KR20200083303A - Apparatus and method for increasing learning data using patch matching - Google Patents

Abstract

Disclosed are an apparatus and a method for increasing learning data using patch matching. The method for increasing learning data using patch matching comprises: a first step of generating a patch image by dividing an image into a size designated at an arbitrary angle in a medical image; a 2a step of performing classification through a clustering technique, and performing determination through a preset valid patch and similarity analysis; a 2b step of analyzing and determining the preset valid patch and a similar pattern through histogram analysis; a 2c step of analyzing and determining the degree of similarity for each patch using a similarity analysis function and a learned artificial neural network; and a third step of rotating the medical image at different angles, dividing the image into a specified size to generate a patch, and analyzing validity through one of the 2a, 2b, and 2c steps. The present invention can increase the valid patches.

Description

Apparatus and method for increasing learning data using patch matching{APPARATUS AND METHOD FOR INCREASING LEARNING DATA USING PATCH MATCHING}

An embodiment of the present invention relates to an apparatus and method for increasing learning data using patch matching.

In order to develop an artificial intelligence-based image analysis system, it is essential to secure a large amount of image data. In order to overcome these limitations, a method of generating an image fragment, that is, a patch, by dividing a large image into small pieces, and learning it is widely used. Nevertheless, it is difficult to secure a sufficient number of image data in the case of special image data such as pathological images for a specific cancer. In addition, in the case of such an image, there is a point in which the size of the repeated pattern to be analyzed is repeated non-uniformly, so it is difficult to secure the accuracy even by learning by dividing the patch.

On the other hand, since medical images have to undergo different image processing processes according to their types, there is a disadvantage that the detection or diagnosis device or its operating system is complicated and large. Most of the conventional lesion detection techniques using medical images require the user's participation, such as a doctor, to assist the user in diagnosing the lesion based on the medical image, or stay at a level that assists the user, and automatically diagnose the lesion or grade the lesion. There are still limitations in the analysis.

In recent years, in the medical field, machine learning or artificial intelligence has been introduced to promote active research and development. However, research and development results for automatically diagnosing lesions and analyzing the grades of lesions in connection with brain diseases such as stroke, etc. are still insignificant.

The present invention is intended to meet the needs of the existing technology, and an object of the present invention is to provide an apparatus and method for automatically dividing a patch effective for learning in a large-capacity image having a pattern that is uniformly repeated.

Another object of the present invention is to provide a computer-readable recording medium recording a program implementing the apparatus and method for increasing learning data using patch matching.

In order to solve the above technical problem, a method of increasing learning data using patch matching according to an aspect of the present invention includes: a first step of generating a patch image by dividing the image into a specified size at an arbitrary angle in a medical image; A second step of classifying through a clustering technique and determining through a preset valid patch and similarity analysis; A second step of analyzing and determining a predetermined effective patch and a similar pattern through histogram analysis; A 2c step of analyzing and determining similarity for each patch using the similarity analysis function and the learned artificial neural network; And a third step of rotating the medical image at a different angle and dividing the image to a specified size to generate a patch and analyzing the effectiveness through one of the 2a, 2b, and 2c.

In one embodiment, the method of increasing learning data using patch matching may further include, after the third step, ending the patch generation when the rotated angle of the medical image is 360 degrees.

In one embodiment, a method of increasing learning data using patch matching may include, after the step of terminating, further classifying patches that have been determined to be valid in the validity analysis using a clustering technique; And generating a single image by connecting a plurality of images in the same cluster according to additional classification.

In one embodiment, a method of increasing learning data using patch matching, after generating the one image, crops the image generated in the generating step to a smaller size than one image to generate a new sized image. It may further include a step.

In one embodiment, the method may further include determining the validity again by analyzing a ratio of the new sized image overlapping the lesion area of the existing valid patch image.

In one embodiment, further comprising the step of generating a patch while rotating the image generated in the new size from 360 to 360 degrees at any angle, and augmenting the training data by analyzing the effectiveness of any one of the 2a to 2c above Can be.

An apparatus for increasing learning data using patch matching according to another aspect of the present invention for solving the above technical problem includes: a patch image generator for dividing an image into a specified size at an arbitrary angle in a medical image to generate a patch image; A first judging unit that classifies through a clustering technique and determines through a preset valid patch and similarity analysis; A second determination unit that analyzes and determines a preset effective patch and similar patterns through histogram analysis; A third determination unit that analyzes and determines similarity for each patch using a similarity analysis function and a learned artificial neural network; And a validity analysis unit that rotates the medical image at a different angle, divides the image to a specified size, generates a patch, and analyzes effectiveness through one of the first to third determination units.

In one embodiment, the apparatus for increasing learning data using patch matching further includes a basic task termination unit that ends patch generation when the rotated angle of the medical image is 360 degrees after validity analysis of the effectiveness analysis unit.

In one embodiment, the apparatus for increasing learning data using patch matching includes: a classification unit that further classifies patches, which have been determined to be valid in the validity analysis, into a clustering technique after the termination operation of the basic operation termination unit; And a first image generator configured to generate a single image by connecting a plurality of images in the same cluster according to additional classification.

In one embodiment, the apparatus for increasing learning data using patch matching further includes a second image generator that crops one image generated by the image generator to a smaller size than the one image to generate a new size image. It can contain.

In one embodiment, the effectiveness analysis unit may determine the validity again by analyzing the ratio of the new size of the image overlapping the lesion area of the existing valid patch image.

In one embodiment, further comprising a data augmentation unit for generating a patch while rotating the image generated in the new size to 360 degrees at any angle and analyzing the effectiveness with any one of the above 2a to 2c to enhance learning data do.

An apparatus according to another aspect of the present invention for solving the above technical problem includes a computer readable recording medium recording a program implementing a method for increasing learning data using patch matching in any one of the above-described embodiments.

In the case of using the apparatus and method for increasing learning data using the above-described patch matching, a patch effective for learning can be automatically divided in a large-capacity image in which a non-uniformly repeated pattern exists, and the effective patch can be increased. By increasing the learning data, there is an advantage that can effectively process the medical image.

1 is a flowchart of a method for increasing learning data using patch matching according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining an additional conversion process in the method of FIG. 1.
FIG. 3 is a diagram illustrating a process of obtaining a patch in a specific region of interest of a pathology image that can be employed in the process of increasing learning data in the method of FIG. 1.
4 is a structural diagram of a convolutional neural network employable in the method and apparatus of the present embodiment.
5 is a block diagram of an apparatus for increasing learning data using patch matching according to another embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are exemplified only for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention It can be implemented in various forms and is not limited to the embodiments described herein.

Embodiments according to the concept of the present invention can be applied to various changes and can have various forms, so the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosure forms, and includes all modifications, equivalents, or substitutes included in the spirit and scope of the present invention.

The terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as “comprises” or “haves” are intended to indicate the presence of features, numbers, steps, operations, elements, parts, or combinations thereof described herein, one or more other features. It should be understood that the existence or addition possibilities of fields or numbers, steps, actions, components, parts or combinations thereof are not excluded in advance.

Terms used in the present specification are as follows.

T2 weighted imaging (T2-weighted imaging): refers to a technique obtained from a specific pulse sequence (magnetic pulse imaging) from magnetic resonance imaging (MRI) or images obtained by this technique. Provide structural information.

FLAIR (Fluid attenuated inversion recovery): A signal acquisition technique using magnetic imaging devices or images obtained by this technique that makes it easier to detect lesions that are easily missed in T2-enhanced images by weakening the signals of the cerebrospinal fluid with long reversal and echo times. Refers to. FLAIR can be referred to as fluid attenuation inversion recovery.

DWI (Diffusion weighted imaging): Diffusion-weighted imaging mainly refers to diffusion-weighted images obtained from magnetic resonance imaging, and provides information on the degree and extent of diffusion of water molecules in cell tissues in a specific direction.

Perfusion weighted imaging (PWI): refers to perfusion-weighted images (simply, perfusion images) obtained from magnetic resonance imaging, and informs the change in concentration over time of the injected contrast agent.

Penumbra: A semi-shaded region in an image caused by an ischemic event or embolism, which indicates that the oxygen transport function is locally reduced to cause hypoxic cell death or to be viable upon proper treatment within a few hours.

ADC (Apparent diffusion coefficient): This is an apparent diffusion coefficient obtained from a magnetic resonance image, and provides information on a diffusion impeding factor in internal tissues of the human body.

Arterial phase (AP): A period of specific perfusion obtained from magnetic resonance imaging, which indicates the time when the contrast medium injected over time passes through the artery.

Capillary phase (CP): A period of specific perfusion obtained from magnetic resonance imaging. It indicates the time when the contrast medium injected over time passes through the capillary portion.

Venous phase (VP): A period of specific perfusion obtained from magnetic resonance imaging, which indicates when the contrast medium injected over time passes through the vein.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings.

1 is a block diagram of an apparatus for increasing learning data using patch matching according to an embodiment of the present invention. FIG. 2 is a diagram for explaining an additional conversion process in the method of FIG. 1. And FIG. 3 is a diagram illustrating a process of obtaining a patch in a specific region of interest of a pathology image that can be employed in the process of increasing learning data in the apparatus of FIG. 1.

Referring to FIG. 1, a method for increasing learning data using patch matching according to the present embodiment (hereinafter simply referred to as a'learning data increase method') is performed by generating a patch at a randomly multiple of a predetermined size in a medical image, and a multiple of a randomly specified size. When it is 360 degrees, it includes a step (first step) of generating a patch image to end patch generation (S10, S12, S14).

Next, the learning data increase method includes classifying through a clustering technique and determining through a preset valid patch and similarity analysis (step 2a) (S21, S31, S32).

The detailed description of step 2a is as follows.

The designated effective patch is divided into K groups according to the number of classes of lesions. Here, K is the number of types of lesions on the medical image. For example, K has 4 lesion classes if the medical image has annotation of late cancer, middle cancer, early cancer, and normal.

And, the distance between two adjacent patches is defined as the following [Equation 1].

In [Equation 1] above, A(i,j,k) means a pixel vector value of a 3D A patch. m is the width of the patch, and n is the height of the patch.

For example, when there is a patch A, if there are at least s points within a certain distance (e) from the patch A, this patch is called a core patch.

로 고려되는 집합의 범위를 고정한다. E와 S는 K개의 그룹과 겹쳐지지 않도록 사용자가 미리 결정하여 설정할 수 있다.here,

The range of the set considered as is fixed. E and S can be set by the user in advance so as not to overlap with the K groups.

Within a group of K, a certain distance (e) is changed according to each group to obtain a patch (A^) in which the number of patches within a certain distance from patch A is the maximum. If the maximum number of patches (A^) is indicated, it is as shown in [Equation 2] below.

이하이면, 유효한 패치로 판정할 수 있다. 유효성 분석에서 유사도 분석 대상이 되는 패치들 사이의 거리가

이하가 아니면 유효하지 않은 패치(비유효패치)로 판정할 수 있다.The distance between patches subject to similarity analysis is based on the core patch found according to each K group in the validity analysis.

If it is as follows, it can be judged as a valid patch. In the effectiveness analysis, the distance between patches subject to similarity analysis is

If not, it can be determined as an invalid patch (ineffective patch).

In addition, the learning data increasing method of the present embodiment may include a step (step 2b) of analyzing and determining a preset effective patch and a similar pattern through histogram analysis after the first step (S22, S31, S32). Step 2b may be performed in parallel with step 2a.

The step 2b will be described in detail as follows.

에 대해 mean(A(i,j,k))가 유효한 패치들의 (max(B(i,j,k))의 최솟값에 대해 기준 설정 배율보다 적으면 유효하다고 판정한다. 기준 설정 배율은 0.9배일 수 있다. 이러한 기준 설정 배율은 실험에 의해 최적화된 것이다.Fixed for RGB channels

For mean(A(i,j,k)), it is determined that it is valid if it is less than the reference setting magnification for the minimum value of (max(B(i,j,k)) of valid patches. These reference setting magnifications are optimized by experiments.

If the step 2b is expressed as an equation, it can be expressed as [Equation 3] below.

In addition, the learning data increasing method of the present embodiment may include a step of analyzing and determining the similarity (step 2c) for each patch using the similarity analysis function after the first step (S23, S31, S32). Step 2c (S23) may be performed in parallel with steps 2a (S21) and 2b (S22).

Step 2c will be described in detail as follows.

의 차원 3mn의 정규분포를

라고 하자. 그리고, 유효하지 않는 패치의 그룹에 대한 차원 3mn의 정규 분포를

라고 하자.Valid patch for each K group

Normal distribution of dimension 3mn

Let's say. Then, the normal distribution of dimension 3mn for the group of invalid patches

Let's say.

라고 하고, Z에 들어갈 확률을

라고 하면

이 된다.The probability that the pixel data of a given patch will enter group K

And the probability of entering Z

If you say

It becomes.

At this time, apply the Bayes rule and solve the following [Equation 4] for each K group.

의 K+1개의 방정식을 연립해서 풀어서 각

값을 구한 뒤 가장 높은 값으로 대상 패치의 유효그룹을 결정한다.

가 가장 클 경우 대상패치는 유효하지 않은 패치로 분류되고,

가 가장 클 경우 K 그룹중 I번째 그룹에 속하게 된다.In [Equation 4] above

Solve by solving K+1 equations of each

After obtaining the value, the effective group of the target patch is determined with the highest value.

If is the largest, the target patch is classified as an invalid patch,

If is the largest, it belongs to the I-th group of the K groups.

In addition, the method for increasing the learning data of the present embodiment can analyze the effectiveness through steps 2a, 2b, and 2c (S21, S22, S23, S31, and S32).

In addition, in the method of increasing the learning data of the present embodiment, after steps 2a, 2b, and 2c, patches (ineffective patches) that have not been judged as valid are added as a normal distribution for RGB clauses of valid pathological images. It may include the step of converting and generating (S33) (see FIG. 2). Through this, the increased number of effective patches can be classified (S40).

In addition, in the method for increasing the learning data of the present embodiment, the effectiveness of the additional transforms and generated patches may be analyzed again through steps 2a, 2b, and 2c.

The process of processing additional transforms and generating steps using the normal distribution of valid patches is as follows.

First, patches that have not been validated are normalized.

Then, the convergence operator converts the patch that has been judged to be a valid distribution for the RGB channel. If this is represented, it is as shown in [Equation 5].

In [Equation 5] above, N nonvalid means the Gaussian distribution for the RGB channel of the patch that has not been validated. Since there are 3 channels in R, G, and B, a total of 3 N valid means the normal distribution for the RGB channels of the valid patch.

The convolution operator means the following equation (6).

According to [Equation 5] and [Equation 6] above, the Gaussian distribution of a patch that does not receive a valid patch by the convolution operator is converted into a normal distribution for the RGB channels of the effective patch. Therefore, in this embodiment, an invalid patch is converted into a valid patch using this.

In addition, in the method of increasing the learning data of the present embodiment, after generating one image by valid patches obtained through the above process, the generated image is cropped at a constant small magnification than one image to generate a new size image. It may further include a step.

In addition, the method for increasing the learning data of the present embodiment may further include generating a new sized image and analyzing the ratio of the new sized image overlapping with the lesion area of the existing valid patch image to determine validity again. .

In addition, in the method of increasing the learning data of the present embodiment, after generating a new sized image, a patch is generated while rotating the image generated at a new size from an angle to 360 degrees, and any of the above steps 2a to 2c is performed. One may further include the step of augmenting the learning data by analyzing the effectiveness.

4 is a structural diagram of a convolutional neural network employable in an artificial neural network learning method and apparatus using the operation of the present embodiment.

The artificial neural network learning apparatus using the motion according to the present embodiment may use a deep learning architecture as an artificial intelligence neural network. The deep learning architecture uses motion convolutional neural network (CNN), deconvolution, and skip connection for medical image signals to learn motion based on motion-based labeling, and according to the learning results. It can be implemented to process a series of processes that generate external signals or generate specific noise.

As an example, the deep learning architecture may have a form including a convolutional network, a deconvolutional network, and a shortcut. As shown in FIG. 3, the deep learning architecture stacks a 3x3 size color convolution layer and an activation layer (ReLU) and extracts a 2x2 size filter to extract local features of the medical image (X). (stride) is applied to 1 to perform the operation of the convolution block that is connected to the next lower depth level 4 times, and then a 2x2 size deconvolution layer and activation layer (ReLU) are applied. After connecting to the next higher depth level, the operation of the inverse convolution block, which stacks a 3x3 size color convolution layer and an activation layer, is repeated 4 times, where each of the convolution networks including the operation of the convolution block of each level is performed. It may be made to copy and contate the result of the convolution of the corresponding level of the inverse convolution network of the same level to the image of the convolution block of the level and perform convolution operations in each block.

The convolution block in the convolution network and the deconvolution network may be implemented by a combination of conv-ReLU-conv layers. Further, the output of the deep learning architecture may be obtained through a classifier connected to a convolutional network or a deconvolutional network, but is not limited thereto. The classifier may be used to extract local features from an image using a fully connectivity network (FCN) technique.

Further, the deep learning architecture may be implemented to additionally use an insulation module or a multi filter pathway in a convolution block depending on the implementation. Different filters in the inception module or multi-filter path may include 1x1 filters.

For reference, in the case of an input image in a deep learning architecture having 32 horizontal, 32 vertical, and RGB channels, the size of the input image X corresponding to the medical image may be [32x32x3]. In the deep learning architecture's convolutional neural network (CNN), the convolutional (CONV) layer is connected to some areas of the input image, and can be designed to calculate the dot product of the connected areas and their weights. .

Here, the ReLU (rectified linear unit) layer is an activation function applied to each element, such as max(0,x). The ReLU layer does not change the size of the volume. The POOLING layer may output a reduced volume by performing downsampling or subsampling on a dimension represented by (horizontal, vertical).

Then, the fully-connected (FC) layer may calculate class scores and output a volume having a size of [1x1x10], for example. In this case, 10 numbers correspond to class scores for 10 categories. The pre-connection layer is connected to all elements of the previous volume. There, some layers may have parameters, while some layers may not. CONV/FC layers may include weight and bias as an activation function, not just input volume. Meanwhile, the ReLU/POOLING layers are fixed functions, and the parameters of the CONV/FC layer can be learned with a gradient descent so that the class score for each image is the same as the label of the corresponding image.

5 is a block diagram of an apparatus for increasing learning data using patch matching according to another embodiment of the present invention.

Referring to FIG. 5, the apparatus for increasing learning data using patch matching according to the present embodiment includes a memory 52 for storing a program and a control unit 51 connected to the memory 52 to perform a program. In addition, a communication unit 53 connected to a network to transmit and receive signals and data such as medical image signals may be further provided.

A program or software module may be mounted in the memory 52. The software module includes a patch image generation unit 521, a first determination unit 522, a second determination unit 523, a third determination unit 524, a validity analysis unit 525, and a basic operation termination unit 526. , A classification unit 527, a first image generation unit 528a, a second image generation unit 528b, and a data enhancement unit 529.

The patch image generation unit 521 generates a patch image by dividing the image into a specified size at an arbitrary angle in the medical image.

The first determination unit 522 classifies through a clustering technique, and determines through a preset valid patch and similarity analysis.

The second determination unit 523 analyzes and determines a preset effective patch and similar patterns through histogram analysis.

The third determination unit 524 analyzes and determines similarity for each patch using the similarity analysis function and the learned artificial neural network.

The validity analysis unit 525 rotates the medical image at a different angle, divides the image to a specified size, and generates a patch, and analyzes the validity through one of the first to third determination units.

After the basic analysis of the validity analysis unit, the basic operation termination unit 526 ends patch generation when the rotated angle of the medical image is 360 degrees.

The classification unit 527 further classifies the patches, which have been determined to be valid in the validity analysis, by the clustering technique after the termination operation of the basic operation termination unit.

The first image generator 528a generates a single image by connecting a plurality of images in the same cluster according to additional classification.

The second image generator 528b cuts one image generated by the image generator to a smaller size than the one image to generate a new size image.

The data augmentation unit 529 generates a patch while rotating the image generated in a new size from an angle to 360 degrees, and augments the learning data by analyzing the validity of any one of the above 2a to 2c.

The above-described first to third determination unit, validity analysis unit, classification unit, etc. may be made to learn with an artificial neural network. The artificial neural network learning apparatus including the learning unit may be mounted on a computing device or implemented to include a computing device.

The control unit 51 may correspond to a microprocessor, arithmetic processing unit or control unit, may include a central processing unit (CPU) or core, and executes a program or software module stored in the memory 52 It can be implemented to implement the method of this embodiment. For example, as illustrated in FIG. 3, a patch may be obtained from a specific region of interest of a pathology image and analyzed to enhance the data.

The communication unit 53 may include at least one sub-communication system for an intranet, the Internet, a wired network, a wireless network, a satellite network, or a combination thereof. The communication unit 53 may be connected to an MRI device or a database to receive or read a medical image signal, and may be implemented to transmit the processed result to a display device or an external device.

The artificial intelligence that can be employed in the above-described device may have a separate preprocessor, recognizer, learner, or a combination of independent functional units or components that are connected to the classification system, or may be mounted on each component. . Artificial intelligence can be designed to include machine learning, artificial neural networks, or deep learning. Deep learning may include a convolutional neural network. Such an artificial intelligence or artificial intelligence system including the same may be referred to as a learning means.

Although the present invention has been mainly described with reference to the embodiments with reference to the drawings, it is only an example, and it is obvious that various modifications and equivalent other embodiments are possible from those skilled in the art. Therefore, the true technical protection scope of the present invention should be defined by the technical spirit of the appended claims.

Claims (12)

A first step of generating a patch image by dividing the image into a specified size at an arbitrary angle in the medical image;
A second step of classifying through a clustering technique and determining through a preset valid patch and similarity analysis;
A second step of analyzing and determining a predetermined effective patch and a similar pattern through histogram analysis;
A 2c step of analyzing and determining similarity for each patch using the similarity analysis function and the learned artificial neural network; And
And a third step of rotating the medical image at a different angle, dividing the image to a specified size, generating a patch, and analyzing the effectiveness through one of the 2a, 2b, and 2c. How to increase. The method according to claim 1,
After the third step, when the rotated angle of the medical image is 360 degrees, ending the patch generation;
After the step of terminating, further classifying the patches that have been determined to be valid in the validity analysis using a clustering technique; And
Further comprising the step of generating a single image by connecting a plurality of images in the same cluster according to the further classification, the method of increasing learning data using patch matching. The method according to claim 2,
After the step of generating the one image,
And generating a new size image by cropping one image generated by the image generation unit to a smaller size than the one image. The method according to claim 3,
A method of increasing learning data using patch matching, further comprising the step of re-determining the validity by analyzing a ratio in which the image of the new size overlaps with the lesion area of the existing valid patch image. The method according to claim 3,
Further comprising a data augmentation unit for generating a patch while rotating the image generated in the new size to 360 degrees at any angle and analyze the effectiveness of any of the above 2a to 2c to enhance the learning data, patch matching How to increase learning data. The method according to claim 1,
Generating a non-valid patch obtained in the validity analysis by additionally transforming it into a normal distribution for the RGB channel of the pathological image; And
Further comprising the step of analyzing the validity again through the steps 2a to 2c for the patches generated by the additional conversion, and converting the invalid patch to a valid patch, further increasing learning data using patch matching Way. A patch image generator for dividing the image into a specified size at an arbitrary angle in the medical image to generate a patch image;
A first judging unit that classifies through a clustering technique and determines through a preset valid patch and similarity analysis;
A second determination unit that analyzes and determines a preset effective patch and similar patterns through histogram analysis;
A third determination unit that analyzes and determines similarity for each patch using a similarity analysis function and a learned artificial neural network; And
Rotating the medical image at a different angle and dividing the image to a specified size to generate a patch and including a validity analysis unit for analyzing the effectiveness through one of the first to third determination unit, increase learning data using patch matching Device. The method according to claim 7,
After the validity analysis of the validity analysis unit, when the rotation angle of the medical image is 360 degrees further comprising a basic operation termination unit to end the patch generation, learning data increase apparatus using a patch matching. The method according to claim 8,
A classification unit further classifying the patches, which have been judged to be valid in the validity analysis, into a clustering technique after the termination operation of the basic operation termination unit; And
Further comprising a first image generating unit for generating a single image by connecting a plurality of images in the same cluster according to the further classification, the apparatus for increasing learning data using patch matching. The method according to claim 9,
And a second image generator configured to crop one image generated by the image generator to a smaller size than the one image to generate a new size image. The method according to claim 10,
The validity analysis unit analyzes the ratio of the new size of the image overlapping the lesion area of the existing valid patch image to determine the validity again, the apparatus for increasing learning data using patch matching. The method according to claim 10,
Further comprising a data augmentation unit for generating a patch while rotating the image generated in the new size to 360 degrees at any angle and analyze the effectiveness of any of the above 2a to 2c to enhance the learning data, patch matching Device used to increase learning data.

Images