KR102215902B1 - Method of correcting an image and apparatuses performing the same - Google Patents

Abstract

Disclosed are an image correction method and devices performing the same. According to one embodiment of the present invention, the image correction method comprises the steps of: generating non-resonant frequency maps for distorted images based on the distorted images obtained from an object; and generating a corrected image in which distortion included in the distorted images is corrected based on the distorted images and the non-resonance frequency maps.

Description

Image correction method and devices that perform the same TECHNICAL FIELD [METHOD OF CORRECTING AN IMAGE AND APPARATUSES PERFORMING THE SAME}

The following embodiments relate to an image correction method and apparatuses for performing the same.

The magnetic resonance imaging apparatus may acquire various information such as an anatomical tomography image of a person, an anatomical structure, and a physiological function. For example, the magnetic resonance imaging apparatus may capture a magnetic resonance image of a person.

The magnetic resonance imaging apparatus can be widely used to diagnose lesions in various parts of the brain, musculoskeletal system, and organ organs.

However, in the MR image captured by the MR imaging apparatus, distortion (or artifacts) may occur (or be included) in the image due to off-resonance frequencies.

The non-resonant frequency may be caused by field inhomogeneities (or magnet inhomogeneity), a difference in susceptibility between tissues, and the presence or absence of foreign substances having a large susceptibility value such as a metal (or metal component).

In the MR image of a metal component, the position information of the MR image is distorted, and distortion (or artifacts) may occur (or include) in the final acquired image in the form of void or pile-up of a signal, or a change in the size of a voxel. .

Distortion included in the magnetic resonance image not only interferes with accurate diagnosis, but also may cause difficulty in follow-up of a surgical procedure.

Embodiments may provide a technique for generating a single corrected image in which distortion included in a magnetic resonance image pair is corrected by using a magnetic resonance image pair including distortion due to a non-resonant frequency.

An image correction method according to an embodiment includes generating non-resonant frequency maps for the distorted images based on distorted images acquired from an object, and the distortion based on the distorted images and the non-resonant frequency maps. And generating a corrected image in which distortion included in the images is corrected.

The distortion images may be pairs of magnetic resonance images including distortion due to non-resonant frequencies.

The MR image pair may include a first MR image and a second MR image obtained from the object by magnetic fields having different polarities.

The generating of the non-resonant frequency maps may include generating the non-resonant frequency maps representing distortion included in the distorted images using an artificial neural network.

The non-resonant frequency maps may represent distortion included in the distorted images based on any one of the distorted images.

The non-resonant frequency maps include a first non-resonant frequency map and a second non-resonant frequency map, and the first non-resonant frequency map is included in the second MR image based on the first MR image. It represents distortion, and the second non-resonance frequency map may represent distortion included in the first MR image based on the second MR image.

The generating of the correction image includes generating a first correction image based on the first MR image and the first non-resonance frequency map, and the second MR image and the second non-resonance frequency map. It may include generating a second corrected image based on the first corrected image and generating the corrected image based on the first and second corrected images.

Generating the first corrected image may include generating the first corrected image based on a half value of the first non-resonant frequency map.

Generating the second correction image may include generating the second correction image based on a half value of the second non-resonance frequency map.

Generating the correction image based on the first correction image and the second correction image may include generating the correction image by averaging the first correction image and the second correction image.

An image correction apparatus according to an embodiment generates non-resonant frequency maps for the distorted images based on a communication module and distorted images obtained from an object through the communication module, and the distorted images and the non-resonant And a processor that generates a corrected image in which distortion included in the distortion images is corrected based on frequency maps.

The distortion images may be pairs of magnetic resonance images including distortion due to non-resonant frequencies.

The MR image pair may include a first MR image and a second MR image obtained from the object by magnetic fields having different polarities.

The processor may include a first generator that generates the non-resonant frequency maps representing distortions included in the distorted images using an artificial neural network.

The non-resonant frequency maps may represent distortion included in the distorted images based on any one of the distorted images.

The non-resonant frequency maps include a first non-resonant frequency map and a second non-resonant frequency map, and the first non-resonant frequency map is included in the second MR image based on the first MR image. It represents distortion, and the second non-resonance frequency map may represent distortion included in the first MR image based on the second MR image.

The processor generates a first correction image based on the first MR image and the first non-resonance frequency map, and generates a second correction image based on the second MR image and the second non-resonance frequency map. And a second generator configured to generate the corrected image based on the first corrected image and the second corrected image.

The second generator may generate the first correction image based on a half value of the first non-resonant frequency map.

The second generator may generate the second correction image based on a half value of the second non-resonant frequency map.

The second generator may generate the correction image by averaging the first correction image and the second correction image.

1 is a diagram illustrating an image correction system according to an exemplary embodiment.
FIG. 2 shows a schematic block diagram of the image correction apparatus shown in FIG. 1.
FIG. 3 shows a schematic block diagram of the processor shown in FIG. 2.
4 shows an example for describing an image correction operation according to an embodiment.
5 shows an example for explaining an unsupervised learning operation and an image correction operation according to an embodiment.
6 shows an example for describing an artificial neural network according to an embodiment.
7 shows an example for explaining a correction result through an artificial neural network learned by various learning methods according to an embodiment.
8 shows another example for explaining a correction result through an artificial neural network learned by various learning methods according to an embodiment.
9 is a flowchart illustrating an operation of the image correction apparatus shown in FIG. 1.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the rights of the patent application is not limited or limited by these embodiments. It should be understood that all changes, equivalents, or substitutes to the embodiments are included in the scope of the rights.

The terms used in the examples are used for illustrative purposes only and should not be interpreted as limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the presence or addition of elements or numbers, steps, actions, components, parts, or combinations thereof, does not preclude in advance.

Terms such as first or second may be used to describe various components, but the components should not be limited by terms. The terms are only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the embodiment, the first component may be named as the second component, and similarly The second component may also be referred to as a first component.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this application. Does not.

In addition, in the description with reference to the accompanying drawings, the same reference numerals are assigned to the same components regardless of the reference numerals, and redundant descriptions thereof will be omitted. In describing the embodiments, when it is determined that a detailed description of related known technologies may unnecessarily obscure the subject matter of the embodiments, the detailed description thereof will be omitted.

1 is a diagram illustrating an image correction system according to an exemplary embodiment.

The image correction system 10 includes an image generating device 100 and an image correcting device 300.

The image generating apparatus 100 may generate a magnetic resonance image by performing magnetic resonance imaging (MRI) of an object. For example, the image generating device 100 may be an MRI device (or an MRI imaging device). In this case, the object may be various objects such as a person, an animal, and an object that generate (or cause) non-resonant frequencies. The magnetic resonance image may be a pair of magnetic resonance images including distortion due to non-resonant frequencies.

The image generating apparatus 100 may acquire (or generate) a magnetic resonance image pair including distortion by photographing an object (or subject) including a metal component through a magnetic field (or a gradient magnetic field) using the principle of nuclear magnetic resonance. have.

First, the image generating apparatus 100 may encode location information of an object through a change in a magnetic field. The positional encoding for the object may be a phase encoding and a frequency encoding for the object. For example, the image generating apparatus 100 may encode the position of the object by converting the polarity of the gradient magnetic field into a positive value (or a positive polarity) and a negative value (or a negative polarity).

Thereafter, the image generating apparatus 100 may acquire a MR image pair including distortion based on the encoding result. The magnetic resonance image pair may be distortion images including distortion generated in at least one of a frequency encoding direction and a phase encoding direction.

The magnetic resonance image pair may include a first magnetic resonance image and a second magnetic resonance image. For example, the first magnetic resonance image may be a distortion image (or a magnetic resonance image including distortion) based on a positive value of a gradient magnetic field. The second magnetic resonance image may be a distortion image due to a negative value of the gradient magnetic field.

The first magnetic resonance image can be represented by Equation 1. The second magnetic resonance image can be represented by Equation 2. In this case, Equations 1 and 2 may include distortion occurring only in the frequency encoding direction.

는 제1 자기 공명 영상이고,

는 왜곡이 없는 자기 공영 영상이고, x는 제1 자기 공명 영상에 대한 주파수 인코딩 방향으로의 좌표값이고, y는 제1 자기 공명 영상에 대한 위상 인코딩 방향으로의 좌표값이고, BW는 제1 자기 공명 영상에 대한 위치 인코딩의 대역폭이고, f는 BW에 따라 해당 픽셀에 대한 비공명 주파수일 수 있다.here,

Is the first magnetic resonance image,

Is a distortion-free magnetic resonance image, x is a coordinate value in the frequency encoding direction of the first magnetic resonance image, y is a coordinate value in the phase encoding direction of the first magnetic resonance image, and BW is the first magnetic resonance image. It is a bandwidth of position encoding for a resonance image, and f may be a non-resonant frequency for a corresponding pixel according to the BW.

는 제2 자기 공명 영상이고,

는 왜곡이 없는 자기 공영 영상이고, x는 제2 자기 공명 영상에 대한 주파수 인코딩 방향으로의 좌표값이고, y는 제2 자기 공명 영상에 대한 위상 인코딩 방향으로의 좌표값이고, BW는 제2 자기 공명 영상에 대한 위치 인코딩의 대역폭이고, f는 BW에 따라 해당 픽셀에 대한 비공명 주파수일 수 있다.here,

Is the second magnetic resonance image,

Is a distortion-free magnetic resonance image, x is a coordinate value in the frequency encoding direction of the second magnetic resonance image, y is a coordinate value in the phase encoding direction of the second magnetic resonance image, and BW is the second magnetic resonance image. It is a bandwidth of position encoding for a resonance image, and f may be a non-resonant frequency for a corresponding pixel according to the BW.

The image generating apparatus 100 may transmit (or provide) the first and second magnetic resonance images, which are magnetic resonance image pairs of the object, to the image correction apparatus 300.

As described above, although the image generating device 100 is an MRI device, it is not limited thereto. For example, the image generating device 100 may be an electronic device that has magnetic resonance image pairs for various objects that generate non-resonant frequencies. The electronic device may include a database storing magnetic resonance image pairs.

For example, the electronic device may be various devices such as a personal computer (PC), a data server, or a portable electronic device. Portable electronic devices include a laptop computer, a mobile phone, a smart phone, a tablet PC, a mobile internet device (MID), a personal digital assistant (PDA), an enterprise digital assistant (EDA). ), digital still camera, digital video camera, portable multimedia player (PMP), personal navigation device or portable navigation device (PND), handheld game console, e-book (e-book), can be implemented as a smart device (smart device). In this case, the smart device may be implemented as a smart watch or a smart band.

The image correction apparatus 300 may generate a correction image in which distortion included in the magnetic resonance image pair is corrected by using a magnetic resonance image pair including distortion due to a non-resonant frequency.

Accordingly, the image correction apparatus 300 may improve image quality by generating an image whose distortion is corrected without label data using only an image containing distortion in an environment in which label data, which is an image without distortion, is generated.

In addition, the image correction apparatus 300 may be extendedly applicable to a distortion correction (or distortion removal) method for correcting and removing distortion caused by various non-resonant frequencies such as incomplete shimming, chemical shift, and distortion in the one-shot imaging method. have.

FIG. 2 is a schematic block diagram of the image correction apparatus illustrated in FIG. 1, FIG. 3 is a schematic block diagram of the processor illustrated in FIG. 2, and FIG. 4 illustrates an image correction operation according to an exemplary embodiment. Here is an example for this.

The image correction apparatus 300 includes a communication module 310, a memory 330, and a processor 350.

The communication module 310 may transmit the magnetic resonance image pair transmitted from the image generating device 100 to the controller 350.

The memory 330 may store instructions (or programs) executable by the processor 350. For example, the instructions may include instructions for executing an operation of the processor 350 and/or an operation of each component of the processor 350.

The processor 350 may process data stored in the memory 330. The processor 350 may execute computer-readable code (eg, software) stored in the memory 330 and instructions induced by the processor 350.

The processor 350 may be a data processing device implemented in hardware having a circuit having a physical structure for executing desired operations. For example, desired operations may include code or instructions included in a program.

For example, a data processing device implemented in hardware is a microprocessor, a central processing unit, a processor core, a multi-core processor, and a multiprocessor. , Application-Specific Integrated Circuit (ASIC), and Field Programmable Gate Array (FPGA).

The processor 350 may control the overall operation of the image correction apparatus 300. For example, the processor 350 may control the operation of each component 310 and 330 of the image correction device 300.

The processor 350 generates non-resonant frequency maps for the distorted images based on the distorted images, and generates a corrected image in which distortion included in the distorted images is corrected based on the distorted images and the non-resonant frequency maps. I can. Non-resonance frequency maps are parameters representing distortion included in the distorted images based on any one of the distorted images, and the distorted images may include a first MR image and a second MR image.

The processor 350 includes a first generator 351 and a second generator 353. The first generator 351 includes an artificial neural network 370. The artificial neural network 370 may be learned from the processor 350 through an unsupervised learning method. The unsupervised learning method will be described in detail in FIG. 5.

The first generator 351 may generate non-resonant frequency maps representing distortions included in the first and second MR images using the artificial neural network 370.

For example, the first generator 351 may input a first magnetic resonance image and a second magnetic resonance image to the artificial neural network 370. Thereafter, the first generator 351 may generate (or estimate) non-resonant frequency maps using the artificial neural network 370 to which the first and second MR images are input. Non-resonant frequency maps may be output from the artificial neural network 370.

The non-resonant frequency maps may include a first non-resonant frequency map and a second non-resonant frequency map. For example, the first non-resonant frequency map may represent distortion included in the first MR image and distortion included in the second MR image based on the first MR image. The second non-resonant frequency map may represent distortion included in the first MR image and distortion included in the second MR image based on the second MR image.

에 있는 한 pixel을 기준으로

에는

에 상응하는(또는 대응하는) pixel이 어디에 존재하는 지를 나타내는 값일 수 있다.The non-resonant frequency map may be a registration parameter in which a degree of distortion of another image from a reference image is expressed in pixels. For example, the non-resonant frequency map is a non-linear translation (or distorted shape) due to metal distortion.

Based on one pixel in

In

It may be a value indicating where the pixel corresponding to (or corresponding to) exists.

The first non-resonant frequency map can be represented by Equation 3. The second non-resonant frequency map can be represented by Equation 4.

는 제1 비공명 주파수 맵이고, G는 자기 공명 영상 생성의 함수를 나타낸다.here,

Is a first non-resonant frequency map, and G is a function of generating an MR image.

는 제2 비공명 주파수 맵을 나타낸다.here,

Represents a second non-resonant frequency map.

또는

이라는 의미일 수 있다.The mathematical meaning of arg min in Equations 3 and 4 is that the value of f that minimizes the following equation is

or

It can mean that.

The first generator 351 may transmit the first non-resonant frequency map and the second non-resonant frequency map to the second generator 353.

The second generator 353 may generate a corrected image based on the distortion images and non-resonant frequency maps.

For example, the second generator 353 may generate a first correction image based on the first MR image and the first non-resonance frequency map. In this case, the second generator 353 may generate the first correction image based on a half value of the first non-resonant frequency map. The first correction image may be a correction image (or a corrected magnetic resonance image) in which distortion included in the first magnetic resonance image is corrected.

The second generator 353 may generate a second correction image based on the second MR image and the second non-resonant frequency map. In this case, the second generator 353 may generate a second correction image based on a half value of the second non-resonant frequency map. The second correction image may be a correction image (or a corrected MR image) in which distortion included in the second MR image is corrected.

The second generator 353 may generate a correction image based on the first correction image and the second correction image. For example, the second generator 353 may generate a final corrected image, which is a magnetic resonance image without distortion, by averaging the first and second corrected images. The final corrected image may be a MR image in which distortion included in the first MR image and distortion included in the second MR image are corrected.

The final corrected image can be represented by Equation 5.

는 최종 보정 영상을 나타낸다.here,

Represents the final corrected image.

5 shows an example for explaining an unsupervised learning operation and an image correction operation according to an embodiment.

Existing methods for learning the artificial neural network 370 include a modeling-based learning method and a supervised learning-based learning method.

The modeling-based learning method, which is an existing method, is a learning method using prior knowledge, and may be methods prior to deep learning using learning data. For example, the modeling-based learning method may be an analytic method and a numerical method through an iterative operation. Modeling-based learning methods have numerous methods depending on each application field, and distortion of an image can be removed through modeling using prior knowledge.

However, the modeling-based learning method has disadvantages such as an error due to abstraction of prior knowledge, an increase in recovery time and limited performance occurring in repetitive operations.

A learning method based on supervised learning, which is an existing method, is a learning method using learning data and may be methods after deep learning. After the success of deep learning, the supervised learning-based learning method showed remarkable performance in various application fields through supervised learning using a lot of data. A learning method based on supervised learning was used to remove distortion caused by various causes in medical images. For example, a learning method based on supervised learning may remove distortion included in an image by learning a correlation between an image with distortion and an image without distortion.

The learning method based on supervised learning showed better performance than the existing modeling-based methods, and it takes a long learning time, but when applied in practice, distortion can be removed with a short restoration time.

However, the learning method based on supervised learning has a disadvantage in that a pair of a distortion-free image and an image containing distortion, which are training data, are required to learn the artificial neural network 370.

In addition to the above-described disadvantages, the existing learning method has difficulty in acquiring training data for training the artificial neural network 370, and in particular, it is difficult to collect patient data from medical images. Also, the existing learning method must use a method such as 3D-PE to obtain the ground truth to be used as a label. In the case of using a method such as 3D-PE, there is a disadvantage of requiring too long an image acquisition time.

The unsupervised learning method is a learning that can obtain an output image, an image whose distortion is corrected, from an input image pair, which is a magnetic resonance image pair including distortion by a non-resonant magnetic field (f), in order to overcome the disadvantages of the existing learning method described above. It could be the way.

)으로부터 제2 자기 공명 영상과 유사한 영상(

)을 생성하고, 제2 자기 공명 영상(

)으로부터 제1 자기 공명 영상과 유사한 영상(

)을 생성하게 하는 비공명 주파수 맵들을 출력할 수 있다. 이때, 인공 신경망(370)은 손실 함수(

)를 반영할 수 있다.The artificial neural network 370 may be trained to output the first and second magnetic resonance images, which are input image pairs, as opposite images, respectively. For example, in the artificial neural network 370, the second generator 353 is a first magnetic resonance image (

) From an image similar to the second magnetic resonance image (

), and a second magnetic resonance image (

An image similar to the first magnetic resonance image from (

) Can be output to generate non-resonant frequency maps. At this time, the artificial neural network 370 is a loss function (

) Can be reflected.

The second generator 353 generates an output image pair different from the input image by using the total value of the input image pair and the non-resonant frequency maps, or included in the input image by using half values of the input image pair and the non-resonant frequency maps. A correction image in which distortion is corrected may be generated.

)을 왜곡시켜 입력 영상과 상이한 출력 영상(

)을 생성할 수 있다.For example, the second generator 353 uses the entire value of the random non-resonant frequency map f to use an arbitrary input image (

) To distort the output image different from the input image (

) Can be created.

The output image can be represented by Equation 6.

Here, K denotes a parameter reflecting information of frequency and phase encoding.

Equation 6 may be an equation composed of a Fourier transform and an inverse Fourier transform when a non-resonant frequency occurs based on the principle of generating a magnetic resonance image. Equation 6 may be an equation expressed only in the frequency encoding direction (x) for simplification of the equation.

6 shows an example for describing an artificial neural network according to an embodiment.

The artificial neural network 370 may be a variety of convolution-based artificial neural networks (or artificial neural network models) in order to use the regional association of non-resonant frequencies.

In the case of a severe change in non-resonant frequency such as metal, the artificial neural network 370 uses a U-net structure for more efficient learning, so that the receptive field of the artificial neural network 370 is large as shown in FIG. Can be configured.

For stabilization of learning, regularization using prior knowledge such as "non-resonant frequency map is smooth" can be promoted.

As shown in FIG. 6, blocks of the artificial neural network 370 may be feature maps (matrix) after passing through each layer. The arrows indicate each layer, and the role of each layer may be different depending on the color indicated at the bottom right. The red line may be a layer that connects the convolution layer-batch normalization-relu function into one.

7 shows an example for explaining a correction result through an artificial neural network learned by various learning methods according to an embodiment.

The corrected image through supervised learning and the corrected image through unsupervised learning may be a magnetic resonance image from which non-resonant frequency artifacts caused by metal are removed. At this time, the synthetic data for learning may be synthetic data through computer simulation for learning.

MAVRIC, the existing method of removing metal artifacts, has artifacts remaining due to non-resonant frequencies as shown by the red arrow.

In the unsupervised learning method, the remaining artifacts can be effectively removed.

When comparing the unsupervised learning method and the supervised learning method, when there is no noise, the performance of the unsupervised learning method may be slightly degraded. However, the unsupervised learning method shows better performance than the supervised learning method as the noise increases.

That is, the unsupervised learning method shows better performance than other learning methods in unseen data.

8 shows another example for explaining a correction result through an artificial neural network learned by various learning methods according to an embodiment.

The corrected images of FIG. 8 may actually be corrected images for an artifact acquired through a magnetic resonance imaging apparatus. In this case, the corrected images may be corrected images acquired using the artificial neural network 370 learned through MAVRIC, model-based, supervised learning, and unsupervised learning.

The corrected image was acquired through an experiment based on fast spin echo in a 3T system. The parameters for the MR image may be matrix size = 160 (readout)x512 (phase encoding)x20 (slice), resolution = 1x1x5mm^3, readout bandwidth = 390hz/px, Gaussian RF with 2.25kHz full-width0half-maximum. .

MAVRIC, the existing method of removing metal artifacts, has artifacts that remain large in areas where the non-resonant frequency is large at the point of the red arrow.

The model-based method has some ripple left at the point of the sky blue arrow.

The unsupervised method removes a lot of remaining artifacts, but the supervised learning method removes the artifacts, but additional artifacts were often caused by overfitting at the point of the yellow arrow.

9 is a flowchart illustrating an operation of the image correction apparatus shown in FIG. 1.

The first generator 351 may input the distortion image pair to the artificial neural network 370 (1010) to generate non-resonant frequency maps representing the distortion included in the distortion image pair (1030).

The second generator 353 may generate corrected images based on the distortion image pair and half values of the non-resonant frequency maps (1050).

The second generator 353 may generate a final corrected image by averaging the corrected images (operation 1070).

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

As described above, although the embodiments have been described by the limited drawings, a person of ordinary skill in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments and claims and equivalents fall within the scope of the following claims.

Claims (20)

Generating non-resonant frequency maps for the distorted images based on the distorted images obtained from the object; And
Generating a corrected image in which distortion included in the distortion images is corrected based on the distortion images and the non-resonant frequency maps
Image correction method comprising a.
The method of claim 1,
The image correction method wherein the distortion images are magnetic resonance image pairs including distortion due to a non-resonant frequency.
The method of claim 2,
The MR image pair includes a first MR image and a second MR image obtained from the object by magnetic fields having different polarities.
The method of claim 1,
Generating the non-resonant frequency maps,
Generating the non-resonant frequency maps representing distortions included in the distorted images using an artificial neural network
Image correction method comprising a.
The method of claim 4,
The non-resonant frequency maps indicate distortion included in the distorted images based on any one of the distorted images.
The method of claim 3,
The non-resonant frequency maps,
A first non-resonant frequency map; And
Second non-resonant frequency map
Including,
The first non-resonant frequency map represents distortion included in the second MR image based on the first MR image,
The second non-resonant frequency map is an image correction method indicating distortion included in the first MR image based on the second MR image.
The method of claim 6,
The step of generating the correction image,
Generating a first correction image based on the first magnetic resonance image and the first non-resonant frequency map;
Generating a second correction image based on the second magnetic resonance image and the second non-resonant frequency map; And
Generating the corrected image based on the first corrected image and the second corrected image
Image correction method comprising a.
The method of claim 7,
Generating the first correction image,
Generating the first correction image based on the half value of the first non-resonant frequency map
Image correction method comprising a.
The method of claim 7,
Generating the second correction image,
Generating the second correction image based on the half value of the second non-resonant frequency map
Image correction method comprising a.
The method of claim 9,
Generating the correction image based on the first correction image and the second correction image,
Generating the corrected image by averaging the first corrected image and the second corrected image
Image correction method comprising a.
Communication module; And
Generate non-resonant frequency maps for the distorted images based on distorted images obtained from an object through the communication module, and distortion included in the distorted images based on the distorted images and the non-resonant frequency maps A processor that generates this corrected correction image
Image correction device comprising a.
The method of claim 11,
The distorted images are magnetic resonance image pairs including distortion due to a non-resonant frequency.
The method of claim 12,
The magnetic resonance image pair includes a first magnetic resonance image and a second magnetic resonance image obtained from the object by magnetic fields having different polarities.
The method of claim 11,
The processor,
A first generator that generates the non-resonant frequency maps representing distortions included in the distorted images using an artificial neural network
Image correction device comprising a.
The method of claim 14,
The non-resonant frequency maps display distortion included in the distorted images based on any one of the distorted images.
The method of claim 13,
The non-resonant frequency maps,
A first non-resonant frequency map; And
Second non-resonant frequency map
Including,
The first non-resonant frequency map represents distortion included in the second MR image based on the first MR image,
The second non-resonant frequency map is an image correction apparatus indicating distortion included in the first MR image based on the second MR image.
The method of claim 16,
The processor,
Generating a first corrected image based on the first MR image and the first non-resonant frequency map, and generating a second corrected image based on the second MR image and the second non-resonant frequency map, A second generator that generates the corrected image based on the first corrected image and the second corrected image
Image correction device comprising a.
The method of claim 17,
The second generator,
An image correction device that generates the first correction image based on a half value of the first non-resonance frequency map.
The method of claim 17,
The second generator,
An image correction apparatus for generating the second correction image based on a half value of the second non-resonance frequency map.
The method of claim 19,
The second generator,
An image correction apparatus for generating the correction image by averaging the first correction image and the second correction image.

Images