Method and device for providing trained model for denoising

Abstract

Disclosed, according to one embodiment, are a method and a device for providing a trained model for denoising, comprising the steps in which: a receiving unit obtains a low-dose image and a high-dose image of an object; a processor obtains low-dose continuous segmented images by segmenting the low-dose image in a manner enabling continuity; the processor obtains global information indicating image characteristics for the entirety of the object by applying the low-dose continuous segmented images to a trained model; the processor obtains low-dose random segmented images by randomly segmenting the low-dose image; the processor obtains local information indicating image characteristics for each region of the object by applying the low-dose random segmented images to the trained model; and an updated trained model is obtained by updating the trained model by using the global information and the local information.

Description

Method and device for providing a learning model for denoising

[0001] The present invention relates to a method and device for providing a learning model for denoising updated based on information obtained by applying an image included in a low-dose image of a subject to a learning model. More specifically, the present invention relates to a technical field capable of efficiently removing noise from a low-dose image and improving the performance of the learning model by dividing and randomly dividing a low-dose image into successive segments, applying each segment to a learning model, and then providing an updated learning model by reflecting the information obtained therefrom.

[0002]

[0003] Traditionally, filtering-based techniques such as Gaussian, Anisotropic Diffusion, and Bilateral have been used to remove noise from low-dose images. However, existing denoising techniques have had difficulties in preserving the edges of the target object, and denoising effects have been particularly poor for noise with nonlinear characteristics. For example, ultra-low-dose CBCT images are already blurred due to severe scatter noise, so even when existing denoising techniques are used, it is inconvenient to perform efficient denoising while maintaining the shape of the target object.

[0004] Due to this, various learning models and algorithms have recently been proposed to improve low-dose image quality. However, efficient resource management for denoising low-dose images has not been considered, resulting in unnecessary resource usage and problems in optimizing learning models.

[0005] Therefore, a technology is required that can improve the performance of a learning model for denoising by efficiently using resources by dividing and randomly dividing images included in low-dose images so that they can be continuous, applying each to a learning model, and improving the learning model for denoising using the information obtained therefrom.

[0006] Prior art literature

[0007] (Patent Document 1) Korean Patent No. 10-2448069 B1 (September 22, 2022)

[0008]

[0009] The present disclosure is intended to solve the above-described problem, and provides a method and device for providing a learning model for updated denoising by reflecting information obtained by applying an image included in a low-dose image to a learning model.

[0010] Specifically, the present invention seeks to provide a learning model for updated denoising by using global information representing image characteristics of the entire object obtained by dividing a low-dose image into successive portions and applying the same to a learning model, and local information representing image characteristics of each region of the object obtained by randomly dividing the image into successive portions and applying the same to a learning model.

[0011] The problems to be solved in the present disclosure are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0012]

[0013] As a technical means for achieving the above-described technical problem, a method for providing a learning model for denoising by a device according to the first aspect of the present disclosure may include the steps of: a receiving unit obtaining a low-dose image and a high-dose image of a target object; a processor dividing the low-dose image so as to be continuous to obtain a low-dose continuous segmented image; the processor applying the low-dose continuous segmented image to a learning model to obtain global information representing image characteristics of the entire target object; a step by the processor randomly dividing the low-dose image to obtain a low-dose random segmented image; a step by the processor applying the low-dose random segmented image to the learning model to obtain local information representing image characteristics of each region of the target object; and a step of updating the learning model using the global information and the local information to obtain an updated learning model.

[0014] In addition, the step of obtaining the global information may include a step in which the processor applies the low-dose continuous segmented image to the learning model to obtain a denoising continuous segmented image for the object; a step in which the processor obtains a denoising full image using the denoising continuous segmented image; and a step in which the processor obtains the global information based on a comparison result between the denoising full image and the high-dose image.

[0015] In addition, the step of obtaining the local information may include a step in which the processor applies the low-dose random segmentation image to the learning model to obtain a denoising random segmentation image for the object; a step in which the processor randomly segments the high-dose image to obtain a high-dose random segmentation image; and a step in which the processor obtains the local information based on a comparison result between the denoising random segmentation image and the high-dose random segmentation image.

[0016] In addition, the method may further include a step of obtaining a new low-dose image of the new object and a denoising new image of the new object using the updated learning model.

[0017] Additionally, the global information may include brightness information for the entire object.

[0018] Additionally, the local information may include information about the shape of a structure included in the target object.

[0019] In addition, the step of obtaining the global information may include a step in which the processor applies the low-dose continuous segmented image to the learning model to obtain a predicted high-dose image for the object; and a step in which the processor obtains the global information based on a comparison result between the predicted high-dose image and the high-dose image.

[0020] In addition, the step of obtaining the local information may include a step in which the processor applies the low-dose random segmentation image to the learning model to obtain a predicted high-dose random segmentation image for the object; a step in which the processor randomly segments the high-dose image to obtain a high-dose random segmentation image; and a step in which the processor obtains the local information based on a comparison result between the predicted high-dose random segmentation image and the high-dose random segmentation image.

[0021] Additionally, the step of acquiring the low-dose continuous segmented images may determine the number of the low-dose continuous segmented images according to the size of the low-dose image and the predicted noise level for the low-dose image.

[0022] In addition, the step of obtaining the global information based on the comparison result between the denoised full image and the high-dose image may include the step of obtaining a first intensity value indicating a degree of brightness or darkness for the denoised full image; the step of obtaining a second intensity value indicating a degree of brightness or darkness for the high-dose image; and the step of obtaining the global information based on an error value output by applying the first intensity value and the second intensity value to a loss function.

[0023] In addition, the step of obtaining the local information based on the comparison result between the denoising random segmentation image and the high-dose random segmentation image may include the step of obtaining a first sharpness value indicating the sharpness of an edge of a shape of a structure included in the object included in the denoising random segmentation image; the step of obtaining a second sharpness value indicating the sharpness of an edge of a shape of a structure included in the object included in the high-dose random segmentation image corresponding to the denoising random segmentation image; and the step of obtaining the local information based on an error value output by applying the first sharpness value and the second sharpness value to a loss function.

[0024] A device providing a learning model for denoising according to a second aspect of the present disclosure may include a receiving unit that acquires a low-dose image and a high-dose image of a target object; and a processor that segments the low-dose image so as to be continuous to acquire a low-dose continuous segmented image, applies the low-dose continuous segmented image to a learning model to acquire global information representing image characteristics of the entire target object, randomly segments the low-dose image to acquire a low-dose random segmented image, applies the low-dose random segmented image to the learning model to acquire local information representing image characteristics of each region of the target object, and updates the learning model using the global information and the local information to acquire an updated learning model.

[0025] In addition, a third aspect of the present disclosure can provide a computer-readable recording medium having recorded thereon a program for executing the method of the first aspect on a computer.

[0026]

[0027] According to one embodiment of the present disclosure, by reflecting both global information representing image characteristics of the entire object obtained by dividing an image included in a low-dose image into successive parts and applying it to a learning model, and local information representing image characteristics of each region of the object obtained by randomly dividing it and applying it to a learning model, a learning model optimization can be performed, and a problem of difficulty in preserving detailed information on a low-dose image and texture on the object can be solved.

[0028] Additionally, as a learning model optimized for low-dose images is performed, the quality of low-dose images can be improved by using resources that are easy and efficient for improving the performance of the learning model.

[0029] The effects of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0030]

[0031] FIG. 1 is a block diagram schematically illustrating the configuration of a device that provides a learning model for denoising according to one embodiment.

[0032] FIG. 2 is a flowchart illustrating a method by which a device provides a learning model for denoising according to one embodiment.

[0033] FIG. 3 is a diagram illustrating a process of applying a low-dose continuous segmented image obtained by segmenting a low-dose image to a learning model according to an embodiment of the present invention.

[0034] FIG. 4 is a diagram illustrating an example of a device according to one embodiment obtaining global information by applying a denoising full image and a high-dose image obtained from a denoising continuous segmentation image to a loss function.

[0035] FIG. 5 is a diagram illustrating a process in which a device, according to one embodiment, randomly divides a low-dose image and applies the acquired low-dose randomly divided image to a learning model.

[0036] FIG. 6 is a diagram illustrating an example of a device according to one embodiment obtaining local information by applying a denoising random segmentation image and a high-dose random segmentation image to a loss function.

[0037] FIG. 7 is a diagram illustrating an example of a device obtaining a denoised new image through an updated learning model using global information and local information according to one embodiment.

[0038] FIG. 8 is a diagram schematically illustrating a flow of interaction between a device, a photographing device, a database, and a user device according to one embodiment.

[0039]

[0040] The advantages and features of the present disclosure, and the methods for achieving them, will become clearer with reference to the embodiments described below in detail together with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided solely to ensure that the disclosure is complete and to fully inform those skilled in the art of the scope of the present disclosure.

[0041] The terminology used herein is for the purpose of describing embodiments only and is not intended to limit the present disclosure. In this specification, the singular also includes the plural unless specifically stated otherwise. As used herein, the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components in addition to the mentioned components. Like reference numerals refer to like components throughout the specification, and "and / or" includes each and any combination of one or more of the mentioned components. Although "first", "second", etc. are used to describe various components, these components are not limited by these terms. These terms are only used to distinguish one component from another. Therefore, it should be understood that a first component mentioned below may also be a second component within the technical spirit of the present disclosure.

[0042] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in the sense commonly understood by those skilled in the art. Furthermore, terms defined in commonly used dictionaries should not be interpreted ideally or excessively unless explicitly and specifically defined otherwise. Spatially relative terms such as "below," "beneath," "lower," "above," and "upper" may be used to easily describe the relationship between one component and other components, as illustrated in the drawings.

[0043] Spatially relative terms should be understood to encompass different orientations of components during use or operation, in addition to the orientation depicted in the drawings. For example, if a component depicted in a drawing is flipped over, a component described as "below" or "beneath" another component may end up "above" the other component. Thus, the exemplary term "below" can encompass both the above and below orientations. Components can also be oriented in other directions, and thus spatially relative terms can be interpreted based on their orientation.

[0044] The term "model" as used herein may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units to solve a specific problem, or an abstract model of a processing process to solve a specific problem. For example, a neural network "model" may refer to the entire system implemented as a neural network that has problem-solving capabilities through learning. In this case, the neural network can have problem-solving capabilities by optimizing the parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a set of neural networks that are a combination of multiple neural networks.

[0045] The term "image" used in this disclosure may refer to multidimensional data composed of discrete image elements. In other words, "image" may be understood as a term referring to a digital representation of an object visible to the human eye. For example, "image" may refer to multidimensional data composed of elements corresponding to pixels in a two-dimensional image. "Image" may refer to multidimensional data composed of elements corresponding to voxels in a three-dimensional image.

[0046] Below, various embodiments are described in detail with reference to the drawings.

[0047] Figure 1 is a block diagram schematically illustrating the configuration of a device (100) that provides a learning model for denoising according to one embodiment.

[0048] Referring to the drawing, the device (100) may include a receiver (110), a processor (120), and a transmitter (130).

[0049] Those skilled in the art will appreciate that, in addition to the components illustrated in FIG. 1, the device (100) may further include other general-purpose components. For example, the device (100) may further include memory (not illustrated). Alternatively, those skilled in the art will appreciate that, according to other embodiments, some of the components illustrated in FIG. 1 may be omitted.

[0050] According to one embodiment, the receiver (110) can acquire low-dose images and high-dose images of a subject from a photographing device or a database. The low-dose and high-dose images may be images included in a medical image in which a patient or a portion of a patient is photographed, or a simulation model corresponding to the shape of a portion of a patient is photographed.

[0051] Additionally, the receiver (110) can obtain the number of pre-stored consecutive segments from the database. The number of pre-stored consecutive segments can be an optimized number of segments for obtaining global information representing image characteristics of the entire object by segmenting the low-dose image so that it can be continuous, and can be updated according to learning results obtained from the learning model.

[0052] Additionally, the receiver (110) can obtain a number of pre-stored random segments from the database. The number of pre-stored random segments may refer to an optimized number of segments for randomly dividing a low-dose image to acquire local information representing image characteristics of each region of the target object, and may be updated according to learning results obtained from the learning model.

[0053]

[0054] According to an embodiment, the processor (120) may obtain a low-dose continuous segmented image by dividing a low-dose image of an object obtained from the receiver (110) into consecutive images. In addition, the processor (120) may obtain global information representing image characteristics of the entire object by applying the low-dose continuous segmented image to a learning model. In addition, the processor (120) may obtain a low-dose random segmented image by randomly dividing the low-dose image of an object obtained from the receiver (110). In addition, the processor (120) may obtain local information representing image characteristics of each region of the object by applying the low-dose random segmented image to a learning model. In addition, the processor (120) may update the learning model using the global information and the local information to obtain an updated learning model.

[0055] A processor (120) according to an embodiment may process computational processes such as processing input data for updating a learning model, extracting features, and calculating error values ​​based on backpropagation. The processor (120) 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), or a field programmable gate array (FPGA). The type of the processor (120) described above is only one example, and thus, the type of the processor (120) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0056] According to an embodiment, a processor (120) may generate training data for a deep learning learning model for denoising images included in low-quality medical images to high-quality images. The processor (120) may generate input data and label data for training the deep learning model using metadata associated with medical data. The processor (120) may generate input data and label data for the deep learning learning model from the medical data, taking into account the characteristics of the metadata of the medical data. At this time, the medical data may include accelerated imaging or accelerated simulated medical images. In addition, the deep learning learning model may include a neural network model that restores the accelerated imaging or accelerated simulated medical images to a normal imaging state. In other words, the deep learning learning model may include a neural network model that denoises the accelerated imaging or accelerated simulated medical images or generates output that improves the resolution. For example, the neural network model may include U-NET, which is based on a convolutional neural network. The above-described type of neural network is only an example, and can be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0057]

[0058] FIG. 2 is a flowchart illustrating a method for providing a learning model for denoising by a device (100) according to one embodiment.

[0059] Referring to step S210, the device (100) can acquire low-dose images and high-dose images of the target object.

[0060] For example, low-dose images and high-dose images can be acquired from one or more frames included in a medical image. The term "medical image" is not limited to a specific medical image and has the same meaning as generally understood by those skilled in the art.

[0061] For example, the device (100) can acquire a low-dose image and a high-dose image of a target object in a pair form from a photographing device. A low-dose image can reduce the dose as the photographing time is reduced, but may refer to an image that requires improvement in image quality due to severe blurring. In addition, when the target object corresponds to a general patient, the general patient is likely to be overexposed to radiation during the process of photographing a high-dose image of the target object, and it may be difficult to acquire information about the target object with the same phase due to the movement of the general patient. Accordingly, the low-dose (e.g., 20 mAs) and high-dose (e.g., 200 mAs) images acquired by the device (100) in a pair form can be acquired by photographing from a simulation model including a skull phantom or a human phantom with a fixed position. Without being limited thereto, the device (100) can acquire low-dose and high-dose images of a specific target object pre-stored from a database rather than a photographing device in a pair form.

[0062]

[0063] Referring to step S220, the device (100) can obtain a low-dose continuous segmented image by dividing the low-dose image so that it can be continuous.

[0064] For example, a low-dose continuous segmented image may include one or more low-dose continuous segmented images (hereinafter also referred to as “forward patches”) that are segmented so as to be continuous from the low-dose image and applied to a learning model. For example, the device (100) may acquire N low-dose continuous segmented images by segmenting the low-dose image into 0 forward patches, 1 forward patch, ..., N forward patches. In addition, the device (100) may determine the size and N number of low-dose continuous segmented images differently depending on a selection input for the size of the low-dose image or the N number of user accounts.

[0065] FIG. 3 is a drawing for explaining a process of applying a low-dose continuous segmented image (310) obtained by dividing a low-dose image (300) so that it can be continuous to a learning model (320) by a device (100) according to one embodiment.

[0066]

[0067] Referring to FIG. 3, a device (100) according to an embodiment can obtain N low-dose continuous segmented images (310) by dividing the obtained low-dose image (300) into N numbers (e.g., N=9) corresponding to the size (e.g., 1000x1000x600). The obtained low-dose continuous segmented images (310) are applied to a learning model (320), so that the device (100) can obtain global information representing image characteristics for the entire object. In addition, the device (100) can determine N low-dose continuous segmented images (310) according to the number of pre-stored continuous segments obtained from a database.

[0068] As an effect of this, the number of consecutive segments of the base memory is the number of optimized forward patches updated by applying one or more existing low-dose images (300) to the learning model (320), thereby enabling the acquisition of precise global information representing the image characteristics of the entire object included in the low-dose image (300).

[0069]

[0070] According to one embodiment, the device (100) can determine the number of low-dose continuous segmentation images (310) according to the size of the low-dose image (300) and the predicted noise level for the low-dose image (300).

[0071] The predicted noise level may mean something included in the image preprocessing performed by the device (100) as the low-dose image (300) is acquired. That is, each pixel included in the low-dose image (300) is scanned, and the device (100) can determine a predicted noise level (e.g., phase) indicating the degree of blurring for the low-dose image (300).

[0072] Specifically, when the size of the low-dose image (300) is greater than or equal to a preset image size and the predicted noise level (e.g., top) for the low-dose image (300) is greater than or equal to a preset noise level (e.g., middle), the device (100) may determine the number of low-dose continuous segmented images (310) obtained by segmenting the low-dose image (300) so that it can be continuous to be greater than or equal to a threshold number (e.g., N = 18).

[0073] As an effect of this, when the noise in the low-dose image (300) is severe before being applied to the learning model (320), the probability of errors in the global information representing the image characteristics of the entire object is reduced by increasing the number of low-dose continuous segmented images (310) to a critical number or more, thereby enabling efficient updating of the learning model (320).

[0074]

[0075] Referring to step S230, the device (100) can obtain global information representing image characteristics of the entire object by applying a low-dose continuous segmentation image (310) to a learning model (320).

[0076] For example, the learning model (320) to which the low-dose continuous segmented image (310) is applied is not limited to a specific learning model, and a deep learning learning model capable of extracting features or characteristics for the entire object from each low-dose continuous segmented image (310) may be used. In addition, the learning model (320) may be updated and added according to the selection input of the user account. The global information acquired by being applied to the learning model (320) is a global feature extracted to improve the denoising performance for the low-dose image (300). Since the low-dose image (300) contains a mixture of bright and dark parts, the brightness information for the entire object to which uniformity for the brightness of the entire part can be applied may be included.

[0077]

[0078] According to one embodiment, a device (100) can obtain a denoising continuous segmented image for a target object by applying a low-dose continuous segmented image (310) to a learning model (320), obtain a denoising full image using the denoising continuous segmented image, and obtain global information based on a comparison result between the denoising full image and the high-dose image.

[0079] Specifically, the denoising continuous segmentation image may refer to a denoising result value for each forward patch when a low-dose continuous segmentation image (310) is applied to a learning model (320). Once the denoising continuous segmentation image is obtained, the device (100) can combine them to obtain a denoising full image for comparison with a high-dose image. Accordingly, global information including brightness information for the entire object can be obtained by comparing each pixel information included in the denoising full image with each pixel information of the high-dose image.

[0080] As an effect of this, there is an effect of preventing the problem of learning time becoming relatively long when processing is performed without considering the amount of computation according to the size of the low-dose image (300). That is, by comparing the denoising full image, in which each denoising continuous segmented image is combined, with the high-dose image, brightness information for the entire object is obtained, thereby efficiently improving the performance of the learning model (320).

[0081]

[0082] According to one embodiment, a device (100) can obtain a predicted high-dose image for a target object by applying a low-dose continuous segmentation image (310) to a learning model (320) and can obtain global information based on a comparison result between the predicted high-dose image and the high-dose image.

[0083] Specifically, the predicted high-dose image may mean a denoised image that is a combined form of a low-dose continuous segmented image (310) applied to a learning model (320) and is an image similar to a high-dose image. That is, the device (100) may obtain global information representing image characteristics of the entire object from the comparison result obtained by comparing the brightness level of the predicted high-dose image with the brightness level of the high-dose image.

[0084] As an effect of this, the process of obtaining an additional denoising full image is reduced by obtaining a predicted high-dose image predicted as a high-dose image rather than a denoising continuous segmentation image from a low-dose continuous segmentation image (310) applied to the learning model (320), thereby shortening the execution time of the learning model (320).

[0085]

[0086] FIG. 4 is a drawing for explaining an example of a device (100) according to one embodiment of the present invention obtaining global information (440) by applying a denoising full image (410) and a high-dose image (420) obtained from a denoising continuous segmentation image (400) to a loss function (430).

[0087] For example, the device (100) may obtain a first intensity value indicating the degree of brightness or darkness for the entire denoising image (410), obtain a second intensity value indicating the degree of brightness or darkness for the high-dose image (420), and obtain global information (440) based on an error value output by applying the first intensity value and the second intensity value to a loss function (430).

[0088] Referring to FIG. 4, the loss function (430) may refer to a function for comparing a first intensity value (e.g., 20%) for the denoising full image (410) and a second intensity value (e.g., 28%) for the high-dose image (420) with a denoising full image (410) and a high-dose image (420) as input values ​​to produce an error value (e.g., 8%). The first intensity value and the second intensity value may be obtained from an average intensity value for all pixels included in each of the denoising full image (410) and the high-dose image (420), and the device (100) may reflect the error value output using the loss function (430) to global information (440) to update the existing learning model (320).

[0089] As an effect of this, when a new low-dose image is applied to the learning model (320), the error value for the brightness value obtained from the loss function (430) is reflected and updated in the error value used in the existing learning model (320), so that an improved denoising continuous segmentation image (400) is obtained from the low-dose continuous segmentation image (310), thereby gradually increasing the effect of improving the image quality of the entire object.

[0090]

[0091] Referring to step S240, the device (100) can randomly divide the low-dose image (300) to obtain a low-dose randomly divided image.

[0092] In the above-described step S230, the device (100) obtains N low-dose continuous segmented images (310) by dividing the low-dose image (300) so that it can be continuous, and applies the obtained N low-dose continuous segmented images (310) to the learning model (320) to obtain global information (440) in which brightness information or global features for the target object are extracted. In step S240, as a process of obtaining local information by extracting information or local features for the shape of a structure included in the target object, the device (100) can obtain a low-dose random segmented image (hereinafter, also referred to as a 'random patch') by randomly dividing a portion of the target object included in the low-dose image (300). For example, the device (100) can obtain M low-dose random segmented images by dividing the low-dose image (300) into 0 random patches, 1 random patch, ..., M random patches. In addition, depending on the size of the low-dose image (300) or the shape of the structure included in the target object, the device (100) may determine the size and the number of M of the low-dose random segmentation image differently. For example, each low-dose random segmentation image may include a structure that is part of the target object, and may also include soft tissue that is part of the target object, but is not limited thereto. That is, when the target object corresponds to a tooth, it is obvious that the low-dose random segmentation image is not limited to any one part of the target object, since a person skilled in the art can intuitively infer the name related to the target object appearing in the image.

[0093]

[0094] FIG. 5 is a drawing for explaining a process in which a device (100) according to one embodiment randomly divides a low-dose image (300) and applies the obtained low-dose randomly divided image (500) to a learning model (320).

[0095]

[0096] Referring to FIG. 5, a device (100) according to an embodiment can divide an acquired low-dose image (300) into M random patches (e.g., M=6) corresponding to the shape of a structure (e.g., tooth length) included in a target object in the low-dose image (300), thereby obtaining M low-dose random segmented images (500). The acquired low-dose random segmented images (500) can be applied to a learning model (320), so that the device (100) can obtain local information representing image characteristics of each region of the target object. In addition, the device (100) can determine M low-dose random segmented images (500) according to the number of pre-stored random segments acquired from a database.

[0097] As an effect of this, the number of base-stored random segments is the number of optimized random patches updated by applying one or more existing low-dose images (300) to the learning model (320), enabling the acquisition of precise local information representing the image characteristics of each region of the target object.

[0098] According to an embodiment, the device (100) may determine the number of low-dose random segmented images (500) differently depending on the size of the low-dose image (300). That is, when the size of the low-dose image (300) (e.g., size 6) is greater than or equal to a preset reference size (e.g., size 5), the device (100) may obtain an additional segmented number representing an additional number of segments from the number of existing low-dose random segmented images (500) and apply the low-dose random segmented images (500) to the learning model (320). The unit for the size of the low-dose image (300) is not limited to a specific unit and may be added and updated according to a user's input.

[0099] As an effect of this, in the case of a low-dose image (300) having a size larger than a preset standard size, a larger number of low-dose random segmented images (500) are acquired and applied to the learning model (320), so that more precise features or features of each region of the object corresponding to the structure or soft tissue of the object can be extracted.

[0100] According to an embodiment, the device (100) may determine the number of low-dose random segmented images (500) differently depending on the area of ​​the structure included in the target object in the low-dose image (300). That is, in the case of a structure area (e.g., molars) for the target object in the low-dose image (300), the shape of the molars is more likely to appear as a complex curvature than the shape of the front teeth, and thus noise may appear severely at the boundary between teeth. Accordingly, the device (100) may acquire the structure area corresponding to the molars in the low-dose image (300) by the number of low-dose random segmented images (500) (e.g., M=5), and may acquire the structure area corresponding to the front teeth by the number of low-dose random segmented images (500) (e.g., M=3).

[0101] As an effect of this, since noise is expected to become severe in some areas with complex curvatures in the shape of the structure in the low-dose image (300), by dividing the number of low-dose random segmented images (500) into a larger number than in some areas with non-complex curvatures, the probability of errors in local information representing the image characteristics of each area of ​​the target object is high.

[0102]

[0103] Referring to step S250, the device (100) can obtain local information representing image characteristics of each region of the target object by applying a low-dose random segmentation image (500) to the learning model (320).

[0104] For example, the learning model (320) to which the low-dose random segmentation image (500) is applied is not limited to a specific learning model, and a deep learning learning model capable of extracting features or characteristics of each region of the target object from each low-dose random segmentation image (500) may be used. The local information acquired by being applied to the learning model (320) is a local feature that represents the characteristics of each region of the target object to improve the denoising performance for the low-dose image (300), and may include information on the shape of a structure included in the target object that can preserve edges that appear in each region of the target object.

[0105]

[0106] According to one embodiment, a device (100) can obtain a denoising random segmentation image for a target object by applying a low-dose random segmentation image (500) to a learning model (320), can obtain a high-dose random segmentation image by randomly segmenting a high-dose image (420), and can obtain local information based on a comparison result between the denoising random segmentation image and the high-dose random segmentation image.

[0107] Specifically, the denoising random segmentation image may mean a denoising result value for each random patch when a low-dose random segmentation image (500) is applied to the learning model (320). When the denoising random segmentation image is obtained, the device (100) may obtain a high-dose random segmentation image corresponding to each area in each denoising random segmentation image. Accordingly, by performing a comparison between pixel information included in each denoising random segmentation image and pixel information of each high-dose random segmentation image, local information including information on the shape of a structure included in the target object may be obtained.

[0108] As an effect of this, there is an effect of preventing the problem of difficult resource management due to the high probability of unexpected overload because the amount of computation varies depending on the size of the low-dose image (300) in the process of acquiring the image characteristics of each region of the target object. That is, by performing a comparison with each denoising random segmentation image and its corresponding high-dose image, local information for each region of the target object (e.g., structure region or soft tissue) is acquired, thereby efficiently improving the performance of the learning model (320).

[0109]

[0110] According to one embodiment, a device (100) can obtain a predicted high-dose random segmentation image for a target object by applying a low-dose random segmentation image (500) to a learning model (320), can obtain a high-dose random segmentation image by randomly segmenting a high-dose image (420), and can obtain local information based on a comparison result between the predicted high-dose random segmentation image and the high-dose random segmentation image.

[0111] Specifically, the predicted high-dose random segmentation image may refer to a random segmentation image that is denoised by applying the low-dose random segmentation image (500) to the learning model (320). That is, the device (100) may obtain local information representing the image characteristics of each region of the object from the comparison result obtained by comparing the edge sharpness (e.g., 40%) of a part of a structure included in the object for each predicted high-dose random segmentation image with the edge sharpness (e.g., 50%) of a part of a structure included in the object for each corresponding high-dose random segmentation image.

[0112] As an effect of this, since a predicted high-dose random segmentation image, rather than a denoising random segmentation image, is obtained from the low-dose random segmentation image (500) applied to the learning model (320), the time for performing comparison with the high-dose image (420) is shortened, so that local information can be quickly obtained.

[0113]

[0114] FIG. 6 is a drawing for explaining an example of a device (100) according to one embodiment obtaining local information (620) by applying a denoising random segmentation image (600) and a high-dose random segmentation image (610) to a loss function (430).

[0115] For example, the device (100) can obtain a first sharpness value indicating the sharpness of an edge of a shape of a structure included in a target object included in a denoising random segmentation image (600), can obtain a second sharpness value indicating the sharpness of an edge of a shape of a structure included in a target object included in a high-dose random segmentation image (610) corresponding to the denoising random segmentation image (600), and can obtain local information (620) based on an error value output by applying the first sharpness value and the second sharpness value to a loss function (430).

[0116] Referring to FIG. 6, the loss function (430) may be a function for calculating an error value (e.g., 10%) by comparing a first sharpness value (e.g., 60%) representing the sharpness of an edge of a shape of a structure included in a target object of each denoising random segmentation image (600) with a second sharpness value (e.g., 70%) representing the sharpness of an edge of a shape of a structure included in a target object of each denoising random segmentation image (600) with a high-dose random segmentation image (610) corresponding thereto, by applying the denoising random segmentation image (600) and the high-dose random segmentation image (610) corresponding thereto as input values. The first sharpness value and the second sharpness value can be obtained from the average sharpness value of the pixels adjacent to the edge of the shape of the structure included in the target object corresponding to the denoising random segmentation image (600) and the high-dose random segmentation image (610), respectively, and the device (100) can update the existing learning model (320) by reflecting the error value output using the loss function (430) to the local information (620).

[0117] As an effect of this, when a new low-dose image is applied to the learning model (320), the error value for the sharpness value obtained from the loss function (430) is reflected and updated in the error value used in the existing learning model (320), so that the image quality of the shape of the structure included in the target object is gradually increased by obtaining an improved denoising random segmentation image (600) from the low-dose random segmentation image (500).

[0118]

[0119] According to an embodiment, a device (100) may obtain a target random segmentation image, which is one of a plurality of denoising random segmentation images (600), and may obtain a target random segmentation image, which is one of a plurality of high-dose random segmentation images (610), and may obtain local information (620) indicating image characteristics of each region of an object according to a similarity level indicating similarity between shape information of a structure included in the target random segmentation image and shape information of a structure included in the target random segmentation image. That is, when the shape of a structure included in an object of the target random segmentation image is similar to the shape of a structure included in an object of the target random segmentation image (e.g., the similarity level is equal to or higher than a preset similarity level), the device (100) may obtain local information (620) including information on the shape of a structure included in the object according to a comparison result between the target random segmentation image and the target random segmentation image. In addition, the device (100) can update the existing learning model (320) by reflecting the target random segmentation image and the target random segmentation image input to the loss function (430) and the output error value in local information (620).

[0120] As an effect of this, the target random segmentation image among the plurality of high-dose random segmentation images (610) that are most similar to the target random segmentation image among the plurality of denoising random segmentation images (600) is compared, thereby determining a precise error value, so that the performance of the existing learning model (320) can be improved more quickly.

[0121]

[0122] According to an embodiment, a device (100) may update a learning model (320) according to an error value output by applying a denoising random full image and a high-dose image, which are formed by combining multiple denoising random segmented images (600) into a single image, to a loss function (430). Accordingly, based on a comparison result between the output error value (e.g., 1) and a reference error value (e.g., 1.2), the device (100) may update the reference error value to a value closer to 0, thereby obtaining an updated learning model that reflects this.

[0123] As an effect of this, as the reference error value gets closer to 0, the learning model (320) is updated, and a new low-dose image for a new subject can be gradually improved to an image close to a high-dose image by obtaining a new denoised image with minimized noise due to the updated learning model.

[0124]

[0125] In addition, the error value between the portion of the object appearing in each random patch of the high-dose random segmentation image (610) corresponding to the portion of the object appearing in each random patch of the denoising random segmentation image (600) is obtained, thereby shortening the comparison execution time and the resources used.

[0126]

[0127] Referring to step S260, the device (100) can obtain an updated learning model by updating the learning model (320) using global information (440) and local information (620).

[0128] Referring to the above-described steps S230 and S250, the device (100) acquires global information (440) representing image characteristics of the entire object and local information (620) representing image characteristics of each region of the object. Accordingly, by reflecting the acquired global information (440) and local information (620) into the existing learning model (320), the device (100) can acquire an updated learning model.

[0129]

[0130] FIG. 7 is a diagram illustrating an example of a device (100) according to one embodiment obtaining a denoised new image (710) through an updated learning model (700) using global information (440) and local information (620).

[0131] Referring to FIG. 7, the device (100) can obtain a denoised new image (710) for the new object using a new low-dose image for the new object and an updated learning model (700). By applying a low-dose continuous segmented image (310) to an existing learning model (320), global information (440) representing image characteristics for the entire object is obtained, and by applying a low-dose random segmented image (500) to the learning model (320), local information (620) representing image characteristics for each region of the object is obtained, and the device (100) can reflect this in the existing learning model (320) to update it, thereby obtaining an updated learning model (700).

[0132] According to an embodiment, the device (100) can obtain an updated learning model (700) by updating the learning model (320) based on different weights assigned to each of the global information (440) and the local information (620). That is, among the update value (e.g., 0.5) corresponding to the global information (440) including brightness information for the entire object and the update value (e.g., 0.3) corresponding to the local information (620) including information on the shape of a structure included in the object, a higher weight is assigned to the local information (620) that is closer to 0, and the device (100) can obtain an updated learning model (700) by reflecting this in the existing learning model (320).

[0133]

[0134] [Mathematical Formula 1]

[0135] Loss Function = Local Information + 0.5 * Global Information

[0136] According to an embodiment, the device (100) can obtain an updated learning model (700) by assigning preset weights to global information (440) and local information (620), respectively. Referring to Equation 1, the device (100) can calculate an error value of a loss function (430) by assigning preset weights to global information (440) obtained according to the comparison result of the obtained denoising full image (410) and high-dose image (420) and local information (620) obtained according to the comparison result of the denoising random segmented image (600) and high-dose random segmented image (610). Accordingly, the device (100) can obtain an updated learning model (700) by reflecting the calculated error value.

[0137] As an effect of this, by assigning different weights to the global information (440) obtained by extracting the global features for the low-dose image (300) and the local information (620) obtained by extracting the local features for the low-dose image (300), a balanced updated learning model (700) can be obtained each time it is updated, and a denoising new image (710) with improved quality can be obtained from a new low-dose image for a new object.

[0138] FIG. 8 is a diagram schematically illustrating a flow of interaction between a device (100) according to one embodiment, a photographing device (800), a database (820), and a user device (810).

[0139] Referring to FIG. 8, the device (100) can obtain a low-dose image (300) and a high-dose image (420) of a target object from a photographing device (800). In addition, if there is a history of low-dose images (300) and high-dose images (420) corresponding to the target object being obtained in the past, the device (100) can obtain the low-dose image (300) and the high-dose image (420) of the target object from a database (820). The device (100) obtains a low-dose continuous segmented image (310) obtained by segmenting the low-dose image (300) so that it can be continuous, and a low-dose random segmented image (500) obtained by randomly segmenting the low-dose image (300), and applies the obtained images to a learning model (320), thereby obtaining global information (440) representing image characteristics of the entire target object and local information (620) representing image characteristics of each region of the target object, respectively. Accordingly, the device (100) can obtain an updated learning model (700) by updating the learning model (320) using the acquired global information (440) and local information (620), and when a denoising low-dose image for an object showing the most similar quality to the high-dose image (420) is requested from the user device (810), the device (100) can provide the user device (810) with a denoising low-dose image that has been trained with the updated learning model (700) and improved to a high quality.

[0140]

[0141] Various embodiments of the present disclosure may be implemented as software comprising one or more instructions stored in a storage medium (e.g., memory) readable by a machine (e.g., a display device or a computer). For example, a processor (e.g., processor 120) of the machine may call at least one instruction from the storage medium and execute it. This enables the machine to operate to perform at least one function according to the at least one called instruction. The one or more instructions may include code generated by a compiler or code executable by an interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' simply means that the storage medium is a tangible device and does not contain signals (e.g., electromagnetic waves), and the term does not distinguish between cases where data is stored semi-permanently or temporarily on the storage medium.

[0142] According to one embodiment, the method according to various embodiments disclosed in the present disclosure may be provided as included in a computer program product. The computer program product may be traded as a commodity between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or may be distributed online (e.g., downloaded or uploaded) via an application store (e.g., Play Store™) or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0143] Those skilled in the art will appreciate that the present invention can be implemented in modified forms without departing from the essential characteristics of the above-described description. Therefore, the disclosed methods should be considered illustrative rather than restrictive. The scope of the present invention is set forth in the claims, not the foregoing description, and all differences within the scope equivalent thereto should be construed as being encompassed by the present invention.

[0144]

[0145] Description of the symbol

[0146] 100: Device 110: Receiver

[0147] 120: Processor 130: Transmitter

[0148] 300: Low-dose image 310: Low-dose sequential segmentation image

[0149] 320: Learning Model

[0150] 400: Denoising continuous segmented image 410: Denoising full image

[0151] 420: Image of the amount of data 430: Loss function

[0152] 440: Global Information

[0153] 500: Low-dose randomly segmented image 600: Denoising randomly segmented image

[0154] 610: High-dose random segmentation image 620: Local information

[0155] 700: Updated training model 710: Denoising new images

[0156] 800: Shooting device 810: User device

[0157] 820: Database

Claims

1. In a method for a device to provide a learning model for denoising, A step of the receiver acquiring a low-dose image and a high-dose image of the target object; A step of obtaining a low-dose continuous segmented image by dividing the low-dose image so that the processor can continue; A step in which the processor applies the low-dose continuous segmented image to the learning model to obtain global information representing image characteristics of the entire object; A step in which the processor randomly divides the low-dose image to obtain a low-dose randomly divided image; The step of the processor applying the low-dose random segmented image to the learning model to obtain local information representing the image characteristics of each region of the object; and A method comprising: a step of obtaining an updated learning model by updating the learning model using the global information and the local information.

2. In paragraph 1, The steps for obtaining the above global information are A step in which the processor applies the low-dose continuous segmentation image to the learning model to obtain a denoised continuous segmentation image for the object; The step of the processor obtaining a denoised full image using the denoised continuous segmented image; and A method comprising: a step of obtaining the global information based on a comparison result between the denoised full image and the high-dose image by the processor; 3. In paragraph 1, The step of obtaining the above local information is A step in which the processor applies the low-dose random segmentation image to the learning model to obtain a denoising random segmentation image for the object; A step in which the processor randomly divides the high-dose image to obtain a high-dose randomly divided image; and A method comprising: a step of obtaining the local information based on a comparison result between the denoising random segmentation image and the high-dose random segmentation image by the processor; 4. In paragraph 1, A method further comprising: obtaining a new low-dose image of a new object and a denoising new image of the new object using the updated learning model.

5. In paragraph 1, A method wherein the global information includes brightness information for the entire object.

6. In paragraph 1, A method wherein the local information includes information about the shape of a structure included in the target object.

7. In paragraph 1, The steps for obtaining the above global information are The step of the processor applying the low-dose continuous segmented image to the learning model to obtain a predicted high-dose image for the object; and A method comprising: a step of obtaining the global information based on a comparison result between the predicted high-dose image and the high-dose image by the processor; 8. In paragraph 1, The step of obtaining the above local information is A step in which the processor applies the low-dose random segmentation image to the learning model to obtain a predicted high-dose random segmentation image for the object; A step in which the processor randomly divides the high-dose image to obtain a high-dose randomly divided image; and A method comprising: a step of obtaining the local information based on a comparison result between the predicted high-dose random segmentation image and the high-dose random segmentation image by the processor; 9. In paragraph 1, The step of acquiring the above low-dose sequential segmented images is A method for determining the number of low-dose continuous segmented images according to the size of the low-dose image and the predicted noise level for the low-dose image.

10. In paragraph 2, The step of obtaining the global information based on the comparison result between the denoising full image and the high-dose image is as follows: A step of obtaining a first intensity value representing the degree of brightness or darkness of the entire denoising image; A step of obtaining a second intensity value indicating the degree of brightness or darkness of the high-dose image; and A method comprising: obtaining the global information based on an error value output by applying the first intensity value and the second intensity value to a loss function; 11. In paragraph 3, The step of obtaining the local information based on the comparison result between the denoising random segmentation image and the high-dose random segmentation image is as follows: A step of obtaining a first sharpness value indicating the sharpness of an edge of a structure included in the target object included in the denoising random segmentation image; A step of obtaining a second sharpness value representing the sharpness of an edge of a structure included in the object included in the high-dose random segmentation image corresponding to the denoising random segmentation image; and A method comprising: obtaining the local information based on an error value output by applying the first sharpness value and the second sharpness value to a loss function; 12. In a device that provides a learning model for denoising, A receiving unit for acquiring low-dose images and high-dose images of a target object; and By dividing the above low-dose image into consecutive low-dose continuous segmented images, By applying the above low-dose continuous segmented image to a learning model, global information representing the image characteristics of the entire object is obtained, By randomly dividing the above low-dose image, a low-dose randomly divided image is obtained, Applying the low-dose random segmented image to the learning model, local information representing the image characteristics of each region of the object is obtained, A device comprising a processor that updates the learning model using the global information and the local information to obtain an updated learning model.

13. A computer-readable recording medium having recorded thereon a program for executing the method of any one of paragraphs 1 to 11 on a computer.

Images