350 is obtained from the first input image data 340 using the first artificial neural network model 500.

[0134] Figure 9 It is a schematic diagram showing the steps of obtaining the first output magnetic resonance image 350 from the first input image data 340 using the first artificial neural network model 500 in an embodiment of the present invention.

[0135] Please refer to Figure 9 , the first artificial neural network model 500 can be an aggregate of algorithms that learn the correlation between at least one subsampled magnetic resonance image and at least one fully sampled magnetic resonance image using statistical machine learning results. The first artificial neural network model 500 can include at least one neural network. The neural network can include, but is not limited to, network models such as a deep neural network (DNN), a recurrent neural network (RNN), a bidirectional recurrent deep neural network (BRDNN), a multilayer perceptron (MLP), and a convolutional neural network (CNN).

[0136] For example, the first artificial neural network model 500 can be a model constructed by using a neural network to learn the correlation between at least one subsampled magnetic resonance image and at least one fully sampled magnetic resonance image in units of many pixels along at least one sampling line stacked in the phase encoding direction Ky.

[0137] Moreover, in addition to the subsampled magnetic resonance image and the fully sampled magnetic resonance image, the first artificial neural network model 500 can be constructed using various additional data. For example, at least one of the k-space data, real image data, imaginary image data, size image data, phase image data, and sensitivity data of the multi-channel RF coil, and the noise pattern image NP data corresponding to the first magnetic resonance image can be used as additional data.

[0138] Figure 10It is a schematic diagram showing the structure of a first artificial neural network model 500 according to an embodiment of the present invention.

[0139] Please refer to Figure 10 , preferably, the first artificial neural network model can have a convolutional network (U-net, Convolutional Networks for Biomedical Image Segmentation) structure for biomedical image segmentation that includes a contracting path 510 and an expanding path 520.

[0140] The contracting path 510 can include multiple convolutional (Conv) layers and pooling layers. For example, the contracting path 510 can include multiple combinations composed of a pooling layer that takes multiple 3x3 convolutional layers as input layers and applies wavelet transform.

[0141] Moreover, the expanding path 520 can include multiple convolutional layers and un-pooling layers. For example, the expanding path 520 includes multiple combinations composed of multiple 3x3 convolutional layers and un-pooling layers that apply inverse wavelet transform, and also includes an output layer 1x1 convolutional layer. On the other hand, each part of the expanding path 520 can input the noise pattern image NP described later to enable the first artificial neural network to learn.

[0142] Figure 11 The noise pattern image NP used as learning data for the artificial neural network model in an embodiment of the present invention. Figure 12 Used to illustrate the hallucination mitigation effect obtained by applying the noise pattern image NP as learning data to the artificial neural network model in an embodiment of the present invention.

[0143] Please refer to Figure 11 , the noise pattern image NP can be an image that is distinguished and displayed by applying a weight value based on the degree of noise occurrence for each part of the object in the reconstructed image. As an example, the noise pattern image NP can be a pattern image generated by comparing the image reconstructed by applying a subsampled magnetic resonance image to a general artificial neural network model with a fully sampled magnetic resonance image and based on the perceived noise.

[0144] Specifically, please refer to Figure 11(a), please view the comparison image 1130. This comparison image 1130 compares the reconstructed image 1120 output after taking the subsampled magnetic resonance image 1110 as the input of the artificial neural network model with the fully sampled magnetic resonance image. It can be seen that a large amount of noise appears in the central part of the image in the form of a vertical ellipse (Poor - conditiond, PC). Moreover, it can also be seen that the outer part of the image has less noise (Good - conditioned, GC). The noise pattern image NP is generated based on the difference in noise generation in each part of this object.

[0145] Please refer to Figure 11 (b), please view the comparison image 1160. This comparison image 1160 compares the reconstructed image 1150 output after taking the subsampled magnetic resonance image 1140 and the noise pattern image NP as the input of the artificial neural network model with the fully sampled magnetic resonance image. It can be seen that the noise has been reduced throughout the image.

[0146] Moreover, please refer to Figure 12 , from the left, the fully sampled magnetic resonance image 1210, the subsampled magnetic resonance image 1220, the reconstructed image 1230 output by the artificial neural network model using the subsampled magnetic resonance image 1220 as the input, and the reconstructed image 1240 output by the artificial neural network model using the subsampled magnetic resonance image 1220 and the noise pattern image NP as the input can be seen in sequence.

[0147] At the lower end of each image, a part of the object image is enlarged. Here, it can be seen that the aliased image in the enlarged part h2 of the subsampled magnetic resonance image 1220 overlaps with the target image and there is a difference from the enlarged part h1 of the fully sampled magnetic resonance image. This phenomenon is called the hallucination phenomenon, and it can also be seen in the enlarged part h3 of the reconstructed image that the image is changed due to the subsampled magnetic resonance image. In contrast, in the enlarged part h4 of the reconstructed image 1240 using the noise pattern image NP, the hallucination phenomenon is alleviated similarly to the enlarged part h1 of the fully sampled magnetic resonance image 1210.

[0148] Referring to the application effect of this noise pattern image NP, as a preferred embodiment, the first artificial neural network model 500 can take the noise pattern image NP generated based on the first magnetic resonance image as the input. That is, the first artificial neural network model 500 can perform the following process: taking the first input image data 340 and the noise pattern image NP as input data and outputting the first output magnetic resonance image 350.

[0149] For example, the noise pattern image NP can be input into at least one of the convolutional layers and the de-pooling layers of the expanding path 520 of the first artificial neural network model 500. Specifically, the noise pattern image NP can be input into the convolutional layer that is the output layer of the expanding path 520, and can be input after being pooled according to the respective image sizes of the de-pooling layers.

[0150] In one embodiment, the magnetic resonance image processing device 100 can construct the aforementioned first artificial neural network model 500 by itself. In another embodiment, the magnetic resonance image processing device 100 can also obtain the first artificial neural network model 500 constructed by an external server or an external device from the external server or the external device. Moreover, the magnetic resonance image processing device 100 obtains the first output magnetic resonance image 350 based on the first artificial neural network model 500 that utilizes a neural network, and the first output magnetic resonance image 350 is a reconstructed image of the subsampled magnetic resonance image, thereby achieving the following purposes, that is, accelerating the image acquisition speed and improving the image quality.

[0151] Here, after the obtained first output magnetic resonance image 350 is provided to the user terminal, it can be applied to the diagnosis of the diseased part of the patient. On the other hand, the magnetic resonance image processing steps can be further performed to improve the accuracy of the first output magnetic resonance image 350.

[0152] Figure 13 It is a flowchart showing the process of a magnetic resonance image processing method that is an extended embodiment of the magnetic resonance image processing method including an embodiment of the present invention. Figure 14 It is a schematic diagram showing the magnetic resonance image processing method of an extended embodiment of the present invention.

[0153] Please refer to Figure 13 and Figure 14 , after step S250, the step S260 of obtaining the second k-space data 360 from the first output magnetic resonance image 350 by using Fourier transform can be performed.

[0154] Figure 15 It is a flowchart showing the step of obtaining the second output magnetic resonance image 400 and the step of obtaining the second k-space data 360 of an extended embodiment of the present invention.

[0155] Please refer to Figure 15 , the step S260 of obtaining the second k-space data 360 may include step S261, that is, decomposing the first output magnetic resonance image 350 to obtain a plurality of second output magnetic resonance images 400.

[0156] Specifically, the step of decomposing the first output magnetic resonance image 350 to obtain a plurality of second output magnetic resonance images 400 can be performed using the following mathematical formula 3 or mathematical formula 4:

[0157] Mathematical formula 3:

[0158]

[0159] where,

[0160] S i = the sensitivity matrix of the i-th coil,

[0161]

[0162] Mathematical formula 3 is a mathematical formula applicable when obtaining the first output magnetic resonance image 350 based on complex-valued information.

[0163] Mathematical formula 4:

[0164]

[0165] where,

[0166] S i = the sensitivity matrix of the i-th coil,

[0167]

[0168]

[0169] Mathematical formula 4 is a mathematical formula applicable when obtaining the first output magnetic resonance image 350 based on real-valued information. Here, in the aforementioned synthesis process, the phase information of the complex value is extracted to obtain the phase information of the first output magnetic resonance image 350. Thus, the first output magnetic resonance image 350 can be decomposed based on the sensitivity information of each RF coil to generate a plurality of magnetic resonance images.

[0170] Moreover, the step S260 of obtaining the second k-space data 360 may include the step S262 of obtaining the second k-space data 360 from the second output magnetic resonance image 400 using Fourier transform. Thus, Fourier transform can be applied to the plurality of magnetic resonance images generated from the first output magnetic resonance image 350 to obtain a plurality of second k-space data 360.

[0171] After the step S260, the step S270 of obtaining the third k-space data 370 from the second k-space data 360 using the second parallel imaging technique can be performed.

[0172] Here, the second parallel imaging technique may be the following parallel imaging technique that can obtain the third k-space data 370 which is a full-sampled correction of the second full-sampled k-space data 360. Preferably, the second parallel imaging technique may use the aforementioned SPIRiT technique.

[0173] Here, the first parallel imaging technique uses GRAPPA while the second parallel imaging technique uses SPIRiT, so that a highly accurate reconstructed image can be obtained during the process of obtaining the reconstructed image.

[0174] After step S270, step S280 of obtaining the third magnetic resonance image from the third k-space data 370 by using inverse Fourier transform can be performed.

[0175] After step S280, step S290 of preprocessing the third magnetic resonance image to generate the second input image data 390 can be performed.

[0176] Here, the method of preprocessing the third magnetic resonance image to generate the second input image data 390 may perform the same process as the following steps, which are the steps of preprocessing the aforementioned first magnetic resonance image to generate the first input image data 340 in step S240.

[0177] On the other hand, in a modified embodiment, step S290 of preprocessing the third magnetic resonance image to generate the second input image data 390 may include the following steps, that is, the step of combining multiple third magnetic resonance images to obtain a single fourth magnetic resonance image 380.

[0178] Here, the method of combining multiple third magnetic resonance images to obtain a single fourth magnetic resonance image 380 may use the same method as the following method, which obtains a single second magnetic resonance image 330 by combining the aforementioned first magnetic resonance images.

[0179] Moreover, step S290 of preprocessing the third magnetic resonance image to generate the second input image data 390 may include the following steps, that is, preprocessing the fourth magnetic resonance image 380 to generate the second input image data 390.

[0180] Here, the method of preprocessing the fourth magnetic resonance image 380 to generate the second input image data 390 may use the same method as the following method, which preprocesses the aforementioned second magnetic resonance image 330 to generate the first input image data 340.

[0181] After step S290, step S300 of obtaining the third output magnetic resonance image from the second input image data 390 by using the second artificial neural network model 600 can be performed.

[0182] Here, the second artificial neural network model 600 may use the same artificial neural network model as the aforementioned first artificial neural network model 500. Alternatively, the first artificial neural network model 500 that has further learned the first input image data 340 may also be used, or an artificial neural network model different from the first artificial neural network model 500 may be used.

[0183] After step S300, the second output magnetic resonance image 400 is further subjected to Fourier transform and / or decomposition and Fourier transform to obtain the fourth k-space data 410. Based on this, the process from step S260 to step S300 can be repeated as a group.

[0184] Moreover, the second output magnetic resonance image 400 can be provided to the user terminal and used as a diagnostic image of the patient's lesion site.

[0185] The following describes an experimental example implemented by the magnetic resonance image processing apparatus 100 and its method according to an embodiment of the present invention.

[0186] Figure 16 An experimental example of the present invention shows a reconstructed image with an acceleration factor of 4 and a comparative example. Here, the acceleration index is 4, and the number of subsampled lines can be 1 / 4 of the number of fully sampled lines.

[0187] Please refer to Figure 16 , from the left, the fully sampled magnetic resonance image 1610, the magnetic resonance image 1620 reconstructed by the GRAPPA technique, the magnetic resonance image 1630 reconstructed by CG-SENSE (conjugate gradient sensitivity encoding), the magnetic resonance image 1640 reconstructed by a general artificial neural network model, and the magnetic resonance image 1650 reconstructed by the magnetic resonance image processing apparatus 100 of the present invention can be seen in sequence. Moreover, some images can be seen below each image. These images are images in which the difference points when compared with the fully sampled magnetic resonance image 1610 are magnified by 4 times as error images, and it can be seen that the error of the magnetic resonance image 1650 reconstructed by the magnetic resonance image processing apparatus 100 of the present invention is the smallest.

[0188] Figure 17 An experimental example of the present invention shows a reconstructed image with an acceleration factor of 6 and a comparative example. Here, the acceleration index is 6, and the number of subsampled lines can be 1 / 6 of the number of fully sampled lines.

[0189] Please refer to Figure 17, from the left, the fully sampled magnetic resonance image 1710, the magnetic resonance image 1720 reconstructed by the GRAPPA technique, the magnetic resonance image 1730 reconstructed by CG-SENSE, the magnetic resonance image 1740 reconstructed by a general artificial neural network model, and the magnetic resonance image 1750 reconstructed by the magnetic resonance image processing apparatus 100 of the present invention can be seen in sequence. Moreover, some images can be seen below each image, and these images are the images obtained by magnifying by 4 times the error images that analyze the difference points in comparison with the fully sampled magnetic resonance image 1710. It can be seen that the error of the magnetic resonance image 1750 reconstructed by the magnetic resonance image processing apparatus 100 of the present invention is the smallest.

[0190] The magnetic resonance image processing apparatus 100 and its method according to the embodiment of the present invention described above can generate a high-accuracy magnetic resonance image through reconstructed images even if the time for capturing a magnetic resonance image using a magnetic resonance imaging apparatus is shortened.

[0191] Moreover, the magnetic resonance image processing apparatus 100 and its method according to the embodiment of the present invention acquire a plurality of magnetic resonance images, perform preprocessing on them to generate input data, and provide a high-accuracy output magnetic resonance image reconstructed using the first artificial neural network model 500 and the second artificial neural network model 600, thereby enabling easy diagnosis of the diseased part of a patient.

[0192] On the other hand, the magnetic resonance image processing method according to an embodiment of the present invention can also be implemented in the form of a recording medium, and the recording medium includes computer-executable instructions such as computer-executable program modules. The computer-readable recording medium can be any available medium accessible by a computer, including volatile and non-volatile storage devices, separable and non-separable media. Moreover, the computer-readable medium can include a computer storage medium. The computer storage medium includes volatile and non-volatile, separable and non-separable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Although the method and system of the present invention have been described in association with specific embodiments herein, some or all of the elements or actions thereof can be implemented using a computer system having a general hardware architecture.

[0193] The foregoing has merely illustrated by way of example the technical idea of the present invention. However, those of ordinary skill in the technical field to which the present invention pertains can implement various deformations and modifications without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in this specification are merely for illustration and not for limiting the technical idea of the present invention, and these embodiments cannot limit the scope of the technical idea of the present invention. The protection scope of the present invention should be interpreted according to the scope of the invention claimed, and all technical ideas within the equivalent scope should be included in the scope of the rights of the present invention.

Claims

1. A magnetic resonance imaging processing method based on a magnetic resonance imaging processing device, characterized in that: It includes the following steps: Preprocessing the first magnetic resonance image obtained from the subsampled magnetic resonance signal to generate first input image data; and Using a first artificial neural network model to obtain a first output magnetic resonance image from the first input image data, If there are multiple subsampled magnetic resonance signals, then multiple first magnetic resonance images are obtained, The step of preprocessing the first magnetic resonance image to generate first input image data includes the following steps: Synthesizing multiple first magnetic resonance images to obtain a single second magnetic resonance image; and Preprocessing the second magnetic resonance image to generate first input image data.

2. The magnetic resonance imaging processing method according to claim 1, characterized in that: The step of synthesizing multiple first magnetic resonance images to obtain a single second magnetic resonance image is performed using the following mathematical formula 1; Mathematical formula 1: c i = S i · m, where c i = the i-th first magnetic resonance image, S i = sensitivity matrix of the i-th coil, m = the second magnetic resonance image based on complex value information, m real = the second magnetic resonance image based on real-valued information.

3. The magnetic resonance imaging processing method according to claim 1, characterized in that: The step of synthesizing multiple first magnetic resonance images to obtain a single second magnetic resonance image is performed using the following mathematical formula 2; Mathematical formula 2: where m = the second magnetic resonance image based on complex value information, S i * = the conjugate transpose matrix of the sensitivity matrix of the i-th coil, c i = the i-th first magnetic resonance image.

4. The magnetic resonance imaging processing method according to claim 1, characterized in that: The first magnetic resonance image includes a target image and an aliased image in the phase encoding direction, The step of preprocessing the first magnetic resonance image to generate first input image data includes the following steps: Moving the first magnetic resonance image to obtain a shifted first magnetic resonance image so that the aliased image matches the target image; and Stacking and grouping the first magnetic resonance image and the shifted first magnetic resonance image along a first direction perpendicular to the phase encoding direction and the readout direction respectively to generate first input image data.

5. The magnetic resonance imaging processing method according to claim 1, characterized in that: The second magnetic resonance image includes a target image and an aliased image in the phase encoding direction, The step of preprocessing the second magnetic resonance image to generate first input image data includes the following steps: Moving the second magnetic resonance image to obtain a shifted second magnetic resonance image so that the aliased image matches the target image; and Stacking and grouping the second magnetic resonance image and the shifted second magnetic resonance image along a first direction perpendicular to the phase encoding direction and the readout direction respectively to generate first input image data.

6. The magnetic resonance image processing method according to claim 1, wherein The first artificial neural network model receives a noise pattern image generated based on the first magnetic resonance image.

7. The magnetic resonance imaging processing method according to claim 6, characterized in that: The first artificial neural network model has an artificial neural network structure including a contracting path and an expanding path, The contracting path includes multiple convolutional layers and pooling layers, and the expanding path includes multiple convolutional layers and de-pooling layers, The noise pattern image is input into at least one of the plurality of convolutional layers and de-pooling layers of the expansion path.

8. The magnetic resonance image processing method according to claim 1, wherein it includes the following steps, that is, obtaining second k-space data from the first output magnetic resonance image by using Fourier operation, the step of obtaining the second k-space data includes the following steps: decomposing the first output magnetic resonance image to obtain a plurality of second output magnetic resonance images; and obtaining second k-space data from the second output magnetic resonance image by using Fourier operation.

9. The magnetic resonance image processing method according to claim 8, wherein the step of decomposing the first output magnetic resonance image to obtain a plurality of second output magnetic resonance images is performed by using the following mathematical formula 3 or mathematical formula 4; Mathematical formula 3: Among them, S i = sensitivity matrix of the i-th coil, Mathematical formula 4: Among them, S i = the sensitivity matrix of the i-th coil, 10. The magnetic resonance image processing method according to claim 8, wherein It includes the following steps: obtaining third k-space data from the second k-space data by using a second parallel imaging technique.

11. The magnetic resonance image processing method according to claim 10, wherein It includes the following steps: obtaining a third magnetic resonance image from the third k-space data by using inverse Fourier operation; preprocessing the third magnetic resonance image to generate second input image data; and obtaining a third output magnetic resonance image from the second input image data by using a second artificial neural network model.

12. A magnetic resonance image processing apparatus for performing the magnetic resonance image processing method, wherein it includes: a memory storing a magnetic resonance image processing program; and a processor executing the program, as the program is executed, the processor preprocesses a first magnetic resonance image obtained from a sub-sampled magnetic resonance signal to generate first input image data, and obtains a first output magnetic resonance image from the first input image data by using a first artificial neural network model, if there are multiple sub-sampled magnetic resonance signals, then multiple first magnetic resonance images are obtained, preprocessing the first magnetic resonance image to generate first input image data includes: synthesizing multiple first magnetic resonance images to obtain a single second magnetic resonance image, and preprocessing the second magnetic resonance image to generate first input image data.

13. A computer-readable recording medium, characterized in that, It stores a program capable of executing the method according to claim 1.

Images