Training data generation method, computer program, and device

Abstract

A method performed by a computing device including at least one processor according to one embodiment of the present disclosure includes the steps of analyzing signal strength and noise characteristics of at least one input image, determining whether there are image pairs among the at least one input image whose signal strength and noise characteristics correspond, and generating training images and labeled images according to the determination results based on a predetermined noise reduction goal.

Description

[Technical Field]

[0001] The present disclosure relates to a training data generation method, a computer program, and an apparatus, and more particularly to a training data generation method, a computer program, and an apparatus using medical images. [Background technology]

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

[0003] Magnetic resonance imaging devices require a considerable amount of imaging time. Therefore, accelerated imaging technology to shorten magnetic resonance imaging time is extremely important in the medical industry. For accelerated magnetic resonance images to be used in clinical practice, they must contain all information about the subject and minimize noise that affects the interpretation of the information. In response to this need, technology has recently been developed that uses artificial intelligence to restore low-quality accelerated magnetic resonance images to a high-quality image that was not accelerated.

[0004] In order to restore accelerated magnetic resonance images (MR images) with high quality using artificial intelligence (AI), it is necessary to effectively train the AI ​​model. To achieve this, it is necessary to secure not only high-quality input data but also corresponding high-quality label data. However, even if accelerated magnetic resonance images are secured at a certain level as input data, there is a problem that corresponding high-quality restored images are rarely available in reality. Therefore, it is necessary to secure training data, which is a fundamental requirement for building an AI model. Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure is intended to solve the problems of the prior art described above, and relates to a training data generation method, computer program, and device for generating training images and labeled images according to the characteristics of an input image and the goal of noise reduction.

[0006] However, the technical objectives to be achieved by this embodiment are not limited to those described above, and other technical objectives may exist. [Means for solving the problem]

[0007] To achieve the above object, one embodiment of the present disclosure provides a method for generating training data performed by a computing device, the method including: analyzing signal strength and noise characteristics of at least one input image; determining whether there are image pairs among the at least one input image having corresponding signal strength and noise characteristics; and generating training images and labeled images according to the determination result based on a predetermined noise reduction goal.

[0008] Alternatively, the step of determining whether an image pair exists may determine a plurality of images having the same signal strength and being noise independent from each other as the image pair.

[0009] Alternatively, the step of generating the training image and the labeled image may include determining, if an image pair exists, a first image included in the image pair as the training image, and combining the first and second images according to the noise reduction goal to generate the labeled image.

[0010] Alternatively, the steps of generating the training image and the label image may generate the label image such that a first noise in the training image and a second noise in the label image are noise-dependent by the noise reduction target.

[0011] Alternatively, in the step of generating the training image and the labeled image, if the image pair does not exist, a first image among the at least one input image is determined to be the labeled image, and the training image is generated based on the first image and the noise reduction target.

[0012] Alternatively, the steps of generating the training image and the label image may be characterized in that the training image is generated such that the signal intensity of the training image and the signal intensity of the first image are identical, and the first noise of the first image and the second noise of the training image are noise-dependent according to the noise reduction goal.

[0013] Alternatively, the training images and the labeled images are input into an artificial neural network model that improves the quality of medical images.

[0014] In order to achieve the above object, one embodiment of the present disclosure provides a training data generation method performed by a computing device, the method including: setting a noise reduction target; generating training images based on a first image; and combining the first image and a second image according to the noise reduction target to generate a labeled image, wherein the first image and the second image have the same signal intensity and are noise-independent from each other.

[0015] Alternatively, the generating of the label image may be performed by generating the label image according to the following formula:

[0016]

number

[0017] In order to achieve the above object, one embodiment of the present disclosure provides a training data generation method performed by a computing device, the method including: setting a noise reduction target; generating a label image based on a first image; and generating a training image based on the first image and the noise reduction target, wherein the training image and the label image have the same signal strength and are noise-dependent to each other in accordance with the noise reduction target. Alternatively, the step of generating the training image may include the steps of: generating a second noise having the same signal strength as a first noise of the first image and being noise-independent of the first noise; and combining the first image and the second noise according to the noise reduction target to generate the training image. Alternatively, the step of generating the training image by combining the first image and the second noise according to the noise reduction target may generate the training image according to the following equation:

[0018]

number

[0019] To achieve the above object, one embodiment of the present disclosure provides a training data generation device, including: a memory configured to store a predetermined noise reduction target and at least one input image; and a processor configured to, when a second image exists among the at least one input image that has the same signal strength as a first image and is noise-independent from the first image, determine the first image as a training image and generate the labeled image by combining the first image and the second image according to the noise reduction target; and, when the second image does not exist, determine the first image as the labeled image and generate the training image based on the first image and the noise reduction target.

[0020] Alternatively, the training image and the label image may have the same signal strength and may be noise-dependent to each other in accordance with the noise reduction goal.

[0021] According to one embodiment of the present disclosure, there is provided a computer program stored on a computer-readable storage medium, which, when executed by one or more processors, performs the following operations: analyzing signal strength and noise characteristics of at least one input image; determining whether there are image pairs among the at least one input image whose signal strength and noise characteristics correspond; and generating training images and labeled images to be input to the artificial neural network model based on the determination result and a noise reduction target set for the artificial neural network model. [Effects of the Invention]

[0022] According to the above-described means for solving the problems of the present disclosure, the present disclosure can precisely adjust the degree of noise reduction according to the user's requirements, thereby preventing the loss of image information due to excessive noise reduction.

[0023] Furthermore, according to the above-mentioned solution to the problem of the present disclosure, the present disclosure discloses various methods for generating training images and label images according to input image characteristics, thereby reducing the time and cost required to provide training data. [Brief explanation of the drawings]

[0024] [Figure 1] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure. [Figure 2] FIG. 1 is a block diagram of a training data generation device according to an embodiment of the present disclosure. [Figure 3] 1 is a flowchart of a training data generation method according to an embodiment of the present disclosure. [Figure 4] 4 is a flowchart illustrating detailed operations of step S130 of FIG. 3 according to an embodiment of the present disclosure. [Figure 5] 4 is a flowchart illustrating detailed operations of step S130 of FIG. 3 according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a block diagram of a training data generation device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0025] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement the present disclosure. The embodiments presented in this disclosure are provided to enable those skilled in the art to use or practice the contents of the present disclosure. Therefore, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be embodied in various different forms and is not limited to the following embodiments.

[0026] Throughout the specification of the present disclosure, the same or similar reference numerals refer to the same or similar components. In addition, in order to clearly explain the present disclosure, reference numerals of parts that are not relevant to the explanation of the present disclosure may be omitted from the drawings.

[0027] The term "or" as used in this disclosure is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or otherwise clear from the context in this disclosure, "X uses A or B" should be understood to mean one of the natural inclusive permutations. For example, unless otherwise specified or otherwise clear from the context in this disclosure, "X uses A or B" can be interpreted as either X uses A, X uses B, or X uses both A and B.

[0028] The term "and / or" as used in this disclosure must be understood to indicate and include all possible combinations of one or more of the associated listed concepts.

[0029] The terms "comprises" and / or "comprising" as used in this disclosure should be understood to mean that the specified features and / or components are present. However, the terms "comprises" and / or "comprising" should not be understood to exclude the presence or addition of one or more other features, other components and / or combinations thereof.

[0030] In this disclosure, unless otherwise specified or clear from the context as referring to the singular form, the singular should generally be construed as including "one or more."

[0031] The term "nth (n is a natural number)" used in this disclosure can be understood as an expression used to distinguish components of the present disclosure from one another based on a predetermined criterion, such as functional, structural, or convenience of description. For example, in this disclosure, components that perform different functional roles can be classified as a first component or a second component. However, components that are substantially identical within the technical concept of the present disclosure but must be distinguished for convenience of description can also be classified as a first component or a second component.

[0032] Meanwhile, the terms "module" or "unit" used in this disclosure may be understood to refer to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a portion thereof, hardware or a portion thereof, or a combination of software and hardware. Here, a "module" or "unit" may refer to a unit composed of a single element or a unit expressed as a combination or collection of multiple elements. For example, as a concept of connotation, a "module" or "unit" may refer to a hardware element or a collection of hardware elements of a computing device, an application program that performs a specific software function, a processing procedure implemented by executing software, or a collection of instructions for executing a program. Furthermore, as a broad concept, a "module" or "unit" may refer to a computing device itself that constitutes a system, or an application executed on a computing device. However, the above concepts are merely examples, and the concepts of "module" and "unit" may be defined in various ways within the scope of understanding of those skilled in the art based on the contents of this disclosure.

[0033] The term "model" as used in this disclosure may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a collection of software units for solving a specific problem, or an abstract model of a processing process for solving a specific problem. For example, a neural network "model" may refer to a system implemented as a neural network that has problem-solving capabilities through learning. Here, a neural network may have problem-solving capabilities by optimizing parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a neural network ensemble in which multiple neural networks are combined.

[0034] As used in this disclosure, the term "image" may refer to multi-dimensional data composed of discrete image elements (e.g., pixels in two-dimensional images and voxels in three-dimensional images). For example, image may include, but is not limited to, medical images acquired by medical imaging devices such as magnetic resonance imaging, computed tomography (CT), ultrasound, or x-ray.

[0035] The term "medical picture archiving and communication system (PACS)" used in this disclosure may refer to a system that stores, processes, and transmits medical images in accordance with the digital imaging and communications in medicine (DICOM) standard. For example, a "medical picture archiving and transmission system" may work in conjunction with digital medical imaging equipment to store medical images such as magnetic resonance imaging (MRI) and computed tomography (CT) images in accordance with the medical digital imaging and communications standard. The "medical picture archiving and transmission system" may transmit medical images to terminals inside and outside the hospital via a communication network, and meta information such as reading results and medical records may be added to the medical images.

[0036] The term "object" used in this disclosure 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 substance having a volume that closely resembles the density and effective atomic number of a living organism, and may include a spherical phantom with properties similar to those of a living organism.

[0037] A magnetic resonance imaging (MRI) system is a system that obtains images of cross-sectional areas of a subject by contrasting the intensity of magnetic resonance (MR) signals in response to RF (Radio Frequency) signals generated by a magnetic field of a specific strength.

[0038] In an MRI system, a main magnet forms a static magnetic field and aligns the magnetic dipole moment of specific atomic nuclei of a subject positioned within 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 in different parts of the subject. An RF coil irradiates magnetic resonance signals according to the resonance frequency of a desired part for image acquisition. Furthermore, the RF coil receives magnetic resonance signals of different resonance frequencies radiated from various parts of the subject by forming a gradient magnetic field. The MRI system acquires images by applying an image reconstruction technique to the magnetic resonance signals received through these steps. The MRI system can also perform serial or parallel signal processing on multiple magnetic resonance signals received by a multi-channel RF coil to reconstruct the multiple magnetic resonance signals into image data.

[0039] The explanations of the above terms are intended to aid in understanding the present disclosure. Therefore, unless the above terms are explicitly stated as matters that limit the contents of the present disclosure, care should be taken not to use them in a way that limits the technical ideas of the contents of the present disclosure.

[0040] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0041] The computing device 100 according to an embodiment of the present disclosure may be a hardware device or part of a hardware device that performs comprehensive data processing and calculations, or may be a software-based computing environment connected via a communication network. For example, the computing device 100 may be a server that performs intensive data processing functions and shares resources, or a client that shares resources by interacting with the server. The computing device 100 may also be a cloud system that enables multiple servers and clients to interact with each other to comprehensively process data. The above description is merely an example of a type of computing device 100, and various types of computing device 100 may be configured within the scope that can be understood by those skilled in the art based on the contents of the present disclosure.

[0042] 1, a computing device 100 according to an embodiment of the present disclosure may include a processor 110, a memory 120, and a network unit 130. However, since FIG. 1 is merely an example, the computing device 100 may include other components for implementing a computer environment. Also, the computing device 100 may include only some of the disclosed components.

[0043] The processor 110 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for performing computing operations. For example, the processor 110 may read a computer program to perform data processing for machine learning. The processor 110 may process operations such as input data processing for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor 110 for performing such data processing may include a central processing unit (CPU), a general-purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc. The types of processor 110 described above are merely examples, and various types of processor 110 may be configured within the scope of what one skilled in the art would understand based on the present disclosure.

[0044] According to one embodiment of the present disclosure, the processor 110 may generate training data to be input to an artificial neural network model. The artificial neural network model may be trained by the training data and may output a high-quality image from a low-quality image. For example, the artificial neural network model may generate a high-quality image by removing noise from the low-quality image.

[0045] The artificial neural network model may include at least one neural network, which 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), or a convolutional neural network (CNN).

[0046] The image may be a medical image, such as a magnetic resonance image. Here, the magnetic resonance image may be acquired by accelerated imaging. Accelerated imaging may be understood as an imaging technique that shortens imaging time by increasing the acceleration factor compared to conventional imaging. The acceleration factor is a term used in parallel imaging techniques and may be understood as the value obtained by dividing the number of signal lines fully sampled in the k-space region by the number of signal lines sampled by imaging.

[0047] The processor 110 may generate training data in various ways depending on the number of input images. In this specification, training data refers to a pair of a training image and a corresponding labeled image.

[0048] The input image may be a medical image, such as a magnetic resonance image or k-space data. The training data may include a training image and a label image corresponding to the training image. The label image may have higher quality than the training image. The training image and the label image may be noise-independent by a predetermined value or noise-dependent by a predetermined value. In this case, an artificial neural network model trained using the training image and the label image may be trained to remove noise corresponding to the predetermined value.

[0049] The processor 110 can directly set the noise reduction target for the artificial neural network model or can receive a noise reduction target setting. The processor 110 can generate training data based on the noise reduction target.

[0050] The processor 110 analyzes the image characteristics of at least one input image. Specifically, the processor 110 analyzes the number, signal strength, and noise characteristics of the input images. The processor 110 determines whether there is an image pair among the at least one input image whose signal strength and noise characteristics correspond. In this specification, an image pair refers to, but is not limited to, two images.

[0051] Meanwhile, when the noise characteristics correspond but the signal intensities are different, the processor 110 may additionally perform a pre-processing process to make the signal intensities of the input images corresponding to the noise characteristics the same.

[0052] The processor 110 generates training images and label images based on the judgment results and the noise reduction goal.

[0053] In one embodiment, if an image pair exists among at least one input image, the processor 110 combines the first and second images included in the image pair to generate a training image and a labeled image. If the noise reduction goal is set to reduce noise by x times, the processor 110 determines weights corresponding to the first and second images based on x, and combines the first and second images based on the weights.

[0054] In one embodiment, when there is only one input image or no image pair exists among the input images, the processor 110 generates training images and labeled images using one image, i.e., the first image. Specifically, the processor 110 generates noise that is noise-independent from the noise of the first image, and generates training data using the first image, the generated noise, and a noise reduction target.

[0055] Alternatively, processor 110 may omit the operation of determining the existence of an image pair and perform the above operations.

[0056] According to the embodiments of the present disclosure, the processor 110 can precisely adjust the degree of noise removal according to the user's request, thereby preventing image information from being lost due to excessive noise removal. Also, various methods for generating training images and label images according to input image characteristics are disclosed, thereby reducing the time and cost required to provide training data.

[0057] According to an embodiment of the present disclosure, the processor 110 can train an artificial neural network model to improve image quality. The image may be, for example, a magnetic resonance image acquired by accelerated imaging. Accelerated imaging can be understood as an imaging technique that shortens imaging time by increasing the acceleration factor compared to conventional imaging. The acceleration factor is a term used in parallel imaging techniques and can be understood as the number of fully sampled signal lines in the K-space region divided by the number of signal lines sampled by imaging. For example, an acceleration factor of 2 can be understood as acquiring half the number of fully sampled signal lines when sampling magnetic resonance signals in the phase encoding direction to acquire lines. Therefore, as the acceleration factor increases, the imaging time of the magnetic resonance image can be proportionally reduced. In other words, by increasing the acceleration factor during magnetic resonance imaging, accelerated imaging can be implemented, with a shortened imaging time for the magnetic resonance image.

[0058] If desired, the processor 110 may generate a user interface that provides an environment for interaction with a user of the computing device 100 or a user of any client. For example, the processor 110 may generate a user interface for receiving an input image and a noise reduction goal for the artificial neural network model.

[0059] The processor 110 may generate a user interface that implements functions such as outputting, correcting, changing, or adding data based on an external input signal applied by a user. The above-described roles of the user interface are merely examples, and the roles of the user interface may be variously defined within a range that can be understood by a person skilled in the art based on the contents of this disclosure.

[0060] The memory 120 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for storing and managing data processed by the computing device 100. That is, the memory 120 may store any type of data generated or determined by the processor 110 and any type of data received by the network unit 130. For example, the memory 120 may include at least one type of storage medium selected from the group consisting of flash memory, hard disk, multimedia card micro, card-type memory, random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. The memory 120 may also include a database system that manages data in a predetermined manner. The types of memory 120 described above are merely examples, and various configurations of the memory 120 are possible within the scope of what would be understood by one skilled in the art based on the present disclosure.

[0061] The memory 120 may structure and organize and manage data, data combinations, and program code executable by the processor 110 required for the processor 110 to perform operations. For example, the memory 120 may store an input image received via the network unit 130 (described later). The memory 120 may also store training images and label images generated from the input image by the processor 110. The memory 120 may also store a noise reduction target set in an artificial neural network model. The memory 120 may also store program code that causes the processor 110 to generate training data.

[0062] The network unit 130 according to an embodiment of the present disclosure may be understood as a component that transmits and receives data via any type of known wired or wireless communication system. For example, the network unit 130 may transmit and receive data using a wired or wireless communication system such as a local area network (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 (WiFi), near field communication (NFC), or Bluetooth. The above-described communication systems are merely examples, and various wired or wireless communication systems for transmitting and receiving data by the network unit 130 may be applied in addition to the above examples.

[0063] The network unit 130 may receive data necessary for the processor 110 to perform calculations via wired or wireless communication with any system or any client. The network unit 130 may also transmit data generated by calculations by the processor 110 via wired or wireless communication with any system or any client. For example, the network unit 130 may receive medical images by communicating with a medical image storage and transmission system, a cloud server that performs tasks such as improving the resolution and standardizing medical images, or the computing device 100. The network unit 130 may transmit improved images output from the artificial intelligence model and the mathematical model by communicating with the above-mentioned system, server, computing device 100, etc.

[0064] If necessary, the network unit 130 may provide a user interface generated by the processor 110 to any system or client via wired or wireless communication with the system or client. For example, the network unit 130 may provide a user interface that visualizes data processed by the processor 110 to the system or device by communicating with a medical image storage and transmission system or a computing device 100 including a display. The network unit 130 may receive an external input signal applied by a user from the system or device via the user interface and transmit the signal to the processor 110. The processor 110 may operate to implement functions such as outputting, correcting, changing, or adding data based on the external input signal transmitted from the network unit 130. Meanwhile, even without communication via the network unit 130, a system or device that receives a user interface may operate to implement functions such as outputting, correcting, changing, or adding data based on the external input signal transmitted by the user.

[0065] FIG. 2 is a block diagram of a training data generation device according to an embodiment of the present disclosure.

[0066] 2, the training data generation device 10 receives at least one input image 210 and generates training data including a training image 310 and a label image 320. The training data generation device 10 may receive a noise reduction target set for an artificial neural network model 400. The artificial neural network model 400 is trained using the training data to improve a low-quality image to a high-quality image by the noise reduction target.

[0067] The training data generating device 10 may determine whether there is an image pair 211, 212 having corresponding signal strength and noise characteristics among at least one input image 210. Here, the at least one input image 210 may be received at different times. In addition, criteria for determining the image pair 211, 212 may be various, such as an image acquisition method, an image shooting environment, etc., in addition to the signal strength and noise characteristics.

[0068] When there is an image pair 211, 212, i.e., a first image 211 and a second image 212 whose image characteristics correspond to each other, the training data generation device 10 combines the first image 211 and the second image 212 based on the noise reduction goal to generate a training image 310 and a labeled image 320.

[0069] If the image pair 211 and 212 does not exist or if the input image 210 is a single image, the training data generation device 10 generates a training image 310 and a label image 320 using the noise reduction target and the first image 211 .

[0070] The training data generation device 10 can provide training data including the generated training images 310 and label images 320 to the artificial neural network model 400 .

[0071] 1 is merely an example, and the training data generation device 10 and the artificial neural network model 400 may be included in different systems. Here, the training data generation device 10 and the artificial neural network model 400 may transmit and receive training data via a network.

[0072] FIG. 3 is a flowchart of a training data generation method according to an embodiment of the present disclosure.

[0073] Referring to FIG. 3, the training data generating device may be an embodiment of the training data generating device 10 in FIG. 1 or the training data generating device 20 in FIG.

[0074] Prior to step S110, the training data generating device may set a noise reduction target for the artificial neural network model, for example, by receiving a noise reduction target input from a user.

[0075] The training data generating device may analyze the signal strength and noise characteristics of at least one input image (S110). For example, the training data generating device may analyze the signal strength and noise characteristics of each of the at least one input image, or may analyze whether two or more input images are noise-independent from each other.

[0076] The training data generating apparatus may determine whether there is an image pair among at least one input image having corresponding signal strength and noise characteristics (S120). The first and second images included in the image pair have the same signal strength and are noise independent.

[0077] The training data generating device can generate training images and label images according to the judgment results based on a predetermined noise reduction goal (S130).

[0078] Specifically, when an image pair exists, the training data generation device may generate training images and labeled images by combining the first and second images included in the image pair, as will be described in detail below with reference to FIG.

[0079] Alternatively, if no image pair exists, the training data generation device can generate training images and labeled images using a single image, i.e., the first image, as will be described in detail below with reference to FIG.

[0080] Here, the training data generating device can generate training images and labeled images to correspond to a noise reduction goal, and the training images and labeled images are included in the training data to be input to an artificial neural network model that improves the quality of medical images.

[0081] According to an embodiment of the present disclosure, a training data generation device can generate training data in different ways depending on the characteristics of an input image, thereby reducing the time and cost required to provide training data that meets specific conditions and thereby enabling the artificial neural network model to be trained quickly.

[0082] FIG. 4 is a flowchart illustrating detailed operations of step S130 of FIG. 3 according to an embodiment of the present disclosure.

[0083] Referring to Fig. 4, the training data generating device may be an embodiment of the training data generating device 10 of Fig. 1 or the training data generating device 20 of Fig. 6. Hereinafter, the first image and the second image may refer to images that are noise-independent from each other and have the same signal strength.

[0084] The training data generator may set a noise reduction target to be applied to the artificial neural network model (S210). Illustratively, the noise reduction target may be a real number between 0 and 1.

[0085] The training data generating apparatus may generate a training image based on the first image (S220). Exemplarily, the training data generating apparatus may determine the first image as the training image.

[0086] The training data generation device may generate a labeled image based on the first image, the second image, and a noise reduction target (S230). Specifically, the training data generation device may generate a labeled image by combining the first image and the second image based on a noise reduction target x. Here, the noise reduction target x refers to a value set to increase noise by x times or to increase the signal-to-noise ratio (SNR) by 1 / x times.

[0087] That is, the training image and the label image are generated by the following equation 3.

[0088]

number

[0089] In Equation 3, T represents the training image 310, L represents the label image 320, Input1 represents the first image 310, Input2 represents the second image 320, and x represents the noise reduction target.

[0090] Meanwhile, the order of steps S220 and S230 may be changed, and each step may be performed in parallel.

[0091] However, the method for generating the training image and the labeled image by combining the first and second images is not limited to this. That is, there may be various methods for setting the noise-independent ratio or the noise-dependent ratio between the training image and the labeled image based on the noise reduction target x, while making the signal intensities of the training image and the labeled image the same.

[0092] FIG. 5 is a flowchart illustrating detailed operations of step S130 of FIG. 3 according to an embodiment of the present disclosure.

[0093] 5, the training data generating device may be an embodiment of the training data generating device 10 in FIG. 1 or the training data generating device 20 in FIG.

[0094] The training data generating apparatus may measure noise included in the first image, i.e., the magnitude of the first noise, before or after step S310. The magnitude of the first noise may refer to the standard deviation of pixel values ​​included in the first image. For example, if the first image is a magnetic resonance image, the background of the first image may be segmented, and the magnitude of the first noise may refer to the standard deviation of pixel values ​​of the segmented background. For example, the first image may be an image in the k-space domain, i.e., k-space data. Here, the magnitude of the first noise refers to the standard deviation of pixel values ​​at a certain distance from the center of the first image.

[0095] The training data generation apparatus may generate a label image based on the first image (S320). Exemplarily, the training data generation apparatus may determine the first image as the label image.

[0096] The training data generating device may generate a training image based on the first image and a noise reduction target (S330). Specifically, the training data generating device may generate a second noise having the same magnitude as the first noise and being noise-independent of the first noise. The second noise may be random noise, which may refer to noise according to a complex Gaussian distribution, a Rician distribution, or a noncentral chi distribution. Specifically, if the first image is k-space data, the second noise according to a complex Gaussian distribution is generated. If the first image is a magnetic resonance image, the second noise according to a Rician distribution or a noncentral chi distribution is generated.

[0097] The training data generating device can generate training images such that the training images and the label images are noise-dependent with respect to the first image, the second noise, and the noise reduction target. That is, the training images and the label images are generated according to the following equation (4):

[0098]

number

[0099] In Equation 4, T is the training image 310, L is the label image 320, Input1 is the first image 310, x is the noise reduction target, and n2 is the second noise.

[0100] Meanwhile, the order of steps S320 and S330 may be changed, and the steps may be performed in parallel.

[0101] Meanwhile, additional pre-processing operations can be performed on the first image, such as transforming the domain of the first image.

[0102] The training data generator can then provide the training images and the labeled images to an artificial neural network model, which can be trained to output high-quality medical images based on low-quality medical images using the training images and the labeled images.

[0103] FIG. 6 is a block diagram of a training data generation device according to an embodiment of the present disclosure.

[0104] 6, the training data generating device 20 is similar to the training data generating device 10 of FIG. 2, and therefore common features will be omitted. The training image 310 and the label image 320 may be magnetic resonance images in the image domain. At least one input image 510 may be an image in the k-space domain, i.e., k-space data.

[0105] The training data generator 20 can generate a k-space training image 610 and a k-space label image 620 that have the same signal strength and are noise-dependent with each other only for the noise reduction purpose.

[0106] The training data generation device 20 can determine whether an image pair 511, 512 exists among at least one input image 500.

[0107] When a first image 511 and a second image 512 included in an image pair 511, 512 exist, the training data generation device 20 combines the first image 511 and the second image 512 based on a noise reduction target to generate a k-space training image 610 and a k-space label image 620. The specific details are the same as those of the method described above with reference to FIG. 4, and therefore will not be described below.

[0108] If there is no image pair 510 or the input image is a single first image 511, the training data generation device 10 generates a k-space training image 610 and a k-space label image 620 using the first image 511 according to the noise reduction goal. The specific details are the same as those of the method described above with reference to FIG. 5, and therefore will not be described again.

[0109] On the other hand, the training data generation device 20 can omit the step of determining whether the image pair 511, 512 exists and generate the k-space training image 610 and the k-space label image 620 using two images, or generate the k-space training image 610 and the k-space label image 620 using one image.

[0110] The training data generator 20 can Fourier transform the k-space training image 610 and the k-space label image 620 to generate the training image 310 and the label image 320, respectively, and provide them to the artificial neural network model 400.

[0111] The above description of the present disclosure is for illustrative purposes only, and those skilled in the art will understand that the present disclosure can be easily modified into other specific forms without changing the technical spirit or essential features of the present disclosure. Therefore, it should be understood that the above-described embodiments of the present disclosure 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.

[0112] The scope of the present disclosure is determined by the claims set forth below rather than the above detailed description, and all modifications and variations derived from the meaning, scope, and equivalent concepts of the claims should be construed as being within the scope of the present disclosure.

Claims

1. 1. A method for generating training data, performed by a computing device including at least one processor, comprising: analyzing signal strength and noise characteristics of at least one input image; determining whether there is an image pair among the at least one input image that has corresponding signal strength and noise characteristics; generating training images and label images according to a judgment result based on a predetermined noise reduction goal; A method comprising:

2. The step of determining whether an image pair exists includes:

2. The method of claim 1, wherein the image pair is determined to be a plurality of images having the same signal strength and being noise-independent from one another.

3. The step of generating the training image and the labeled image includes: If the image pair exists, a first image included in the image pair is determined to be the training image; 3. The method of claim 2, further comprising combining the first and second images according to the noise reduction objective to generate the label image.

4. 4. The method of claim 3, wherein the generating the training image and the label image comprises generating the label image such that a first noise in the training image and a second noise in the label image are noise-dependent by the noise reduction target.

5. The step of generating the training image and the labeled image includes: If the image pair does not exist, determining a first image of the at least one input image as the label image; 3. The method of claim 2, further comprising generating the training images based on the first image and the noise reduction target.

6. The step of generating the training image and the labeled image includes: The signal strength of the training image and the signal strength of the first image are the same; 6. The method of claim 5, wherein the training images are generated such that a first noise in the first image and a second noise in the training images are noise-dependent according to the noise reduction goal.

7. 3. The method of claim 2, wherein the training images and the labeled images are input to an artificial neural network model that improves the quality of medical images.

8. 1. A method for generating training data, performed by a computing device including at least one processor, comprising: setting a noise reduction target; generating a training image based on a first image, and combining the first image and the second image according to the noise reduction goal to generate a labeled image; Including, The method of claim 1, wherein the first image and the second image have identical signal strengths and are noise independent of each other.

9. 9. The method of claim 8, wherein the step of generating the label image generates the label image according to the following formula: [Equation 1] L = the label image x = the noise reduction target Input １ = The first image Input ２ = The second image

10. 1. A method for generating training data, performed by a computing device including at least one processor, comprising: setting a noise reduction target; generating a label image based on a first image, and generating a training image based on the first image and the noise reduction target; Including, The method, wherein the training image and the label image have the same signal strength and are noise-dependent to each other to correspond to the noise reduction goal.

11. The step of generating training images includes: generating a second noise having the same signal strength as the first noise of the first image and being noise-independent from the first noise; generating the training image by combining the first image and the second noise according to the noise reduction goal; The method of claim 10, comprising:

12. 12. The method of claim 11, wherein generating the training image by combining the first image and the second noise according to the noise reduction goal generates the training image according to the following equation: [Equation 2] T = learning image Input １ = The first image n ２ = the second noise x = the noise reduction target

13. A training data generation device, a memory for storing a predetermined noise reduction target and at least one input image; a processor for determining the first image as a learning image when a second image exists among the at least one input image, the second image having the same signal intensity as a first image and being noise-independent from the first image, and combining the first image and the second image according to the noise reduction target to generate a labeled image; and for determining the first image as the labeled image when the second image does not exist, and generating the learning image based on the first image and the noise reduction target; An apparatus comprising:

14. The apparatus of claim 13, wherein the training image and the label image have the same signal strength and are noise-dependent to each other to correspond to the noise reduction goal.

15. A computer program stored in a computer-readable storage medium, the computer program performing the following operations when executed by one or more processors: The operation is analyzing the signal strength and noise characteristics of at least one input image; determining whether there are image pairs among the at least one input image that have corresponding signal strengths and noise characteristics; generating training images and label images to be input to the artificial neural network model based on the judgment result and a noise reduction target set for the artificial neural network model; A computer program comprising:

Images