KR102221711B1 - X-ray INSPECTION SYSTEM BY X-ray IMAGE OPTIMIZATION - Google Patents

Abstract

According to an embodiment of the present invention
An optimized image conversion unit that converts the original X-ray image into another image; A deep learning program for receiving the original X-ray image and the image converted by the optimized image conversion unit-the converted X-ray image-and outputting a similarity of each; A determination unit that determines whether or not there is a disease in the original X-ray image by referring to similarities output by the deep learning program; Including,
The deep learning program calculates and outputs the similarity between the model and the original X-ray image and the similarity between the model and the converted X-ray image. The -ray reading system is started.

Description

X-ray inspection system with improved read rate through X-ray image optimization {X-ray INSPECTION SYSTEM BY X-ray IMAGE OPTIMIZATION}

The present invention relates to an X-ray reading system with improved read rates through X-ray image optimization.

Computerized databases can be used to store different types of data including voice, images and text. This flexibility allows designers to construct databases with data records that organize and store information in different formats, such as text and voice, to provide a database system that can be applied particularly in the near future.

Image identification technology is currently an important field and has been rapidly developing in recent years, and is being used in each field due to its wide use. For example, handwriting input, postal code identification, Chinese character identification, human face identification, fingerprint identification and iris identification, and other technologies have been very advanced.

On the other hand, a technology for analyzing diseases in X-ray images is being developed, and since this is a part directly connected to human life, technology for more accurate reading is always required.

(1) Korean Patent Registration No. KR 10-1710050 (2017.02.20) Image identification system and method (2) Korean Patent Registration No. 10-1690264 B1 (2016.12.21) Multiple code recognition method and system

According to one or more embodiments of the present invention, an X-ray reading system with improved read rates through X-ray image optimization is provided.

According to an embodiment of the present invention, in the X-ray reading system, the read rate is improved through X-ray image optimization,

An optimized image conversion unit for converting the original X-ray image into another image; A deep learning program for receiving the original X-ray image and the image converted by the optimized image conversion unit-the converted X-ray image-and outputting the similarity of each; A determination unit that determines whether or not there is a disease in the original X-ray image by referring to similarities output by the deep learning program; Computer processor; And a memory;

The deep learning program is loaded into the memory under the control of the computer processor to calculate and output a similarity between a model and the original X-ray image and a similarity between the model and the converted X-ray image. An X-ray reading system with improved read rates is provided through -ray image optimization.

According to one or more embodiments of the present invention, the read rate of X-ray images is improved.

1 is a view for explaining an X-ray reading system having an improved read rate through X-ray image optimization according to an embodiment of the present invention.
2 and 3 are diagrams for explaining an operation of generating a model used in the X-ray reading system 100 with improved read rates through X-ray image optimization according to an embodiment of the present invention.
FIG. 4 is a diagram for explaining an X-ray reading method in which a read rate is improved through X-ray image optimization according to an embodiment of the present invention.

The above objects, other objects, features, and advantages of the present invention will be easily understood through the following preferred embodiments related to the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided so that the disclosed content may be thorough and complete, and the spirit of the present invention may be sufficiently conveyed to those skilled in the art.

In various embodiments of the present specification, terms such as first and second are used to describe various elements, but these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. The embodiments described and illustrated herein also include complementary embodiments thereof.

The terms used in the present specification are for describing exemplary embodiments and are not intended to limit the present invention. In this specification, the singular form also includes the plural form unless specifically stated in the phrase. As used in the specification, "comprises" and/or "comprising" does not exclude the presence or addition of one or more other components.

Definition of Terms

In this specification, the term'software' refers to a technology that moves hardware in a computer, and the term'hardware' refers to a type of device or device (CPU, memory, input device, output device, peripheral device, etc.) constituting a computer. And, the term'step' means a series of processes or manipulations connected in a time series to achieve a predetermined order, the term'program means a set of instructions for processing by a computer, and the term'program recording medium' means It refers to a recording medium that can be read by a computer that records programs used to install, execute, or distribute programs.

In the present specification, the terms'image' and'image file' will be used without distinction unless there is a particular benefit of distinction.

In this specification, a machine learning program is a machine learning program that attempts a high level of abstraction (summarizing key content or functions in a large amount of data or complex data) through a combination of several nonlinear transformation methods. ) Means a collection of algorithms. Meanwhile, a deep learning program is a kind of machine learning program.

In the present specification, the term'model' refers to a classifier generated by training (learning) a machine learning program using image data.

In this specification, the term'image data' refers to a collection of digital data that can rasterize an image with a monitor or printer. An image consists of a unit called a pixel, and each pixel has a brightness value or a color value, and the image data may be, for example, a brightness value or color values of each pixel constituting an image. For a more specific example, in the case of a black-and-white image, each pixel may have a brightness value between 0 and 255, and in the case of a color image, each pixel uses three color channels (e.g., R, G, B). May be displayed, and each channel may have a brightness value between 0 and 255. Here, it will be understood by those skilled in the art that the numerical value or the type of color channel is exemplary for description of the present invention.

In the present specification, the term'management' is used as a meaning including'receiving','sending','storing','modifying', and'deleting' of data.

In the present specification, the same or similar sentence as'Step A and Step B have no precedence' means that step A is performed before step B, step A is performed after step B, or steps A and B are performed at the same time. Means you can be.

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

1 is a view for explaining an X-ray reading system with improved read rate through X-ray image optimization according to an embodiment of the present invention.

Referring to FIG. 1, an X-ray reading system (hereinafter referred to as'X-ray reading system') 100 having an improved read rate through X-ray image optimization according to an embodiment of the present invention is An image is provided from the X-ray image providing apparatus 100, and it is determined whether there is a disease in the X-ray image using a deep learning program. Here, the X-ray image providing apparatus 100 and the X-ray reading system 100 are directly connected to each other by wire or a communication network (e.g., a wide area communication network (WAN), a metropolitan area communication network (MAN), a local area network (LAN)). ), personal communication network (PAN)) to transmit and receive data.

The X-ray reading system 100 having an improved read rate through X-ray image optimization according to an embodiment of the present invention includes an optimized image conversion unit 201, a deep learning program 203, a determination unit 205, and an operating system. , A communication unit, a computer processor, a memory device (including a volatile and/or nonvolatile memory device), and other resources (software and hardware).

Here, the operating system (OS) may operatively connect hardware and an application program (eg, the optimized image conversion unit 201, the deep learning program 203, and the determination unit 205). The communication unit refers to a module composed of software and hardware for transmitting and receiving data with the outside.

Here, the computer processor, the memory device, the operating system, the optimized image conversion unit 201, the deep learning program 203, the determination unit 205, the communication unit, and other resources are operatively connected to each other.

The optimized image conversion unit 201, the deep learning program 203, and the determination unit 205 are each loaded into a storage device such as a memory under the control of a computer processor to perform their own operations.

When the optimized image conversion unit 201 receives an X-ray image from the X-ray image providing apparatus 100, the optimized image conversion unit 201 converts the provided X-ray image (hereinafter, referred to as “original X-ray image”) into an optimized form. An example of the form of optimization may be to clarify the boundaries of objects (eg, those with different brightness from the surroundings-such as cancerous tissues) included in the original X-ray image. Meanwhile, in the present specification, the optimized image conversion unit 201 converted X-ray image will be referred to as a'converted X-ray image'.

The optimized image conversion unit 201 may be performed by a method called Filtered back-projection (FBP), for example, in which the original X-ray image is converted. As another example in which the optimized image conversion unit 201 converts the original X-ray image, the optimized image conversion unit 201 obtains a projection image through a projection operation, and converts the obtained projection image to the detector direction. It can be transformed through filtering operation (filtartion) and back-projcttion of filtered projections (The 30th Workshop on Image Processing and Understanding 2018. 2. 7-2.9, Optimized Filter Based X using CNN) -ray CT image restoration method).

On the other hand, the contents described in the'X-ray CT image restoration method based on an optimization filter using CNN' (the 30th workshop on image processing and understanding 2018. 2. 7-2.9) are combined as part of the specification of the present application.

The deep learning program 203 receives the original X-ray image and the converted X-ray image individually, and outputs the result.

When the deep learning program 203 receives the original X-ray image, data representing the same or similarity with the model using the model (e.g., outputting '1' means exactly the same, and '0' If is output, it means complete dissimilarity, and a value between '0' and '1' indicates similarity according to the value).

When receiving the transformed X-ray image, the deep learning program 203 outputs data representing the same or similarity with the model (hereinafter, referred to as “similarity”) using the model.

That is, the deep learning program 203 receives the original X-ray image and the transformed X-ray image individually, and outputs the similarity with the model for each.

According to this embodiment, for one original X-ray image, the deep learning program 203 outputs at least two degrees of similarity.

The determination unit 205 determines whether or not there is a disease in the original X-ray image by referring to the two similarities.

For example, if the similarity of the original X-ray image and the similarity of the converted X-ray image are the same, the determination unit 205 determines that the similarity by the deep learning program 203 is reliable, and the similarity Make a judgment according to it.

For another example, the determination unit 205, although the similarity of the original X-ray image and the similarity of the converted X-ray image are different from each other, the different degree is not out of the error range (the error range is predetermined by the user). Otherwise, it is determined that the similarity calculated by the deep learning program 203 is reliable, and a determination is made according to the similarity (either of the similarity of the original X-ray image and the similarity of the converted X-ray image).

For another example, the determination unit 205 determines the similarity of the original X-ray image and the similarity of the converted X-ray image, and the different degree of the error range (the error range is predetermined by the user). Otherwise, it may be determined that the degree of similarity calculated by the deep learning program 203 is not reliable, and that the presence or absence of a disease cannot be determined.

Meanwhile, in the above-described embodiment, it has been described that the optimized image conversion unit 201 generates one converted X-ray image, but it may be possible to generate two or more optimized images for one original X-ray image. . That is, the optimized image conversion unit 201 may generate at least one or more converted X-ray images, and the deep learning program 203 provides a degree of similarity to the original X-ray image and the optimized image conversion unit 201. Similarity can be output for all of the at least one transformed X-ray image generated by. The determination unit 205 refers to all the similarities output from the deep learning program 203 to determine whether there is a disease.

2 and 3 are diagrams for explaining an operation of generating a model used in the X-ray reading system 100 with an improved read rate through X-ray image optimization according to an embodiment of the present invention.

The model generation device 400 includes an optimized image conversion unit 401, a deep learning program 403, and an operating system (not shown), a communication unit (not shown), a computer processor (not shown), and a storage device (volatile and/or Including non-volatile storage devices) (not shown), and other resources (software and hardware) (not shown).

Here, the operating system (OS: OPERATING SYSTEM) may operatively connect hardware and an application program (eg, an optimized image conversion unit, a deep learning program). The communication unit (not shown) refers to a module composed of software and hardware for transmitting and receiving data with the outside. Meanwhile, a computer processor (not shown) and a storage device (not shown), an operating system (not shown), an optimized image conversion unit 401, a deep learning program 403, a communication unit (not shown), and other resources (not shown) Are operatively linked to each other.

The model generation apparatus 400 generates a model by performing machine learning using a labeled X-ray image, and provides the generated model to the X-ray reading system 100 described with reference to FIG. 1.

Here, the label means data indicating whether there is a disease in the X-ray image.

2 and 3, referring to these drawings, a deep learning program (the deep learning program 203 described with reference to FIG. 1) is applied to a plurality of images (X-ray images) labeled as to whether or not there is a disease. ) Is trained, and the operation of generating a model for the X-ray image is performed by such learning. Hereinafter, the labeled X-ray image will be referred to as a'labeled X-ray image' for the purpose of explanation.

According to the present embodiment, upon receiving a label X-ray image, the optimized image conversion unit 401 converts the optimized X-ray image into an optimized X-ray image. An example of optimization may be to clarify the boundary of an object included in the original X-ray image (e.g., something with a different brightness from the surrounding-something like cancerous tissue). The optimization method of the optimized image conversion unit 401 should be the same as the conversion method of the optimized image conversion unit 201 in FIG. 1.

Meanwhile, the same label as the original X-ray image is attached to the optimized X-ray image. The deep learning program 403 receives an original labeled X-ray image and an optimized labeled X-ray image, respectively, and generates a model through machine learning.

For example, if the model generation device 400 receives 10,000 original X-ray images, the deep learning program generates a model by receiving 20,000 labeled X-ray images.

In the above-described embodiment, it has been described that the optimized image conversion unit 401 generates one converted X-ray image, but it may be possible to generate two or more optimized images for one original X-ray image. That is, the optimized image conversion unit 401 may generate at least one transformed X-ray image, and the deep learning program 403 uses both an original X-ray image and one or more transformed X-ray images. You can create a model.

Meanwhile, the model generation apparatus 400 may be configured separately from the X-ray reading system 100 of FIG. 1 or may be configured integrally with the X-ray reading system 100.

When the model generation device 400 and the X-ray reading system 100 are integrally configured, the optimized image conversion unit 201 and the deep learning program 203 included in the X-ray reading system 100, The functions of the optimized image conversion unit 401 and the deep learning program 403 of the model generation device 400 are also performed in parallel.

Meanwhile, the deep learning program 403 for generating a model and the deep learning program 203 for calculating the similarity are the same programs and may be neural network programs. According to an embodiment, the deep learning program may be a program implemented by a Long Short-Term Memory Network algorithm.

3 is a diagram for explaining a method of generating a model according to an embodiment of the present invention.

According to an embodiment, a neural network program consists of a plurality of nodes, and these nodes are operatively connected to each other. Hereinafter, for the purpose of explanation, operations performed at the node will be described with reference to FIG. 3.

Referring to FIG. 9, the weights (w1, w2, w3, ..., wn) are multiplied with respect to signals (x1, x2, x3, ..., xn) entering a node, and a transfer function (e.g. For example, an operation by weighted sum is performed, the operation result is used as an input to an activation function, and the output value of the activation function becomes an output of the node. Typically, the operating characteristics of each node are determined by the activation function. Here, the signals (x1, x2, x3, ..., xn) entering the node are at least some of the image data constituting one X-ray image.

As described above, each node of the neural network obtains the weighted sum of the input signal, and outputs a value obtained by inputting the weighted sum to the activation function to the outside of the node. Meanwhile, the shape of the neural network is determined according to how the nodes of the neural network are connected.

According to an embodiment, generating a model means determining weights for input signals input from each node in the above description, and comparing models means comparing weights.

That is, in each node, the transfer function and the activation function are determined, and the process of determining the weights is the process of creating the model. In a neural network program, the process of generating a model is described as an example.

First, the weight calculated for the signal input to each node constituting the neural network is initialized to an appropriate value.

Second, in the form of supervised learning learning data, {input, correct answer},'input' is input into a neural network to obtain an output value. The error is calculated by comparing this output value with the'correct answer' of the corresponding input.

Third, the weights of the neural network are adjusted to reduce the error.

FIG. 4 is a diagram for explaining an X-ray reading method in which a read rate is improved through X-ray image optimization according to an embodiment of the present invention.

The X-ray reading method according to the present embodiment may be implemented by an X-ray reading system having an improved read rate through X-ray image optimization described with reference to FIG. 1, for example.

Referring to FIG. 4, the X-ray reading method according to an embodiment of the present invention includes receiving an X-ray image (S101) and converting the X-ray image received in the step S101 into an optimized X-ray image. Transformation (S102), the deep learning algorithm using the model to calculate the similarity to the model (S103), and using the similarity calculated in the step S103 to determine the disease (S105). have.

In the step S101 of receiving an X-ray image, the X-ray reading system 200 receives an X-ray image from the X-ray image providing apparatus 100.

In the step of converting the X-ray image (S102), the optimized image conversion unit 201 converts at least one X-ray image (hereinafter, referred to as'original X-ray image') received as a result of performing step S101. This is the step of creating -ray image. As an example of optimization, it may be to clarify the boundary of an object (eg, an object having a different brightness from the surrounding-such as a cancerous tissue) included in the original X-ray image.

As another example of converting the original X-ray image, it may be performed by a method called Filtered back-projection (FBP). As another example in which the optimized image conversion unit 201 converts the original X-ray image, the optimized image conversion unit 201 obtains a projection image through a projection operation, and converts the obtained projection image to the detector direction. It can be transformed through filtering operation (filtartion) and back-projcttion of filtered projections (The 30th Workshop on Image Processing and Understanding 2018. 2. 7-2.9, Optimized Filter Based X using CNN) -ray CT image restoration method).

The step of calculating the similarity with the model (S103) is a step in which the deep learning program 203 calculates the similarity with the model for each of the original X-ray image and the converted X-ray image as a result of the execution of step S102. As a result of performing this step, at least two degrees of similarity are calculated.

In the step S105 of determining whether there is a disease, the determination unit 205 determines whether there is a disease in the original X-ray image by referring to at least two similarities.

In the step of determining whether there is a disease (S105), for example, if the similarity of the original X-ray image and the similarity of the converted X-ray image are the same, the deep learning program 203 It is determined that the degree of similarity is reliable, and a judgment is made according to the degree of similarity.

In the step of determining whether there is a disease (S105), for another example, the determination unit 205 has different degrees of similarity between the original X-ray image and the converted X-ray image, but the different degrees of the error range (error If the range is not outside the user's pre-determined range), it is determined that the similarity calculated by the deep learning program 203 is reliable, and the similarity (either the similarity of the original X-ray image or the similarity of the converted X-ray image) is determined. Make a judgment according to one).

In the step of determining whether there is a disease (S105), for another example, the determination unit 205, the similarity of the original X-ray image and the similarity of the converted X-ray image are different from each other, and the different degree is an error range ( If the error range is outside the user's pre-determined), it may be determined that the similarity calculated by the deep learning program 203 is not reliable, and thus the presence or absence of a disease cannot be determined.

As described above, the present invention is not limited to the described embodiments, and it is apparent to those of ordinary skill in the art that various modifications and variations can be made without departing from the spirit and scope of the present invention. Therefore, it should be said that such modifications or variations belong to the scope of the claims of the present invention.

100: X-ray reading system
201: Optimized image conversion unit
203: deep learning program
205: judgment unit

Claims (7)

In the X-ray reading system with improved read rate through X-ray image optimization,
An optimized image conversion unit that converts the original X-ray image into another image;
A deep learning program for receiving the original X-ray image and the image converted by the optimized image conversion unit-the converted X-ray image-and outputting a similarity of each;
A determination unit that determines whether or not there is a disease in the original X-ray image by referring to similarities output by the deep learning program;
Computer processor; And
Including;
The deep learning program is loaded into the memory under the control of the computer processor to calculate and output a degree of similarity between a model and the original X-ray image, and a degree of similarity between the model and the converted X-ray image,
The determination unit,
If the similarity of the original X-ray image and the similarity of the converted X-ray image are the same, it is determined that the similarity calculated by the deep learning program is reliable, and the determination is made according to the similarity X-ray reading system with improved read rate through ray image optimization. delete The method of claim 1,
The determination unit,
The similarity of the original X-ray image and the similarity of the converted X-ray image are different from each other, but if the different degree does not exceed the error range, it is determined that the similarity calculated by the deep learning program is reliable, and the deep learning program An X-ray reading system with improved read rates through X-ray image optimization, which is to judge according to the similarity of any one of the similarities calculated by. The method of claim 1,
The determination unit,
If the similarity of the original X-ray image and the similarity of the converted X-ray image are different from each other, and the different degree is out of the error range, it is determined that the similarity calculated by the deep learning program is not reliable, and the presence or absence of a disease is determined. X-ray reading system with improved read rate through X-ray image optimization, which judges that it cannot be determined. The method of claim 1,
A labeled X-ray image and a model generating device for generating the model using a converted X-ray image from the labeled X-ray image; further comprising, read rate through X-ray image optimization This improved X-ray reading system. The method of claim 1,
The deep learning program, also,
Label-Data indicating whether there is a disease in the X-ray image-The X-ray image with and the X-ray image with the label can be converted using an X-ray image to generate the model. The X-ray reading system with improved read rate through X-ray image optimization. The method of claim 6,
The optimized image conversion unit converts the labeled original X-ray image into another labeled image,
The label attached to the other image is the same as the label attached to the original X-ray image, wherein the read rate is improved through X-ray image optimization.

Images