Magnetic resonance image processing method and apparatus using scan parameters

Abstract

A magnetic resonance image processing method according to one embodiment of the present invention includes the steps of acquiring a low-quality training image, a first parameter group including at least one scan parameter applied when acquiring the training image, a high-quality label image, and a second parameter group including at least one scan parameter applied when acquiring the label image, and training an artificial neural network model using the training image, the first parameter group, the label image, and the second parameter group.

Description

[Technical field]

[0001] The present invention relates to a method and apparatus for processing magnetic resonance images, and more particularly to a method and apparatus for processing magnetic resonance images in which an artificial neural network is trained using scan parameters to acquire magnetic resonance images. [Background technology]

[0002] In general, medical imaging devices are devices that obtain physical information of a patient and provide images, including X-ray devices, ultrasound diagnostic devices, computed tomography devices, and magnetic resonance imaging (MRI) devices.

[0003] Here, magnetic resonance imaging uses a harmless magnetic field and non-ionizing radiation to induce nuclear magnetic resonance in hydrogen nuclei in the human body, thereby imaging the density and physical / chemical properties of the nuclei. Magnetic resonance imaging devices have a relatively flexible imaging condition, contain various diagnostic information on soft tissues, and provide images with excellent contrast, so they occupy an important position in the field of diagnosis using medical images.

[0004] Meanwhile, a radiographer can perform magnetic resonance imaging after setting specific parameters indicating imaging conditions in order to obtain desired data. If the parameters are changed, the acquired magnetic resonance signal changes, and therefore the work to be performed for image restoration may change. That is, the image restoration work needs to be performed taking into account not only the magnetic resonance signal but also the parameters used for imaging. Summary of the Invention [Problem to be solved by the invention]

[0005] The present invention is directed to solving the above-mentioned problems of the prior art, and relates to a method and apparatus for processing a magnetic resonance image using scan parameters, which reconstructs a magnetic resonance image using scan parameters used in magnetic resonance imaging.

[0006] However, the technical objectives that the present invention aims to achieve are not limited to those mentioned above, and other technical objectives may exist. [Means for solving the problem]

[0007] As a technical means for achieving the above-mentioned technical object, according to one embodiment of the present invention, a magnetic resonance image processing method is characterized by including a step of acquiring a low-quality training image, a first parameter group including at least one scan parameter applied when acquiring the training image, a high-quality label image, and a second parameter group including at least one scan parameter applied when acquiring the label image, and a step of training an artificial neural network model using the training image, the first parameter group, the label image, and the second parameter group.

[0008] As an embodiment, the scan parameters may include at least one of a number of excitations, which is a repetition number when repeatedly acquiring a line of first k-space data of a magnetic resonance signal, a phase resolution, which is a value obtained by dividing the number of sampled lines in a phase encoding direction of the first k-space data by a predetermined reference value, and an acceleration factor, which is a value obtained by dividing the number of fully sampled signal lines of the first k-space data by the number of sub-sampled signal lines.

[0009] In one embodiment, the training image may be obtained by subsampling a magnetic resonance signal, and the label image may be obtained by fully sampling the magnetic resonance signal.

[0010] In one embodiment, the number of excitations of the training images may be less than the number of excitations of the label images.

[0011] In one embodiment, a phase resolution of the training image may be smaller than a phase resolution of the label image.In one embodiment, training the artificial neural network model may include pre-processing the first parameter group and the second parameter group so that the size of the first parameter group and the size of the second parameter group correspond to the number of channels of the artificial neural network model.

[0012] In one embodiment, the pre-processing step may include inputting the first parameter group and the second parameter group into a pre-processing artificial neural network model to generate first attention data and second attention data, respectively, and training the artificial neural network model may include inputting the first attention data and the second attention data into the artificial neural network model.

[0013] As an example, the pre-processing artificial neural network model may include a fully connected layer.

[0014] In one embodiment, the method may further include inputting an input image and a third parameter group including at least one scan parameter applied when acquiring the input image into the artificial neural network model to generate a reconstructed image.

[0015] As a technical means for achieving the above-mentioned technical object, according to one embodiment of the present invention, a magnetic resonance image processing method includes a step of inputting an input image and a third parameter group including at least one scan parameter applied when acquiring the input image to an artificial neural network model trained using a low-quality training image, a first parameter group including at least one scan parameter applied when acquiring the training image, a high-quality label image, and a second parameter group including at least one scan parameter applied when acquiring the label image, and a step of outputting a restored image from the artificial neural network model.

[0016] As a technical means for achieving the above-mentioned technical object, according to one embodiment of the present invention, there is provided a magnetic resonance image processing device for executing a magnetic resonance image processing method, comprising: a memory in which a magnetic resonance image processing program is stored; and a processor for executing the program, wherein the processor is characterized in that, by executing the program, it trains an artificial neural network model using a low-quality training image, a first parameter group including at least one scan parameter applied when acquiring the training image, a high-quality label image, and a second parameter group including at least one scan parameter applied when acquiring the label image. Effect of the Invention

[0017] According to the above-mentioned solution to the problem of the present invention, the present invention performs learning of an artificial neural network using not only magnetic resonance signals but also scan parameters used for imaging, thereby improving the accuracy of magnetic resonance images acquired using the artificial neural network.

[0018] In addition, by adjusting the values ​​of the scan parameters used for learning, the magnetic resonance image can be restored so that data desired by the user is emphasized, thereby improving user convenience. [Brief description of the drawings]

[0019] [Figure 1] 1 is a block diagram showing the configuration of a magnetic resonance imaging processing apparatus according to an embodiment of the present invention; [Figure 2a] 2 is a diagram illustrating a method for processing magnetic resonance images in the k-space domain according to an embodiment of the present invention; [Figure 2b] 2 is a diagram illustrating a method for processing magnetic resonance images in the k-space domain according to an embodiment of the present invention; [Figure 3a] 1 is a diagram illustrating a method for processing magnetic resonance images in the image domain according to an embodiment of the present invention; [Figure 3b] 1 is a diagram illustrating a method for processing magnetic resonance images in the image domain according to an embodiment of the present invention; [Figure 4] FIG. 2 is a schematic diagram illustrating the difference between sub-sampling and full sampling according to an embodiment of the present invention. [Diagram 5] FIG. 2 is a schematic diagram illustrating obtaining second k-space data from first k-space data using a first artificial neural network model according to one embodiment of the present invention. [Figure 6] 4 is a schematic diagram illustrating acquiring a second magnetic resonance image from a first magnetic resonance image using a second artificial neural network model according to an embodiment of the present invention. FIG. [Figure 7] FIG. 4 is a schematic diagram illustrating a method of training a second artificial neural network model according to an embodiment of the present invention. [Figure 8] FIG. 4 is a schematic diagram illustrating a method of inputting scan parameters into a second artificial neural network model according to an embodiment of the present invention. [Figure 9] 4 is a flowchart showing the procedure of a magnetic resonance image processing method using scan parameters according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] Hereinafter, the embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. In the drawings, in order to clearly explain the present invention, parts that are not related to the description are omitted, and similar parts are designated by similar reference numerals throughout the specification.

[0021] This specification does not describe all elements of the embodiments, and contents common in the technical field to which the present invention belongs or contents overlapping between the embodiments are omitted. The term "part" or "portion" used in the specification can be embodied by software or hardware, and depending on the embodiment, multiple "parts" can be embodied by a single element, or a single "part" can include multiple elements.

[0022] As used herein, an "image" may refer to 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 acquired by a medical imaging device such as a magnetic resonance imaging device, a computed tomography (CT) device, an ultrasound imaging device, or an X-ray imaging device.

[0023] In this specification, an "object" refers to an object to be imaged, and may include a human, an animal, or a part thereof. For example, an object may include a part of a body (such as an organ) or a phantom. A phantom refers to a material having a volume that closely resembles the density and effective atomic number of a living organism, and may include a spherical phantom having properties similar to those of a human body.

[0024] The Magnetic Resonance Image (MRI) system is a system that obtains images of the tomographic regions of a subject by expressing the intensity of the Magnetic Resonance (MR) signal in terms of light and dark contrast in response to the Radio Frequency (RF) signal generated by a magnetic field of a specific strength.

[0025] In an MRI system, a main magnet forms a static magnetic field, and aligns the direction of a magnetic dipole moment of a specific atomic nucleus of a subject located in the static magnetic field with the direction of the static magnetic field. A gradient coil applies a gradient signal to the static magnetic field to form a gradient magnetic field, thereby inducing different resonance frequencies for different parts of the subject. An RF coil can irradiate a magnetic resonance signal according to the resonance frequency of a desired part for image acquisition. In addition, the RF coil can receive magnetic resonance signals of different resonance frequencies radiated from various parts of the subject as a gradient magnetic field is formed. The MRI system acquires an image by applying an image reconstruction technique to the magnetic resonance signal received in this step. In addition, the MRI system can reconstruct the multiple magnetic resonance signals into image data by performing serial or parallel signal processing on the multiple magnetic resonance signals received by the multi-channel RF coil.

[0026] Hereinafter, a magnetic resonance imaging apparatus according to an embodiment of the present invention will be described.

[0027] FIG. 1 is a block diagram showing the configuration of a magnetic resonance imaging processing apparatus according to an embodiment of the present invention.

[0028] 1, the magnetic resonance image processing device 100 may be, but is not limited to, an MRI system capable of acquiring a magnetic resonance image by sensing a magnetic resonance signal by itself, an image processing device for processing an image acquired from an external source, a smartphone, a tablet PC, a PC, a smart TV, a micro server, a cloud server, other home appliances, and other mobile or non-mobile computing devices having a processing function for a magnetic resonance image, etc. Also, the magnetic resonance image processing device 100 may be a wearable device such as a watch, glasses, a hair band, or a ring having a communication function and a data processing function.

[0029] In addition, the magnetic resonance image processing device 100, 100 according to one embodiment of the present invention may relate to a magnetic resonance image processing device 100 that communicates with a medical image archiving and communication system (PACS) used in a medical institution or the magnetic resonance image processing device 100 to transmit and receive medical image data, and restores magnetic resonance image data using an artificial neural network model.

[0030] The magnetic resonance image processing apparatus 100 according to an embodiment of the present invention may be implemented in the form of a cloud computing system. Cloud computing is a computing environment in which IT-related services such as data storage, network, and content use can be used comprehensively through a server on the Internet. Alternatively, the magnetic resonance image processing apparatus 100 may be implemented by various types of computing systems capable of executing a magnetic resonance image processing method, such as server computing, edge computing, and serverless computing.

[0031] Specifically, each of the magnetic resonance imaging processing devices 100 according to an embodiment of the present invention may include a communication module 110 , a memory 120 , a processor 130 , and a database 140 .

[0032] The communication module 110 provides a communication interface to the magnetic resonance image processing apparatus 100 in cooperation with a communication network. The magnetic resonance image processing apparatus 100 can transmit and receive data to and from a client terminal, a PACS terminal, and a PACS server, which will be described later, using the communication module 110. Here, the communication module 110 may be a device including hardware and software necessary to transmit and receive signals, such as control signals or data signals, to and from other network devices via wired or wireless connections.

[0033] For example, the communication module 110 may communicate via LAN, Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), Wireless Broadband Internet (WiBro), 5th Generation Mobile Communication (5G), Ultra Wide-Band, ZigBee, Radio Frequency (RF) communication, Wireless LAN, Wireless Fidelity, Near Field Communication (NFC), Bluetooth, infrared communication, etc. However, this is merely an example, and various wired and wireless communication technologies applicable in the technical field may be used depending on the embodiment to which the present invention is applied.

[0034] Meanwhile, in the present invention, a "terminal" is a wireless communication device that is guaranteed to be portable and mobile, and may be any kind of handheld-based wireless communication device such as a smartphone, a tablet PC, or a notebook PC. A "terminal" may also be a wearable device such as a watch, glasses, a hair band, or a ring that has a communication function and a data processing function. A "terminal" may also be a wired communication device such as a PC that can be connected to another terminal or a server via a network.

[0035] In general, a medical institution may be provided with a client terminal that controls the medical image data capturing device in cooperation with the medical image data capturing device or manages the transmission of the medical image data, and a PACS terminal that is embedded with a PACS program that allows medical staff to view, process, and manage the medical image data. The client terminal may be a terminal embedded with a program that provides a user interface (UI) for outputting user login, worklist, and image processing details. The PACS terminal may be a terminal embedded with a program that provides a user interface for transmitting medical image data and personal information data stored in the PACS server to the magnetic resonance image processing device 100, receiving medical image data restored by an artificial neural network model, and storing it in the PACS server or outputting it on a display.

[0036] The memory 120 may be a recording medium in which a magnetic resonance image processing program is recorded. The memory 120 may also 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 thereto.

[0037] The processor 130 can control the entire process of the magnetic resonance image processing program executed in the magnetic resonance image processing apparatus 100. Each step of the process executed by the processor 130 will be described later.

[0038] Here, the processor 130 may include all kinds of devices capable of processing data, such as a processor. Here, the term "processor" may refer to a data processing device built into hardware, having a circuit physically structured to achieve a function expressed by a code or instruction included in a program. Examples of such data processing devices built into hardware include a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), and the like, but the scope of the present invention is not limited thereto.

[0039] The database 140 may store various data necessary for the magnetic resonance imaging processing apparatus 100 to execute a program. For example, the database may store learning data for an artificial neural network, magnetic resonance signal data, k-space data, magnetic resonance imaging data, and the like, which will be described later.

[0040] A method for acquiring a magnetic resonance image by performing accelerated imaging will be described in detail below with reference to FIGS.

[0041] Figures 2a and 2b are drawings showing a method of processing magnetic resonance images in the k-space domain according to one embodiment of the present invention, Figures 3a and 3b are drawings showing a method of processing magnetic resonance images in the image domain according to one embodiment of the present invention, Figure 4 is a schematic diagram for explaining the difference between sub-sampling and full sampling according to one embodiment of the present invention, Figure 5 is a schematic diagram showing obtaining second k-space data from first k-space data using a first artificial neural network model according to one embodiment of the present invention, and Figure 6 is a schematic diagram showing obtaining a second magnetic resonance image from a first magnetic resonance image using a second artificial neural network model according to one embodiment of the present invention.

[0042] In this specification, imaging in which a magnetic resonance image is obtained using sub-sampled magnetic resonance signals is referred to as accelerated imaging.

[0043] Accelerated imaging may mean imaging in which the magnetic resonance signal obtained is reduced compared to normal imaging. For example, normal imaging may mean acquiring a fully sampled magnetic resonance signal, and accelerated imaging may mean acquiring a sub-sampled magnetic resonance signal. For example, normal imaging may be imaging that takes one hour to acquire 100% of the magnetic resonance signal, and accelerated imaging may be imaging that takes 30 minutes to acquire 50% of the magnetic resonance signal. Thus, a high-quality magnetic resonance image can be acquired by normal imaging, and a relatively low-quality magnetic resonance image can be acquired by accelerated imaging. A low-quality magnetic resonance image is an image with a lower resolution or an image with more noise than a high-quality magnetic resonance image.

[0044] For example, the accelerated imaging may be imaging with a reduced number of excitations (NEX) compared to imaging used clinically. For example, the accelerated imaging may be imaging with a reduced phase resolution compared to imaging used clinically. For example, the accelerated imaging may be imaging with a higher acceleration factor compared to imaging used clinically and applying a parallel imaging technique. In addition, the accelerated imaging may be not only an image actually captured through a magnetic resonance imaging device, but also an image obtained by processing a general image captured by a computing simulation in the same or similar manner as the accelerated imaging.

[0045] Referring to Figures 2a and 2b, a magnetic resonance image processing method performed by a magnetic resonance image processing device according to one embodiment of the present invention can perform a step of acquiring a sub-sampled magnetic resonance signal 210 by accelerated imaging prior to step S110.

[0046] 4, the sub-sampled magnetic resonance signal 210 may be a magnetic resonance signal sampled at a sampling rate lower than the Nyquist sampling rate. Also, the sub-sampled magnetic resonance image is an image acquired by sampling the magnetic resonance signal at a sampling rate lower than the Nyquist sampling rate. Meanwhile, the fully sampled magnetic resonance image may be an image acquired by sampling k-space data at a sampling rate equal to or higher than the Nyquist sampling rate.

[0047] For example, the number of lines of the fully sampled magnetic resonance signal may be n, and the number of lines of the sub-sampled magnetic resonance signal 210 may be n / 2. Here, if the reduction in the sampling lines is a 1 / 2 multiple, the acceleration factor of the magnetic resonance imaging is said to be 2. If the reduction in the sampling lines is a 1 / 3 multiple or a 1 / 4 multiple, the acceleration factors are said to be 3 and 4, respectively.

[0048] The subsampling method can be classified into uniform subsampling and non-uniform subsampling. Uniform subsampling can be performed by maintaining a constant interval between sampled lines. Meanwhile, non-uniform subsampling can generally mean performing more sampling by decreasing the interval between sampled lines toward the center of the sampled data, and performing less sampling by increasing the interval between sampled lines away from the center.

[0049] Meanwhile, the magnetic resonance imaging processing device is included in the MRI system and can obtain input data corresponding to the sub-sampled magnetic resonance signal 210 based on the magnetic resonance signal received from the RF coil. Also, the magnetic resonance imaging processing device can obtain input data corresponding to the sub-sampled magnetic resonance signal 210 from at least one of an external magnetic resonance imaging device, an external server, and a database.

[0050] 2a and 2b, a step of acquiring first k-space data 220 from a sub-sampled magnetic resonance signal 210 using a parallel imaging technique (S110) may be performed.

[0051] Here, the parallel imaging technique is a type of image restoration technique for obtaining high-accuracy k-space data and / or magnetic resonance images, such as fully sampled k-space data and / or magnetic resonance images, from sub-sampled magnetic resonance signals 210 and / or k-space data.

[0052] In performing image restoration using parallel imaging techniques, any technique that is applicable to parallel imaging techniques, such as well-known techniques 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 Sensitivity), GRAPPA (Generalized Autocalibrating Partially Parallel Acquisitions), and SPIRiT (Iterative Self-consistent Parallel Imaging Reconstruction), can be applied without any restrictions.

[0053] After step S110, a step S120 of acquiring second k-space data 230 from the first k-space data 220 by the first artificial neural network model 300 may be performed.

[0054] 5, the first artificial neural network model 300 may be a set of algorithms that use statistical machine learning results to learn correlations between at least one sub-sampled first k-space data 220 and at least one full-sampled first k-space data 220. The first artificial neural network model 300 may include at least one neural network. The neural network may include, but is not limited to, a network model such as a deep neural network (DNN), a recurrent neural network (RNN), a bidirectional recurrent deep neural network (BRDNN), a multilayer perceptron (MLP), a convolutional neural network (CNN), or a U-net.

[0055] For example, the first artificial neural network model 300 may be a model constructed by using a neural network to learn the correlation between at least one sub-sampled first k-space data 220 and at least one fully sampled first k-space data 220 by pixel of at least one sampling line stacked according to the phase encoding direction (Ky).

[0056] 4, the phase encoding direction (Ky) may be a direction extending parallel to a direction in which sampled lines are stacked in the process of sampling a magnetic resonance signal, and the readout direction (Kx) may be a direction in which the sampled lines extend, while the Kz direction may be the axial direction of the coil or a first direction (Kz) perpendicular to the phase encoding direction (Ky) and the readout direction (Kx) described later.

[0057] Meanwhile, when there are a plurality of sub-sampled magnetic resonance signals, a plurality of first k-space data 220 and a plurality of second k-space data 230 can be acquired. That is, when there are a plurality of RF coils receiving magnetic resonance signals, a plurality of restored images generated corresponding to a plurality of magnetic resonance signals received for each channel become the first magnetic resonance image.

[0058] In addition to the first subsampled k-space data and the first fully sampled k-space data, the first ANN model 300 can be constructed using various additional data, such as at least one of real data, imaginary data, magnitude data, and phase data corresponding to the first k-space data 220.

[0059] According to an embodiment of the present invention, the first artificial neural network model 300 is trained in a manner described below with reference to Figures 7 to 9, and is capable of generating high-quality second k-space data 230 from low-quality first k-space data 220.

[0060] 3a and 3b, a step S210 of acquiring a first magnetic resonance image 240 from arbitrary k-space data by performing an inverse Fourier transform (IFFT) may be performed. The Grapper and / or an artificial neural network model may perform an inverse Fourier transform on image data in the complete k-space form to generate the first magnetic resonance image 240, which is a restored image.

[0061] After step S210, a step S220 of acquiring a second magnetic resonance image 250 from the first magnetic resonance image 240 by a second artificial neural network model 400 may be performed.

[0062] Referring to FIG. 6, the second artificial neural network model 400 may be a collection of algorithms that use statistical machine learning results to learn correlations between at least one sub-sampled magnetic resonance image and at least one fully sampled magnetic resonance image.

[0063] The second artificial neural network model 400 may include at least one neural network, which may include, but is not limited to, network models such as Deep Neural Network (DNN), Recurrent Neural Network (RNN), Bidirectional Recurrent Deep Neural Network (BRDNN), Multilayer Perceptron (MLP), and Convolutional Neural Network (CNN).

[0064] For example, the second artificial neural network model 400 may be a model constructed by using a neural network to learn the correlation between at least one sub-sampled magnetic resonance image and at least one fully sampled magnetic resonance image, with the pixels of at least one sampling line stacked in the phase encoding direction (Ky) as a unit.

[0065] In addition to the subsampled and full sampled magnetic resonance images, the second artificial neural network model 400 may be constructed using various additional data, such as at least one of k-space data, real image data, imaginary image data, magnitude image data, and phase image data corresponding to the first magnetic resonance image 240.

[0066] The following describes how the second artificial neural network model 400 further uses the scan parameters to perform training and reconstruction.

[0067] FIG. 7 is a schematic diagram illustrating a method for training a second artificial neural network model according to one embodiment of the present invention.

[0068] In this specification, the low quality magnetic resonance image is referred to as a training image 510, and the high quality magnetic resonance image is referred to as a label image 530. For example, the training image 510 may refer to an image acquired by subsampling a magnetic resonance signal, and the label image 530 may refer to an image acquired by full sampling of a magnetic resonance signal. For example, the number of excitations of the training image 510 may be smaller than the number of excitations of the label image 530. For example, the phase resolution of the training image 510 may be smaller than the phase resolution of the label image 530. The differences between the training image 510 and the label image 530 are various and are not limited to the above.

[0069] On the other hand, in order to obtain the training image 510 and the label image 530, a single imaging is performed, and the training image 510 and the label image 530 can be generated by performing pre-processing using the magnetic resonance signals acquired by the single imaging.

[0070] 7, the second ANN model 400 can receive not only the training image 510 and the label image 530, but also a first parameter group 520 including at least one scan parameter used in the magnetic resonance imaging of the training image 510 and a second parameter group 540 including at least one scan parameter used in the magnetic resonance imaging of the label image 530. Thus, the second ANN model 400 can learn a correlation between the low quality training image 510 and the high quality label image 530.

[0071] The scan parameters refer to environmental variables set during the process of taking a magnetic resonance image, and the acquired magnetic resonance signal and the reconstructed magnetic resonance image may change depending on the values ​​of the scan parameters. Specifically, the imaging time, resolution, noise, etc. of the magnetic resonance image can be adjusted by setting the scan parameters.

[0072] For example, the scan parameters can include, by way of example only and not limitation, number of excitation (NEX), phase resolution, and acceleration factor.

[0073] Specifically, the excitation count refers to the number of repetitions when repeatedly acquiring lines of magnetic resonance signals in k-space. The excitation count can also be called the number of averages. As the excitation count increases, the imaging time of magnetic resonance images can increase proportionally. For example, when the excitation count is 2 or more, the imaging time can be shortened by reducing the excitation count and setting it to capture magnetic resonance images.

[0074] The phase resolution means a value obtained by dividing the number of lines sampled in the phase encoding direction in the k-space by a predetermined reference value. For example, the phase resolution may have values ​​such as 0.5, 0.8, and 1.0. As the phase resolution increases, the imaging time of the magnetic resonance image may increase proportionally. Therefore, when the phase resolution is set to a lower value and the magnetic resonance image is acquired, the imaging time may be shortened.

[0075] The acceleration factor is a term used in parallel imaging techniques and refers to the number of signal lines fully sampled in k-space divided by the number of signal lines sampled by accelerated imaging. For example, an acceleration factor of 2 means that when magnetic resonance signals are sampled in the phase encoding direction to acquire lines, half the number of signal lines of the fully sampled signal lines are acquired, for example, lines are acquired at intervals of two rows. As the acceleration factor increases, the acquisition time of magnetic resonance images can be proportionally reduced.

[0076] 7, the first parameter group 520 and the second parameter group 540 each include values ​​for the number of excitations, phase resolution, and acceleration factor when acquiring the training image 510 and the label image 530, and therefore may have a size of 3. This is merely an example, and the size of the first parameter group 520 and the size of the second parameter group 540 may vary depending on the type of scan parameters included therein.

[0077] According to an embodiment of the present disclosure, a task to be performed in the process of restoring a magnetic resonance image may be determined according to scan parameter values ​​applied to magnetic resonance imaging. For example, at least one of a pre-processing algorithm and a post-processing algorithm to be applied to a training image 510 may be determined according to scan parameter values ​​included in a first parameter group 520. Also, at least one of a pre-processing algorithm and a post-processing algorithm to be applied to a label image 530 may be determined according to scan parameter values ​​included in a second parameter group 540. For example, tasks such as denoising and super resolution may be performed in the restoration process according to the number of excitations, phase resolution, and acceleration factor.

[0078] Therefore, if a restored image is generated taking into account scan parameters during the magnetic resonance image processing process, the accuracy of image restoration can be improved. Also, by adjusting the second parameter group 540 input in the learning step of the second artificial neural network model 400, it is possible to set it so that a restored image of a degree desired by the user is output.

[0079] FIG. 8 is a schematic diagram illustrating a method for inputting scan parameters into a second artificial neural network model according to one embodiment of the present invention.

[0080] 8, the first parameter group 520 is one-dimensional data in which the values ​​of the scan parameters are arranged, and the learning image corresponding to the first parameter group 520 is two-dimensional data. Therefore, since data of different dimensions cannot be concatenated and input to a single artificial neural network model, it is necessary to modify the first parameter group 520.

[0081] As an example, the first parameter group 520 may be input to a pre-processing artificial neural network model to generate first attention data 521. The first attention data 521 may indicate weights by which the scanning parameters included in the first parameter group 520 are applied to a particular layer channel of the second artificial neural network model 400. For example, the pre-processing artificial neural network model may be configured as a fully connected layer.

[0082] A portion of layers of the second artificial neural network model 400 is multiplied by the first attention data 521 along the channel axis. For example, the portion of layers may include layers located at the rear of the second artificial neural network model 400.

[0083] The pre-processing artificial neural network model is not limited to any one, but can be designed such that the length of the first attention data 521 is the same as the number of channels of the layer to which the first attention data 521 is multiplied.

[0084] 8, the preprocessing artificial neural network model is shown separate from the second artificial neural network model 400, this is for illustrative purposes only and the preprocessing artificial neural network model and the second artificial neural network model 400 may be embodied as a single artificial intelligence model, or each of the preprocessing artificial neural network model and the second artificial neural network model 400 may be embodied as multiple artificial intelligence models.

[0085] 8 has been described as an example of the first parameter group 520, the above content can be similarly applied to the second parameter group 540. That is, the second parameter group 540 can be input to a pre-processing artificial neural network model to generate second attention data, and the second attention data can be input to the second artificial neural network model 400. Also, the first parameter group 520 and the second parameter group 540 can be input together to a pre-processing artificial neural network model to output the attention data.

[0086] FIG. 9 is a flow chart showing the procedure of a magnetic resonance image processing method using scan parameters according to an embodiment of the present invention.

[0087] 9, the magnetic resonance imaging processing device may acquire a low-quality training image, a first parameter group including scan parameters used when acquiring the training image, a high-quality label image, and a second parameter group including scan parameters used when acquiring the label image (S310). As described above, the training image may have a smaller number of samples, a smaller number of excitations, or a smaller phase resolution than the label image.

[0088] The magnetic resonance imaging processing device can train an artificial neural network model using the training image, the first parameter group, the label image, and the second parameter group (S320). Specifically, the artificial neural network model can receive the training image, the first parameter group, and the second parameter group as input data and the label image as label data. For example, the artificial neural network model can correspond to the artificial neural network model 400 of Figures 3, 6, 7, and 8.

[0089] Here, the magnetic resonance imaging processing device may perform pre-processing such that the size of the first parameter group and the size of the second parameter group correspond to the number of channels of the artificial neural network model.

[0090] Specifically, the magnetic resonance imaging processing device may input the first parameter group and the second parameter group to a pre-processing artificial neural network model to generate first attention data and second attention data, respectively, and input the first attention data and the second attention data to the artificial neural network model, where the pre-processing artificial neural network may refer to the fully connected hierarchy of FIG.

[0091] The magnetic resonance imaging processing device may input a third parameter group including the input image and scan parameters used in acquiring the input image to the trained artificial neural network model to generate a reconstructed image (S330).

[0092] The above description of the present invention is for illustrative purposes only, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential features of the present invention. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and are not limiting. For example, each component described as a single type can be implemented in a distributed form, and similarly, each component described as a distributed type can be implemented in a combined form.

[0093] The scope of the present invention is determined by the claims set forth below rather than the above detailed description, and all modifications and variations that come within the meaning and scope of the claims and their equivalents should be construed as being within the scope of the present invention. [Explanation of symbols]

[0094] 100: Magnetic resonance imaging processing device 110: Communication module 120:Memory 130: Processor 140: Database 210: Magnetic resonance signal 240: First magnetic resonance image 250: Second magnetic resonance image 300: First artificial neural network model 400: Second artificial neural network model 510: Learning Image 520: First parameter group 521: First Attention Data 530: Label image 540: Second parameter group

Claims

1. 1. A method for processing magnetic resonance images, comprising: obtaining a low quality training image, a first parameter group including at least one scan parameter applied when obtaining the training image, a high quality label image, and a second parameter group including at least one scan parameter applied when obtaining the label image; training an artificial neural network model using the training image, the first parameter group, the label image, and the second parameter group; A magnetic resonance image processing method comprising:

2. 2. The magnetic resonance imaging method of claim 1, wherein the scan parameters include at least one of a number of excitations, which is a number of repetitions when repeatedly acquiring a line of first k-space data of a magnetic resonance signal, a phase resolution, which is a value obtained by dividing a number of sampled lines in a phase encoding direction of the first k-space data by a predetermined reference value, and an acceleration factor, which is a value obtained by dividing a number of fully sampled signal lines of the first k-space data by a number of sub-sampled signal lines.

3. The training image is acquired by subsampling the magnetic resonance signal; 2. The method of claim 1, wherein the label image is obtained by fully sampling the magnetic resonance signal.

4. 2. The magnetic resonance image processing method of claim 1, wherein the number of excitations of the learning image is smaller than the number of excitations of the label image.

5. 2. The magnetic resonance image processing method of claim 1, wherein the phase resolution of the training image is smaller than the phase resolution of the label image.

6. 2. The magnetic resonance image processing method of claim 1, wherein the step of training the artificial neural network model includes a step of preprocessing the first parameter group and the second parameter group so that the size of the first parameter group and the size of the second parameter group correspond to the number of channels of the artificial neural network model.

7. the pre-processing step includes inputting the first group of parameters and the second group of parameters into a pre-processing artificial neural network model to generate first attention data and second attention data, respectively; 7. The method of claim 6, wherein the step of training the artificial neural network model comprises inputting the first attention data and the second attention data to an artificial neural network model.

8. 8. The method of claim 7, wherein the pre-processing artificial neural network model includes a fully connected layer.

9. 2. The magnetic resonance image processing method of claim 1, further comprising inputting an input image and a third parameter group including at least one scan parameter applied when acquiring the input image to the artificial neural network model to generate a reconstructed image.

10. 1. A method for processing magnetic resonance images, comprising: inputting an input image and a third parameter group including at least one scan parameter applied when acquiring the input image into an artificial neural network model trained using a low quality training image, a first parameter group including at least one scan parameter applied when acquiring the training image, and a high quality label image, and a second parameter group including at least one scan parameter applied when acquiring the label image; outputting a reconstructed image from the artificial neural network model; A magnetic resonance image processing method comprising:

11. A magnetic resonance image processing apparatus for performing a magnetic resonance image processing method, comprising: a memory in which a magnetic resonance image processing program is stored; A processor for executing the program; The processor, by executing the program, 1. A magnetic resonance imaging processing apparatus, comprising: a first parameter group including a low-quality training image, at least one scan parameter applied when acquiring the training image; a high-quality label image; and a second parameter group including at least one scan parameter applied when acquiring the label image, the second parameter group being configured to train an artificial neural network model using the low-quality training image and the high-quality label image.

12. A computer readable recording medium having stored thereon a program for carrying out the method according to any one of claims 1 to 9.

Images