KR20200023029A - Diagnosis assistance system and control method thereof - Google Patents

Abstract

The present invention relates to a diagnosis assistance system for assisting diagnosis of a plurality of diseases based on an eye fundus image. The diagnosis assistance system comprises: an eye fundus image acquisition part acquiring an eye fundus image; a first processing part acquiring a first result related to a first opinion of a subject by using a first neural network model for the eye fundus image; a second processing part acquiring a second result related to a second opinion of the subject by using a second neural network model for the eye fundus image; a third processing part determining diagnostic information for the subject based on the first result and the second result; and a diagnostic information output part providing a user with the determined diagnostic information.

Description

Diagnostic Assistance Systems and Control Methods {DIAGNOSIS ASSISTANCE SYSTEM AND CONTROL METHOD THEREOF}

The present invention relates to a diagnostic assistance system and a control method thereof, and more particularly to a diagnostic assistance system and a control method for providing diagnostic assistance information based on an image.

Fundus examination is a diagnostic aid that is frequently used in ophthalmology because it can observe the abnormalities of the retina, optic nerve, and macular area. Recently, the use of the fundus examination is increasing the use of non-invasive methods to observe the degree of vascular damage caused by chronic diseases such as hypertension, diabetes, as well as eye diseases.

On the other hand, with the rapid development of the recent deep learning technology, the development of diagnostic artificial intelligence in the field of medical diagnosis, especially image-based diagnostics is actively being made. Global companies such as Google and IBM are also investing in the development of artificial intelligence for analyzing various image medical data, such as inputting large-scale data in collaboration with medical field.In some companies, AI diagnosis that outputs excellent diagnosis results He also succeeded in developing the tool.

However, when trying to predict a plurality of values from one test data through the deep learning model, there is a problem that the accuracy of prediction is reduced and the processing speed is reduced. Accordingly, there is a need for a learning and diagnostic system for enabling accurate prediction of a plurality of diagnostic characteristics at a high data processing speed.

An object of the present invention is to provide a method of learning a neural network model for obtaining diagnostic assistance information from a fundus image.

Another object of the present invention is to provide a method of training a plurality of neural network models in parallel in order to obtain a plurality of diagnostic assistance information from a fundus image.

Another object of the present invention is to provide a method for rapidly obtaining a plurality of diagnostic assistance information from an ocular fundus image using a machine trained neural network model.

The problem to be solved by the present invention is not limited to the above-described problem, the objects that are not mentioned will be clearly understood by those skilled in the art from the present specification and the accompanying drawings. .

According to an aspect of the present invention, a diagnostic assistance system for assisting in the diagnosis of a plurality of diseases based on the fundus image, the fundus image acquisition unit for obtaining a target fundus image as a basis for obtaining diagnostic assistance information for a subject, Obtaining a first result associated with a first finding for the subject using a first neural network model for the subject fundus image, wherein the first neural network model is machine learned based on the first set of fundus images And a second neural network model for the target fundus image, wherein the second neural network model is machine-learned based on a second set of fundus images that are at least partially different from the first fundus image set. A second processing unit for obtaining a second result associated with a second finding, and based on the first result and the second result, Article may be a diagnostic supporting system including a diagnostics output for providing the determined diagnostic information provided to the third processing unit and the user to determine the diagnostic information. In this case, the first finding and the second finding may be used to diagnose different diseases.

According to another aspect of the present invention, a first training data set including a plurality of fundus images is obtained, the fundus image included in the first training data set is processed, and the first training data set is used. In addition, a learning apparatus for training the first neural network model may be provided.

A diagnostic apparatus for acquiring an object fundus image for acquiring diagnosis assistance information, and acquiring the diagnosis aid information based on the object fundus image using the learned first neural network model; A control method of the learning apparatus included in a system comprising: preprocessing the first fundus image such that a first fundus image included in the first training data set is converted into a form suitable for learning of the first neural network model. Performing the training, serializing the preprocessed first fundus image, and using the serialized first fundus image, training the first neural network model to classify the target fundus image into a first label or a second label. A control method of a learning apparatus including a step may be provided.

Means for solving the problems of the present invention are not limited to the above-described solutions, and the solutions not mentioned will be clearly understood by those skilled in the art from the present specification and the accompanying drawings. Could be.

According to the present invention, the rapid acquisition of diagnostic assistance information can be made based on the fundus image.

In addition, according to the present invention, it is possible to more accurately predict diagnostic assistance information associated with a plurality of diseases based on the fundus image.

In addition, according to the present invention, obtaining an optimized neural network model for predicting diagnostic assistance information associated with a plurality of diseases can be facilitated.

The effects of the present invention are not limited to the above-described effects, and effects that are not mentioned will be clearly understood by those skilled in the art from the present specification and the accompanying drawings.

1 illustrates a diagnostic assistance system according to an embodiment of the present invention.
2 is a block diagram illustrating a learning apparatus according to an embodiment of the present invention.
3 is a block diagram illustrating in more detail a learning apparatus according to another embodiment of the present invention.
4 is a block diagram illustrating a diagnostic apparatus according to an embodiment of the present invention.
5 is a view for explaining a diagnostic apparatus according to another embodiment of the present invention.
6 illustrates a diagnostic assistance system according to an embodiment of the present invention.
7 is a block diagram illustrating a client device according to an embodiment of the present invention.
8 is a view for explaining a diagnostic assistance process according to an embodiment of the present invention.
9 is a view for explaining the configuration of the learning unit according to an embodiment of the present invention.
10 is a conceptual diagram illustrating an image data set according to an embodiment of the present invention.
11 is a view for explaining image resizing according to an embodiment of the present invention.
12 is a diagram for explaining extension of an image data set according to an embodiment of the present invention.
13 is a block diagram illustrating a learning process of a neural network model according to an embodiment of the present invention.
14 is a block diagram illustrating a learning process of a neural network model according to an embodiment of the present invention.
15 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment.
16 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment.
17 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment.
18 is a view for explaining the configuration of the diagnostic unit according to an embodiment of the present invention.
19 is a diagram for describing diagnosis target data according to an embodiment of the present invention.
20 is a diagram illustrating a diagnosis process according to an embodiment of the present invention.
21 is a diagram for explaining a parallel diagnosis assistance system according to some embodiments of the present invention.
FIG. 22 is a diagram for describing a parallel diagnosis assistance system according to some exemplary embodiments.
FIG. 23 is a diagram for describing a configuration of a learning apparatus including a plurality of learning units according to an exemplary embodiment.
24 illustrates a parallel learning process according to an embodiment of the present invention.
25 illustrates a parallel learning process according to another embodiment of the present invention.
26 is a block diagram illustrating a diagnosis unit according to an embodiment of the present invention.
27 is a diagram for describing a diagnostic assistance process according to an embodiment of the present invention.
28 is a diagram for describing a diagnosis assistance system according to an embodiment of the present invention.
29 is a diagram for describing a graphic user interface according to one embodiment of the present invention.
30 is a diagram for explaining a graphic user interface according to one embodiment of the present invention.

The above objects, features and advantages of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. However, the present invention may be modified in various ways and may have various embodiments. Hereinafter, specific embodiments will be illustrated in the drawings and described in detail.

In the drawings, the thicknesses of layers and regions are exaggerated for clarity, and elements or layers are referred to as "on" or "on" of other components or layers. This includes both intervening other layers or other components in the middle as well as directly above other components or layers. Like numbers refer to like elements throughout. In addition, the components with the same functions within the scope of the same idea shown in the drawings of each embodiment will be described using the same reference numerals.

If it is determined that the detailed description of the known function or configuration related to the present invention may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. In addition, numbers (eg, first, second, etc.) used in the description of the present specification are merely identification symbols for distinguishing one component from another component.

In addition, the suffixes "module" and "unit" for components used in the following description are given or mixed in consideration of ease of specification, and do not have meanings or roles distinguished from each other.

The method according to the embodiment may be embodied in the form of program instructions that can be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the media may be those specially designed and constructed for the purposes of the embodiments, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

One. Diagnostic Assistant Systems and Processes

1.1 Purpose and definition

Hereinafter, a diagnosis assisting system and method for assisting the determination of the presence or absence of a disease or a disease or an abnormality that is the basis of the determination, etc. will be described based on the fundus image. In particular, a diagnostic assistance system or method for establishing a neural network model for diagnosing a disease using a deep learning technique and assisting detection of a disease presence or abnormality using the constructed model will be described.

According to an embodiment of the present invention, a diagnostic assistance system or method for acquiring diagnosis information related to the presence or absence of a disease or finding information used for the diagnosis of a disease, and the like, and assisting diagnosis using the fundus image is provided. Can be.

According to an embodiment of the present invention, a diagnostic assistance system or method may be provided to assist in the diagnosis of ocular disease based on the fundus image. For example, a diagnostic assistant system or method may be provided to assist diagnosis by acquiring diagnosis information related to the presence or absence of glaucoma, cataracts, macular degeneration, and prematurity retinopathy of the test subject.

According to another embodiment of the present invention, a diagnostic assistance system or method may be provided to assist in the diagnosis of other diseases (eg, systemic diseases or chronic diseases) that are not eye diseases. For example, a diagnostic assistance system or method may be provided that assists in diagnosis by acquiring diagnostic information of systemic diseases such as hypertension, diabetes, Alzheimer's, cytomegalovirus, stroke, heart disease, atherosclerosis, and the like.

According to another embodiment of the present invention, a diagnostic assistant system or method for detecting abnormal fundus findings that can be used for the diagnosis of eye diseases or other diseases can be provided. For example, color abnormalities across the fundus, lens opacities, cup-to-disc ratios (C / D ratio), macular abnormalities (e.g., macular pores), vessel diameter, running A diagnostic assistant system or method for acquiring findings such as abnormalities of the back, abnormal diameters of the retinal arteries, bleeding of the retina, the occurrence of exudate, and drusen may be provided.

In the present specification, the diagnostic assistance information may be understood to encompass diagnostic information or findings on which the diagnosis is based on the presence or absence of a disease.

1.2 Diagnostic Assist System Configuration

According to one embodiment of the invention, a diagnostic assistance system may be provided.

1 illustrates a diagnostic assistance system 10 according to an embodiment of the present invention. Referring to FIG. 1, the diagnostic assistance system 10 includes a learning apparatus 1000 for training a diagnostic model, a diagnostic apparatus 2000 for performing a diagnosis using the diagnostic model, and a client device 3000 for obtaining a diagnostic request. It may include. The diagnostic assistance system 10 may include a plurality of learning devices, a plurality of diagnostic devices, or a plurality of client devices.

The learning apparatus 1000 may include a learning unit 100. The learning unit 100 may perform training of the neural network model. For example, the learner 100 may acquire a fundus image data set and perform training of a neural network model for detecting a disease or abnormal finding from the fundus image.

The diagnosis apparatus 2000 may include a diagnosis unit 200. The diagnosis unit 200 may perform an acquisition of supplementary information used for diagnosing or diagnosing a disease using a neural network model. For example, the diagnosis unit 200 may acquire the diagnosis assistance information by using the diagnosis model trained by the learner.

The client device 3000 may include an imaging unit 300. The imaging unit 300 may capture an eye fundus image. The client device may be an ophthalmic fundus imaging device. Alternatively, the client device 3000 may be a handheld device, such as a smartphone or tablet PC.

In the diagnosis assistance system 10 according to the present embodiment, the learning apparatus 1000 determines a neural network model for use in diagnosis assistance by acquiring a data set and learning a neural network model, and the diagnostic apparatus requests information from a client. When the information is obtained, the diagnosis assistance information according to the diagnosis target image is obtained using the determined neural network model, and the client device may request information from the diagnosis apparatus and obtain the diagnosis assistance information transmitted in response thereto.

The diagnostic assistance system according to another embodiment may include a diagnostic device and a client device that perform training of a diagnostic model and diagnosis using the same. The diagnostic assistance system according to another embodiment may include a diagnostic apparatus for learning a diagnostic model, obtaining a diagnostic request, and performing a diagnosis. The diagnostic assistance system according to another embodiment may include a learning device for learning a diagnostic model and a diagnostic device for obtaining a diagnosis request and performing a diagnosis.

The diagnostic assistance system disclosed herein is not limited to the above-described embodiments, but includes a learning unit that performs model learning, a diagnostic unit that acquires diagnostic assistance information according to the learned model, and an imaging unit that acquires a diagnosis target image. It may be implemented in any form, including.

In the following, some embodiments of each device constituting the system will be described.

1.2.1 Learning device

The learning apparatus according to an embodiment of the present invention may perform training of a neural network model to assist diagnosis.

2 is a block diagram illustrating a learning apparatus 1000 according to an embodiment of the present invention. Referring to FIG. 2, the learning apparatus 1000 may include a controller 1200 and a memory unit 1100.

The learning apparatus 1000 may include a controller 1200. The controller 1200 may control an operation of the learning apparatus 1000.

The control unit 1200 may include one or more of a central processing unit (CPU), a random access memory (RAM), a graphics processing unit (GPU), one or more microprocessors, and other electronic components capable of processing input data according to predetermined logic. It may include.

The controller 1200 may read a system program and various processing programs stored in the memory unit 1100. For example, the controller 1200 may develop a data processing process, a diagnostic process, and the like for performing the diagnostic assistant, which will be described later, on the RAM, and perform various processes according to the deployed program. The controller 1200 may learn a neural network model to be described later.

The learning apparatus 1000 may include a memory unit 1100. The memory unit 1100 may store data necessary for learning and a learning model.

The memory unit 1100 may include a nonvolatile semiconductor memory, a hard disk, a flash memory, a RAM, a read only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), or a tangible nonvolatile recording medium. It can be implemented as.

The memory unit 1100 may store various processing programs, parameters for performing processing of the programs, or such processing result data. For example, the memory unit 1100 may include a data processing process program, a diagnostic process program, a parameter for performing each program, and data obtained by performing such a program (for example, processed data or diagnostics) for performing diagnostic assistant, which will be described later. Result value).

The learning apparatus 1000 may include a separate learning unit (or learning module). The learning unit may perform training of the neural network model. The performance of learning is described in more detail in Table 2 below.

The learner may be included in the controller 1200 described above. The learning unit may be stored in the memory unit 1100 described above. The learning unit may be implemented by some components of the controller 1200 and the memory unit 1100 described above. For example, the learning unit may be stored in the memory unit 1100 and driven by the controller 1200.

The learning apparatus 1000 may further include a communication unit 1300. The communication unit 1300 may communicate with an external device. For example, the communication unit 1300 may communicate with a diagnostic device, a server device, or a client device described later. The communication unit 1300 may perform wired or wireless communication. The communication unit 1300 may perform bi-directional or unidirectional communication.

3 is a block diagram illustrating the learning apparatus 1000 according to another embodiment of the present invention in more detail. Referring to FIG. 3, the learning apparatus 1000 may include a processor 1050, a volatile memory 1010, a nonvolatile memory 1030, a mass storage device 1070, and a communication interface 1090.

The processor 1050 of the learning apparatus 1000 may include a data processing module 1051 and a learning module 1053. The processor 1050 may process a data set stored in a mass storage device or a nonvolatile memory through the data processing module 1051. The processor 1050 may perform training of the diagnostic assistance neural network model through the learning module 1053. Processor 1050 may include local memory. The communication interface 1090 may be connected with the network 1110.

However, the learning apparatus 1000 illustrated in FIG. 3 is merely an example, and the configuration of the learning apparatus 1000 according to the present invention is not limited thereto. In particular, the data processing module or learning module may be provided at a location different from that shown in FIG. 3.

1.2.2 Diagnostic device

The diagnostic apparatus may acquire diagnostic assistance information using a neural network model.

4 is a block diagram illustrating a diagnostic apparatus 2000 according to an embodiment of the present invention. Referring to FIG. 4, the diagnostic apparatus 2000 may include a controller 2200 and a memory 2100.

The controller 2200 may generate diagnostic assistance information using the diagnostic assistance neural network model. The controller 2200 may acquire diagnostic data for diagnosis (eg, fundus data of a subject) and acquire diagnostic assistance information predicted by the diagnostic data using the trained diagnostic assistance neural network model.

The memory unit 2100 may store the learned diagnostic assistance neural network model. The memory unit 2100 may store parameters, variables, and the like of the diagnostic assistance neural network model.

The diagnostic apparatus 2000 may further include a communication unit 2300. The communicator 2300 may communicate with a learning device and / or a client device. For example, the diagnosis apparatus 2000 may be provided in the form of a server communicating with the client device. In this regard, it will be described in more detail below.

5 is a diagram for describing a diagnostic apparatus 2000 according to another exemplary embodiment. Referring to FIG. 5, the diagnostic apparatus 2000 according to an embodiment of the present invention may include a processor 2050, a volatile memory 2030, a nonvolatile memory 2010, a mass storage device 2070, and a communication interface 2090. It may include.

The processor 2050 of the diagnostic apparatus may include a data processing module 2051 and a diagnostic module 2053. The processor 2050 may process the diagnostic data through the data processing module 2051, and may obtain diagnostic assistance information according to the diagnostic data through the diagnostic module 2053.

1.2.3 Server devices

According to an embodiment of the present invention, the diagnostic assistance system may include a server device. The diagnostic assistance system according to an embodiment of the present invention may include a plurality of server devices.

The server device may store and / or drive a neural network model. The server device may store weight values constituting the trained neural network model. The server device may collect or store data used for diagnostic assistance.

The server device may output a result of the diagnosis assistance process using the neural network model to the client device. The server device may obtain feedback from the client device. The server device may operate similar to the diagnostic device described above.

6 illustrates a diagnostic assistance system 20 according to an embodiment of the present invention. Referring to FIG. 6, the diagnostic assistance system 20 according to an embodiment of the present invention may include a diagnostic server 4000, a learning device, and a client device.

The diagnostic server 4000, that is, the server device may communicate with a plurality of learning devices or a plurality of diagnostic devices. Referring to FIG. 6, the diagnostic server 4000 may communicate with the first learning apparatus 1000a and the second learning apparatus 1000b. Referring to FIG. 6, the diagnostic server 4000 may communicate with the first client device 3000a and the second client device 3000b.

For example, the diagnostic server 4000 trains the first learning apparatus 1000a for training the first diagnostic assistance neural network model for obtaining the first diagnostic assistance information and the second diagnostic assistance neural network model for acquiring the second diagnostic assistance information. The second learning apparatus 1000b may communicate.

The diagnostic server 4000 stores the first diagnostic assistance neural network model for obtaining the first diagnostic assistance information and the second diagnostic assistance neural network model for acquiring the second diagnostic assistance information, and the first client device 3000a or the second client device. In response to the request for obtaining the diagnostic assistance information from 3000b, the diagnostic assistance information may be obtained, and the obtained diagnostic assistance information may be transmitted to the first client device 3000a or the second client device 3000b.

Alternatively, the diagnostic server 4000 may communicate with the first client device 3000a requesting the first diagnostic assistance information and the second client device 3000b requesting the second diagnostic assistance information.

1.2.4 Client device

The client device may request diagnostic assistance information from the diagnostic device or the server device. The client device may acquire data necessary for diagnosis and transmit the acquired data to the diagnostic device.

The client device may include a data acquirer. The data acquirer may acquire data necessary for assistance of diagnosis. The data acquisition unit may be an imaging unit that acquires an image used for the diagnostic assistance model.

7 is a block diagram illustrating a client device 3000 according to an embodiment of the present invention. Referring to FIG. 7, the client device 3000 according to an exemplary embodiment may include an imaging unit 3100, a controller 3200, and a communication unit 3300.

The imaging unit 3100 may acquire image or image data. The imaging unit 3100 may acquire a fundus image. However, the client device 3000 may be replaced with a data acquisition unit other than the imaging unit 3100.

The communication unit 3300 may communicate with an external device, such as a diagnostic device or a server device. The communicator 3300 may perform wired or wireless communication.

The controller 3200 may control the imaging unit 3100 to acquire an image or data. The controller 3200 may control the image capturer 3100 to acquire a fundus image. The controller 3200 may transmit the acquired fundus image to the diagnosis apparatus. The controller may transmit the image acquired through the imaging unit 3100 to the server device through the communication unit 3300, and acquire the diagnostic assistance information generated based on the image.

Although not shown, the client device may further include an output unit. The output unit may include a display that outputs an image or an image or a speaker that outputs an audio. The output unit may output the image or the image data acquired by the obtained imaging unit. The output unit may output the diagnostic assistance information obtained from the diagnostic apparatus.

Although not shown, the client device may further include an input unit. The input unit may obtain a user input. For example, the input unit may obtain a user input for requesting the diagnostic assistance information. The input unit may obtain user information for evaluating the diagnostic assistance information obtained from the diagnostic apparatus.

In addition, although not shown, the client device may further include a memory unit. The memory unit may store an image acquired by the imaging unit.

1.3 Diagnostic Assistant Process Overview

The diagnostic assistant process may be performed by the diagnostic assistant system or the diagnostic assistant apparatus disclosed herein. The diagnostic assistant process can be considered as a largely divided into a training process for learning a diagnostic assistant model used for diagnosis assistance and a diagnostic process using the diagnostic assistant model.

8 is a view for explaining a diagnostic assistance process according to an embodiment of the present invention. Referring to FIG. 8, the diagnostic assistance process according to an embodiment of the present disclosure acquires and processes data (S110), learns a neural network model (S130), and obtains variables of the learned neural network model (S150). It may include a diagnostic assistant process to obtain the learning process and the diagnosis target data (S210) and to obtain diagnosis assistance information (S250) using the trained neural network model (S230) based on the diagnosis target data.

More specifically, the training process may include a data processing process of processing the input training image data into a state that can be used for training the model, and a training process of training the model using the processed data. The training process may be performed by the learning apparatus described above.

The diagnostic process may include a data processing process of processing the input inspection target image data to a state capable of performing a diagnosis using a neural network model, and a diagnostic process of performing diagnosis using the processed data. The diagnostic process may be performed by the above-described diagnostic apparatus or server apparatus.

Hereinafter, each process is demonstrated.

2. Training process

According to one embodiment of the invention, a process for training a neural network model may be provided. As a specific example, a process of training a neural network model to perform or assist with a diagnosis based on the fundus image can be disclosed.

The training process described below may be performed by the learning apparatus described above.

2.1 Learning Department

According to an embodiment of the present invention, the training process may be performed by the learning unit. The learning unit may be provided in the above-described learning apparatus.

9 is a view for explaining the configuration of the learning unit 100 according to an embodiment of the present invention. Referring to FIG. 9, the learner 100 may include a data processing module 110, a queue module 130, a learning module 150, and a learning result obtaining module 170. Each module may perform separate steps of the data processing process and the learning process as described below. However, not all of the components described in FIG. 9 and the functions performed by the elements are essential, and some elements may be added or some elements may be omitted depending on a learning form.

2.2 Data processing process

2.2.1 Image data acquisition

According to one embodiment of the invention, a data set may be obtained. According to one embodiment of the invention, the data processing module may obtain a data set.

The data set may be an image data set. Specifically, it may be a fundus image data set. The fundus image data set may be acquired using a general non-dongdong fundus camera or the like. The fundus image may be a panoramic image. The fundus image may be a red-free image. The fundus image may be an infrared photographed image. The fundus image may be an autofluorescence image. The image data may be obtained in any one of JPG, PNG, DCM (DICOM), BMP, GIF, and TIFF.

The data set may comprise a training data set. The data set may comprise a test data set. The data set may comprise a validation data set. In other words, the data set may be allocated to at least one data set of the training data set, the test data set, and the verification data set.

The data set may be determined in consideration of diagnostic assistance information to be obtained using a neural network model trained through the data set. For example, in order to train a neural network model for obtaining diagnostic assistance information related to cataracts, the data set obtained may be determined as an infrared fundus image data set. Alternatively, when the neural network model for acquiring diagnostic assistance information related to macular degeneration is to be trained, the acquired data set may be a self-fluoresced fundus image data set.

Individual data included in the data set may include a label. The label may be plural. In other words, individual data included in the data set may be labeled for at least one feature. In one example, the data set is a fundus image data set comprising a plurality of fundus image data, each fundus image data being a diagnostic information label (eg, the presence or absence of a specific disease) and / or finding information (eg, specific) according to the image. Abnormalities of the site) may include a label.

As another example, the data set is a fundus image data set and each fundus image data may include a peripheral information label for that image. For example, each fundus image data includes left and right eye information on whether the fundus image is an image of the left eye or right eye, gender information on whether the fundus image is a female or male fundus image, and a subject photographing the fundus image. Peripheral information labels, including age information about the age of may include.

10 is a conceptual diagram illustrating an image data set DS according to an embodiment of the present invention. Referring to FIG. 10, an image data set DS according to an embodiment of the present invention may include a plurality of image data IDs. Each image data ID may include an image I and a label L assigned to the image. Referring to FIG. 10, the image data set DS may include first image data ID1 and second image data ID2. The first image data ID1 may include a first image I1 and a first label L1 corresponding to the first image.

In FIG. 10, one image data includes one label. However, as described above, one image data may include a plurality of labels.

2.2.2 Image resizing

According to an embodiment of the present invention, the size of the acquired image data may be adjusted. That is, the images can be resized. According to an embodiment of the present invention, image resizing may be performed by the data processing module of the learning unit described above.

The size or aspect ratio of the image can be adjusted. The obtained plurality of images may be sized to have a constant size. Alternatively, the images can be scaled to have a constant aspect ratio. Resizing an image may be applying an image conversion filter to the image.

If the size or capacity of the individual images obtained is excessively large or small, the size or capacity of the image may be adjusted to convert to an appropriate size. Alternatively, if the sizes or capacities of the individual images vary, resizing may unify the sizes or capacities.

According to one embodiment, the capacity of the image may be adjusted. For example, when the capacity of an image exceeds an appropriate range, downsampling can reduce the image. Alternatively, when the capacity of the image falls short of an appropriate range, the image may be enlarged through upsampling or interpolating.

According to another embodiment, the size or aspect ratio of the image may be adjusted by cropping the image or adding pixels to the acquired image. For example, if an image contains an unnecessary part for learning, a part of the image may be cropped to remove it. Alternatively, if a portion of the image is cut off and the aspect ratio is not correct, a column or row may be added to adjust the image aspect ratio. In other words, the aspect ratio can be adjusted by adding margins or padding to the image.

According to another embodiment, the capacity and size or aspect ratio of the image may be adjusted together. For example, when the size of the image is large, the image size may be reduced by downsampling the image, and unnecessary portions included in the reduced image may be cropped and converted into appropriate image data.

In addition, according to another embodiment of the present invention, the orientation of the image data may be changed.

As a specific example, when a fundus image data set is used as the data set, each fundus image may be adjusted in size or resized. Cropping to remove the marginal portions other than the fundus portion of the fundus image or padding to adjust the aspect ratio by replenishing the cropped portion of the fundus image may be performed.

11 is a view for explaining image resizing according to an embodiment of the present invention. Referring to FIG. 11, an acquired fundus image may be resized by an image resizing process according to an embodiment of the present invention.

Specifically, the original fundus image (a) may be cropped (b) the margins unnecessary for obtaining diagnostic information (b) and reduced in size (c) to improve learning efficiency.

2.2.3 Image preprocessing

According to one embodiment of the invention, preprocessing of the image may be performed. If the image is used as it is input to the learning, overfitting may occur as a result of learning about unnecessary characteristics, and the learning efficiency may also be reduced.

In order to prevent this, it is possible to improve learning efficiency and performance by appropriately pre-processing and using the image data to meet the purpose of learning. For example, pretreatment may be performed on the fundus image to facilitate highlighting abnormalities of the ocular disease or pretreatment to highlight retinal vessel to blood flow changes.

The preprocessing of the image may be performed by the data processing module of the learning unit described above. The data processing module may acquire the resized image and perform preprocessing required for learning.

The preprocessing of the image may be performed on the image in which the above-mentioned resizing process is completed. However, the content of the present invention disclosed herein is not limited thereto, and the resizing process may be omitted and preprocessing may be performed on the image. Applying the preprocessing of the image may be to apply a preprocessing filter to the image.

According to an embodiment, a blur filter may be applied to the image. Gaussian filters may be applied to the image. Gaussian blur filters may be applied to the image. Alternatively, a deblur filter for sharpening an image may be applied to the image.

According to another embodiment, a filter for adjusting or modulating the color of the image may be applied. For example, a filter for changing the value of some of the RGB values constituting the image or binarizing the image may be applied.

According to another embodiment, a filter may be applied to emphasize specific elements in an image. For example, for fundus image data, preprocessing may be performed to highlight the vascular element from each image. At this time, the pretreatment for emphasizing the vascular element may be to apply one or more filters sequentially or in combination.

According to an embodiment of the present invention, the preprocessing of the image may be performed in consideration of the characteristics of the diagnostic assistance information to be obtained. For example, when it is desired to obtain diagnostic assistance information related to findings such as bleeding, drusen, microvascular flow, and exudate of the retina, pretreatment may be performed to convert the acquired fundus image into a red free fundus image.

2.2.4 Image augmentation

According to one embodiment of the invention, the image may be augmented or expanded. Augmentation of the image may be performed by the data processing module of the learning unit described above.

Augmented images can be used to improve the training performance of the neural network model. For example, when the amount of data for training the neural network model is insufficient, by modulating existing training image data to expand the number of data for training, and using the modulated (or changed) image together with the original image The number of training image data can be increased. As a result, overfitting can be suppressed, a layer of a model can be formed deeper, and prediction accuracy can be improved.

For example, the expansion of the image data may be performed by inverting the left and right sides of the image, clipping out a part of the image, correcting a color value of the image, or adding artificial noise. As a specific example, clipping a portion of an image may be performed by clipping some regions of an element constituting the image or randomly clipping some regions. For example, the image data can be extended by inverting left and right, upside down, rotating, scaling ratio resizing, cropping, padding, color adjustment, or brightness adjustment.

As an example, the augmentation or extension of the image data described above may generally be applied to a training data set. However, the present invention may be applied to other data sets, for example, test data sets, that is, data sets for testing the verified model using training and validation data using training data.

As a specific example, when a fundus image data set is used as the data set, one or more processes of inverting, cropping, adding noise, or changing color in order to increase the number of data are randomly applied. An augmented fundus image data set may be obtained.

12 is a diagram for explaining extension of an image data set according to an embodiment of the present invention. Referring to FIG. 12, an image according to embodiments of the present disclosure may be modified to improve prediction accuracy of a neural network model.

Specifically, referring to FIG. 12, an image according to an exemplary embodiment of the present invention may include a partial area dropout (a), a left and right inversion (b), a center point shifted (c, d), and a partial area of the image. Color may be modulated (e).

2.2.5 Image serialization

According to one embodiment of the invention, the image data may be linearized. The image may be serialized by the data processing module described above. The serialization module can serialize the preprocessed image data and pass it to the queue module.

If the image data is used as it is for training, decoding is necessary because the image data has an image file type such as JPG, PNB, or DCM. However, if the training is performed after decoding each time, the performance of model training may be degraded. . Accordingly, the learning can be performed by serializing the image file without using it for learning. Therefore, serialization of image data can be performed to improve learning performance and speed. The image data to be serialized may be image data to which at least one of the above-described image resizing and image preprocessing is applied, or both of the image data may not be processed.

Each image data included in the image data set may be converted into a string form. Image data may be converted into binary data form. In particular, the image data may be transformed into a data form suitable for use in training the neural network model. For example, the image data may be converted into a TFRecord form for use in neural network model training using tensorflow.

As a specific example, when the fundus image set is used as the data set, the acquired fundus image set may be converted into a TFRecord form and used for learning a neural network model.

2.2.6 Queue

Queues can be used to eliminate data bottlenecks. The queue module of the learning unit may store image data in a queue and transmit the image data to the learning model module.

In particular, when using the CPU (Central Processing Unit) and GPU (Graphic Processing Unit) together to conduct the learning process, the use of queues minimizes bottlenecks between the CPU and GPU, facilitates access to the database, and facilitates memory. It can improve the use efficiency.

The queue may store data used to train the neural network model. The cue can store image data. The image data stored in the queue may be image data processed or at least one of the above-described data processing processes (ie, resizing, preprocessing and augmentation) is processed.

The queue may store image data, preferably serialized image data as described above. The queue may store image data and supply image data to the neural network model. The queue may deliver image data in a batch size unit to a neural network model.

The queue may provide image data. The queue may provide data to a learning module described later. As data is extracted from the learning module, the number of data accumulated in the queue may be reduced.

If the number of data stored in the queue decreases below the reference level as the neural network model is trained, the queue may request replenishment of the data. The queue may request replenishment of certain kinds of data. When the learning unit is requested to supplement the data, the queue may supplement the data in the queue.

The queue may be provided in the system memory of the learning device. For example, the queue may be formed in a random access memory (RAM) of a central processing unit (CPU). In this case, the size of the queue, that is, the capacity may be determined according to the RAM capacity of the CPU. The queue may be a first in first out (FIFO) queue, a primary queue, or a random queue.

2.3 Learning process

According to one embodiment of the invention, the learning process of the neural network model can be initiated.

According to an embodiment of the present invention, the training of the neural network model may be performed by the above-described learning apparatus. The learning process may be performed by the controller of the learning apparatus. The learning process may be performed by the learning module described above.

13 is a block diagram illustrating a learning process of a neural network model according to an embodiment of the present invention. Referring to FIG. 13, in the training process of a neural network model according to an embodiment of the present invention, data is acquired (S1010), a neural network model is trained (S1030), a trained model is verified (S1050), and a learned model. This may be performed by acquiring a variable of S1070.

Hereinafter, referring to FIG. 13, some examples of the learning process of the neural network model will be described.

2.3.1 Data entry

A data set for training the diagnostic assistance neural network model can be obtained.

The data obtained may be an image data set processed by the data processing process described above. As an example, the data set may include serialized fundus image data after it has been sized, preprocessed filters applied, and data is augmented.

In the training phase of the neural network model, a training data set may be obtained and used. In the verification stage of the neural network model, a verification data set may be obtained and used. In the testing phase of the neural network model, a test data set may be obtained and used. Each data set may include a fundus image and a label.

The data set can be obtained from the queue. The data set may be obtained in batch size units from the queue. For example, if 60 are specified as the batch size, the data set may be extracted from the queue in 60 units. The size of the batch size may be limited by the RAM capacity of the GPU.

The data set can be randomly obtained from the queue into the learning module. The data set may be obtained in the order accumulated in the queue.

The learning module may specify and extract the configuration of the data set obtained from the queue. For example, the learning module may extract the fundus image having the left eye label and the fundus image data having the right eye label of a specific subject to be used together for learning.

The learning module may obtain a data set of a specific label from the queue. As an example, the learning module may obtain a fundus image data set having an abnormal diagnostic information label from a queue. The learning module may obtain a data set by specifying a ratio of the number of data according to the label from the queue. For example, the learning module may acquire a fundus image data set from the queue such that the number of fundus image data having an abnormal diagnostic information label and the number of fundus image data having a normal diagnostic information label are one to one.

2.3.2 Model design

The neural network model may be a diagnostic assistance model that outputs diagnostic assistance information based on the image data. The structure of the diagnostic assistance neural network model for obtaining diagnostic assistance information may have a predetermined shape. The neural network model may include a plurality of layers or layers.

The neural network model may be implemented in the form of a classifier that generates diagnostic assistance information. The classifier may perform double classification or multiple classification. For example, the neural network model may be a binary classification model that classifies input data into a normal or abnormal class for target diagnosis assistance information such as a specific disease or abnormality. Alternatively, the neural network model may be a multiple classification model that classifies input data into a plurality of class classes for particular characteristics (eg, degree of disease progression). Alternatively, the neural network model may be implemented as a regression type model that outputs a specific value related to a specific disease.

The neural network model may include a convolutional neural network (CNN). Can be. As the CNN structure, at least one of AlexNet, LENET, NIN, VGGNet, ResNet, WideResnet, GoogleNet, FractaNet, DenseNet, FitNet, RitResNet, HighwayNet, MobileNet, and DeeplySupervisedNet may be used. The neural network model may be implemented using a plurality of CNN structures.

As an example, the neural network model may be implemented to include a plurality of VGGNet blocks. More specifically, the neural network model includes a CNN layer having 64 filters of 3x3 size, a batch normalization (BN) layer and a ReLu layer, and a CNN layer having a 128x filter of 3x3 size, a ReLu layer, The second block, in which the BN layers are sequentially combined, may be provided in combination.

The neural network model may include a max pooling layer, followed by a CNN block, and an end may include a global average pooling layer, a fully connected layer, and an activation layer (eg, sigmoid, softmax, etc.). have.

2.3.3 Model learning

Neural network models can be trained using training data sets.

Neural network models can be trained using labeled data sets. However, the learning process of the diagnostic assisted neural network model described herein is not limited thereto, and the neural network model may be trained in an unsupervised form using unlabeled data.

The training of the neural network model, based on the training image data, obtains a result using a neural network model to which arbitrary weight values are assigned, compares the obtained result with a label value of the training data and back propagates according to the error. By optimizing the weight values. In addition, the training of the neural network model may be influenced from feedback from the verification results, test results, and / or diagnostic steps of the models described below.

Training of the neural network model described above may be performed using TensorFlow. However, the present invention is not limited thereto, and frameworks such as Theano, Keras, Caffe, Torch, and CNTK (Microsoft Cognitive Toolkit) are used for learning neural network models. May be

2.3.4 Validation of the model

The neural network model can be verified using the validation data set. The verification of the neural network model may be performed by obtaining a result value for the verification data set from the neural network model in which the learning is performed, and comparing the result value with a label of the verification data set. Verification can be performed by measuring the accuracy of the results. Depending on the verification result, parameters of the neural network model (eg, weights and / or biases) or hyper parameters (eg, learning rates) may be adjusted.

For example, the learning apparatus according to an embodiment of the present invention may learn a neural network model predicting diagnosis assistance information based on the fundus image, and correspond to the diagnosis fundus image by diagnosing the assistance information of the verification fundus image of the trained model. Validation of the diagnostic assistance neural network model may be performed by comparison with the validation label.

For verification of the neural network model, a separate verification set, i.e., a data set having distinct factors not included in the training data set, may be used. For example, the separate verification set may be a data set in which the training data set and the factors such as race, environment, age, and gender are distinguished.

2.3.5 Model testing

Neural network models can be tested using test data sets.

Although not shown in FIG. 13, according to an exemplary embodiment of the present invention, the neural network model may be tested using a test data set that is distinguished from a training data set and a verification data set. Depending on the test results, the parameters (eg, weights and / or biases) or hyperparameters (eg, learning rates) of the neural network model may be adjusted.

For example, the learning apparatus according to an embodiment of the present invention obtains a result value of inputting test fundus image data that is not used for training and verification, from a neural network model trained to predict diagnostic assistance information based on the fundus image. Thus, a test of the trained and verified diagnostic auxiliary neural network model may be performed.

For the testing of the neural network model, a separately prepared external data set, i.e., a data set having factors distinct from training and / or validation data, may be used.

2.3.6 Output of the result

As a result of the training of the neural network model, parameter values of the optimized model may be obtained. As described above, as the model is repeatedly trained using the test data set, more appropriate parameter (or variable) values may be obtained. If the learning proceeds sufficiently, an optimized value of weight and / or bias can be obtained.

According to one embodiment of the invention, the learned neural network model and / or parameters or variables of the learned neural network model may be stored in the learning apparatus and / or the diagnostic apparatus (or the server). The trained neural network model may be used to predict diagnostic assistance information by a diagnostic device and / or a client device. In addition, the parameters or variables of the learned neural network model may be updated by feedback obtained from the diagnostic apparatus or the client apparatus.

2.3.7 Model ensemble

According to an embodiment of the present invention, a plurality of sub-models may be learned at the same time in the course of learning one diagnostic auxiliary neural network model. The plurality of submodels may have different hierarchical structures.

In this case, the diagnostic assistance neural network model according to an embodiment of the present invention may be implemented by combining a plurality of sub-neural network models. In other words, the neural network model may be trained using an ensemble technique that combines a plurality of sub-neural networks.

When a diagnostic auxiliary neural network model is formed by forming an ensemble, prediction may be performed by combining the results predicted from various types of sub-neural network models, thereby improving accuracy of the result prediction.

14 is a block diagram illustrating a learning process of a neural network model according to an embodiment of the present invention. Referring to FIG. 14, in the training process of the neural network model according to an embodiment of the present invention, a data set is obtained (S1011) and a first model (ie, a first neural network model) and a second are obtained using the acquired data. The model (ie, the second neural network model) is trained (S1031, S1033), the trained first neural network model and the second neural network model are verified (S1051), the final neural network model is determined, and its parameters or variables are obtained (S10721). )can do.

Hereinafter, referring to FIG. 14, some examples of the learning process of the neural network model will be described.

According to an embodiment of the present invention, the plurality of sub-neural network models may acquire the same training data set and generate output values separately. In this case, the ensemble of the plurality of sub-neural network models may be determined as the final neural network model, and parameter values for each of the plurality of sub-neural network models may be obtained as a learning result. The output value of the final neural network model may be determined as an average value of the output values by each sub-neural network model. Alternatively, the output value of the final neural network model may be determined as a weighted average value for the output value of each sub-neural network model in consideration of the accuracy obtained as a result of verification for each sub-neural network model.

More specifically, when the neural network model includes a first sub-neural network model and a second sub-neural network model, the optimized parameter value for the first sub-neural network model and the optimized parameter value of the second sub-neural network model by machine learning Can be obtained. In this case, an average value of output values (eg, probability values for specific diagnostic assistance information) obtained from the first sub-neural network model and the second sub-neural network model may be determined as the output value of the final neural network model.

According to another embodiment of the present invention, the accuracy of the individual sub-neural network model may be evaluated based on output values of each of the plurality of sub-neural network models. In this case, one of the plurality of sub-neural network models may be selected and determined as the final sub-neural network model based on the accuracy. The structure of the determined sub neural network model and the parameter values of the determined sub neural network model obtained as a result of the training may be stored.

As a more specific example, when the neural network model includes a first sub-neural network model and a second sub-neural network model, accuracy is obtained according to each of the first and second sub-neural network models, and the final sub-neural network model is more accurate. This can be determined by a neural network model.

According to another embodiment of the present invention, combining at least one sub-neural network among a plurality of neural network models, forming an ensemble of at least one sub-neural network model combined, and evaluating each ensemble, The combination of sub-neural network models forming a highly accurate ensemble can be determined as the final neural network model. At this time, the ensemble may be performed for all possible cases of selecting at least one of the plurality of sub-neural network models, and the sub-neural network combination evaluated as the highest accuracy may be determined as the final neural network model.

As a more specific example, when the neural network model includes a first sub-neural network model and a second sub-neural network model, the accuracy of the first sub-neural network model, the accuracy of the second sub-neural network model and the ensemble of the first and second sub-neural network models By comparing the accuracy by and the sub-neural network model configuration of the most accurate case can be determined as the final neural network model.

2.4 Embodiment 1-Control Method of a Learning Device.

15 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment.

Referring to FIG. 15, in the method of controlling a learning apparatus according to an embodiment of the present disclosure, performing preprocessing of the first fundus image (S110), serializing the preprocessed first fundus image (S130), and first It may include the step (S150) for training the neural network model.

A control method of a learning apparatus according to an embodiment of the present invention obtains a first training data set including a plurality of fundus images, processes the fundus image included in the first training data set, and the first training. A training apparatus for training a first neural network model, a target fundus image for obtaining diagnostic assistance information, using the data set, and the diagnosis assistance based on the target fundus image using the trained first neural network model. The method may be a control method of a learning apparatus included in a system including a diagnostic apparatus for obtaining information.

Performing preprocessing of the first fundus image (S110) may perform preprocessing of the first fundus image such that the first fundus image included in the first training data set is converted into a form suitable for learning of the first neural network model. It may further include doing.

The control method of the learning apparatus according to an embodiment of the present invention may include serializing the preprocessed first fundus image (S130). The first fundus image may be serialized in a form that is easy for learning of the neural network model.

In this case, the training of the first neural network model (S150) further includes training the first neural network model classifying the target fundus image into a first label or a second label by using the serialized first fundus image. can do.

The learning apparatus may obtain a second training data set including a plurality of fundus images and at least partially different from the first training data set, and train the second neural network model using the second training data set.

According to an embodiment of the present invention, a method of controlling a learning apparatus may include performing preprocessing of a second fundus image such that a second fundus image included in a second training set is suitable for learning a second neural network model. Serializing the two fundus images and using the serialized second fundus image, training a second neural network model that classifies the subject fundus image into a third label or a fourth label.

16 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment. Referring to FIG. 16, in the method of controlling a learning apparatus according to an embodiment of the present disclosure, performing preprocessing of the second fundus image (S210), serializing the preprocessed second fundus image (S230), and second Training of the neural network model (S250) may be included.

In FIG. 16, for convenience of description, preprocessing of the first fundus image, serialization of the first fundus image, and learning using the first fundus image, followed by preprocessing of the first fundus image, serialization of the second fundus image, and first fundus image Although described as being performed following learning with, the content of the present invention is not limited thereto.

Preprocessing the second fundus image included in the second training set, and serializing the second fundus image and learning using the second fundus image, include the above-described preprocessing of the first fundus image, serialization of the first fundus image and first fundus. It can be performed independently of learning using images. Preprocessing the second fundus image included in the second training set, and serializing the second fundus image and learning using the second fundus image, include the above-described preprocessing of the first fundus image, serialization of the first fundus image and first fundus. It can be performed in parallel with learning using images. In other words, the pre-processing of the second fundus image included in the second training set, and the serialization of the second fundus image and the learning using the second fundus image, the pre-processing of the first fundus image described above, serialization of the first fundus image must be performed. And after the learning using the first fundus image, or not previously performed, the processing on the first fundus image and the processing on the second fundus image may be performed without mutual dependence.

The first preprocessing performed on the fundus image included in the first training data set may be distinguished from the second preprocessing performed on the fundus image included in the second training data set. For example, the first pretreatment may be vascular emphasis pretreatment, and the second pretreatment may be color modulation pretreatment. Each preprocessing may be determined in consideration of diagnostic assistance information to be obtained through each neural network model.

The control method of the learning apparatus according to an embodiment of the present invention may evaluate an accuracy of the trained first neural network model by using a first verification data set that is at least partially distinguished from a first training data set. Validating the first neural network model and evaluating the accuracy of the trained first neural network model using a second verification data set that is at least partially distinct from a second training data set, thereby evaluating the second neural network model It may further comprise the step of verifying. At this time, the verification of the first neural network model and the second neural network model may be performed independently.

The serialized first fundus image is sequentially stored in a first cue, and the serialized fundus image stored in the first cue is used for learning the first neural network model in a predetermined capacity unit from the first cue, and serialized. A second fundus image is sequentially stored in a second cue that is distinct from the first cue, and the serialized fundus image stored in the second cue is subjected to learning of the first neural network model in a predetermined capacity unit from the second cue. Can be used.

The first neural network model may include a first sub neural network model and a second sub neural network model. In this case, classifying an object fundus image into the first label or the second label may be performed by considering a first prediction value predicted by the first sub-neural network model and a second prediction value predicted by the second sub-neural network model. Can be performed.

The second neural network model may include a third sub neural network model and a fourth sub neural network model. In this case, classifying the target fundus image into the third label or the fourth label may be performed by considering a third prediction value predicted by the third sub-neural network model and a fourth prediction value predicted by the fourth sub-neural network model. Can be performed.

The first training data set may include at least a portion of the fundus image labeled with the first label, and the second training data set may include at least a portion of the fundus image labeled with the third label. In this case, the fundus image labeled with the first label and the at least some fundus image labeled with the third label may be partially common.

The first label is a normal label indicating that the subject corresponding to the subject fundus image is normal with respect to the first finding, and the second label is an abnormal label indicating that the subject is abnormal with respect to the second finding. Can be.

Performing preprocessing of the first fundus image may include cropping the first fundus image to meet a reference aspect ratio and changing the size of the first fundus image.

The pre-processing of the first fundus image may further include applying a blood vessel enhancement filter to the fundus image such that the processing unit highlights blood vessels included in the first fundus image.

The serialized first fundus image may be sequentially stored in a queue, and the serialized first fundus image stored in the queue may be used for learning the first neural network model in a predetermined capacity unit from the queue. The cue may request replenishment of the serialized first fundus image if the serialized first fundus image not used for the first learning is reduced below a reference dose.

The first finding may be any one of retinal bleeding findings, retinal exudation findings, lens turbidity findings, and diabetic retinopathy findings.

17 is a diagram for describing a method of controlling a learning apparatus, according to an exemplary embodiment.

Referring to FIG. 17, the control method of the learning apparatus according to an embodiment of the present invention may further include verifying the first neural network model (S170) and updating the first neural network model (S190).

The step S170 of validating the first neural network model may be performed by evaluating the accuracy of the trained first neural network model by using a first verification data set that is at least partially distinguished from the first training data set. The method may further include verifying.

The method may further include updating the first neural network model by reflecting the verification result obtained from updating the first neural network model (S190) and validating the first neural network model (S170).

Meanwhile, the first neural network model may include a first sub neural network model and a second sub neural network model. The training of the first neural network model may include verifying the first sub-neural network model using a first verification data set to obtain an accuracy of the first sub-neural network model, and using the first verification data set. The second sub-neural network model is verified to obtain an accuracy of the second sub-neural network model, and the accuracy of the first sub-neural network model and the accuracy of the second sub-neural network model are compared to make a more accurate sub-neural network model as the final neural network model. May include determining.

3. Diagnostic Assistant Process

According to an embodiment of the present invention, a diagnostic assistance process (or diagnostic process) for acquiring diagnostic assistance information using a neural network model may be provided. As a specific example, by the diagnostic assistant process, the diagnostic assistant information (eg, diagnostic information or finding information) may be predicted using the fundus image and the trained diagnostic assistant neural network model.

The diagnostic assistance process described below may be performed by a diagnostic apparatus.

3.1 Diagnostic department

According to an embodiment of the present invention, the diagnostic process may be performed by the diagnosis unit 200. The diagnosis unit 200 may be provided in the above-described diagnosis apparatus.

18 is a view for explaining the configuration of the diagnostic unit 200 according to an embodiment of the present invention. Referring to FIG. 18, the diagnosis unit 200 may include a diagnosis request obtaining module 210, a data processing module 230, a diagnosis module 250, and an output module 270.

Each module may perform separate steps of the data processing process and the learning process as described below. However, not all of the components described in FIG. 16 and the functions performed by each of the elements are essential, and some elements may be added or some elements may be omitted depending on the aspect of diagnosis.

3.2 Data Acquisition and Diagnostic Requests

The diagnostic apparatus according to an embodiment of the present invention may acquire diagnostic target data and obtain diagnostic assistance information based on the diagnostic target data. The diagnosis target data may be image data. The data acquisition and the acquisition of the diagnosis request may be performed by the above-described diagnosis request acquisition module of the diagnosis unit.

19 is a diagram for describing diagnosis target data TD according to an example of the present invention. Referring to FIG. 19, the diagnosis target data TD may include a diagnosis target image TI and diagnosis object information PI.

The diagnosis target image TI may be an image for obtaining diagnosis assistance information on the diagnosis target object. For example, the diagnosis target image may be a fundus image. The diagnostic target (TI) JPG, PNG, DCM (DICOM), BMP, GIF, TIFF may have any one of the formats.

The diagnostic object information PI may be information for identifying a diagnosis object. Alternatively, the diagnostic object information PI may be characteristic information of a diagnostic object or an image. For example, the diagnostic object information PI may include information such as the date and time of imaging the diagnosis target image, the imaging equipment, the identification number of the subject to be diagnosed, the ID, the name, the gender, the age, or the weight. When the diagnosis target image is an ocular fundus image, the diagnosis object information PI may further include eye related information such as binocular information indicating whether the left eye or the right eye is present.

The diagnostic device may obtain a diagnostic request. The diagnostic apparatus may acquire the diagnosis target data together with the diagnosis request. When the diagnostic request is obtained, the diagnostic apparatus may obtain diagnostic assistance information using the trained diagnostic assistance neural network model. The diagnostic device may obtain a diagnostic request from the client device. Alternatively, the diagnostic apparatus may obtain a diagnostic request from the user through an input means provided separately.

3.3 Data processing process

The obtained data can be processed. Processing of the data may be performed by the data processing module of the above-described diagnostic unit.

The data processing process may generally be performed similarly to the data processing process in the aforementioned learning process. Hereinafter, the data processing process in a diagnostic process is demonstrated centering on the difference with the data processing process in a learning process.

In the diagnostic process, the diagnostic device may acquire data as in the learning process. In this case, the obtained data may be in the same format as the data obtained in the learning process. For example, in the learning process, when the learning apparatus trains the diagnostic assistance neural network model using the image data in the DCM format, the diagnostic apparatus may acquire the DCM image and obtain the diagnosis assistance information using the trained neural network model. .

In the diagnostic process, the acquired diagnosis subject image may be resized, similar to the image data used in the learning process. The image to be diagnosed may be shaped to have an appropriate dose, size, and / or aspect ratio in order to efficiently perform diagnostic assistance information prediction through the trained diagnostic assistance neural network model.

For example, when the diagnosis target image is a fundus image, resizing such as cropping an unnecessary portion of the image or reducing the size thereof may be performed to predict diagnosis information based on the fundus image.

In the diagnosis process, a preprocessing filter may be applied to the obtained diagnosis object image, similar to the image data used in the learning process. An appropriate filter may be applied to the object to be diagnosed so that the accuracy of predicting the diagnosis assistance information through the trained diagnosis assistance neural network model is further improved.

For example, when the image to be diagnosed is a fundus image, preprocessing for facilitating prediction of positive diagnostic information, for example, image preprocessing for highlighting blood vessels, or image preprocessing for highlighting or weakening a specific color, may be applied to the image to be diagnosed. have.

In the diagnostic process, the acquired diagnostic subject image may be serialized, similar to the image data used in the learning process. The image to be diagnosed can be converted or serialized into a form that can easily drive a diagnostic model in a specific workframe.

Serialization of the diagnosis target image may be omitted. This may be because, unlike in the learning stage, the processor does not have a large number of data processed at a time in the diagnostic stage, so that the burden on the data processing speed is relatively small.

In the diagnosis process, the obtained diagnosis subject image may be stored in a queue, similar to the image data used in the learning process. However, in the diagnostic process, since the number of processing data is smaller than that of the learning process, the step of storing the data in the queue may be omitted.

On the other hand, in the diagnostic process, since an increase in the number of data is not required, it may be desirable not to use the data enhancement or image enhancement procedure unlike the learning process in order to obtain accurate diagnosis assistance information.

3.4 Diagnostic process

According to one embodiment of the invention, a diagnostic process using a trained neural network model may be initiated. The diagnostic process may be performed in the above-described diagnostic apparatus. The diagnostic process can be performed at the aforementioned diagnostic server. The diagnostic process may be performed by the controller of the above-described diagnostic apparatus. The diagnostic process may be performed by the diagnostic module of the above-described diagnostic unit.

20 is a diagram illustrating a diagnosis process according to an embodiment of the present invention. Referring to FIG. 20, the diagnosis process may be performed by acquiring diagnosis target data (S2010), using a learned neural network model (S2030), and obtaining a result corresponding to the acquired diagnosis target data (S2050). However, the processing of the data may be selectively performed.

Hereinafter, each step of the diagnostic process will be described with reference to FIG. 20.

3.4.1 Data entry

According to an embodiment of the present invention, the diagnostic module may acquire diagnostic target data. The obtained data may be processed data as described above. For example, the acquired data may be fundus image data of a subject to which the pre-processing to adjust the size and emphasize the blood vessel is applied. According to an embodiment of the present invention, the left eye image and the right eye image of one subject may be input together as diagnosis target data.

3.4.2 Data classification

The diagnostic assistive neural network model provided in the classifier form may classify the input diagnosis target image into a positive or negative class for a predetermined label.

The trained diagnostic assistance neural network model may receive diagnostic target data and output a predicted label. The trained diagnostic assistance neural network model may output a predicted value of the diagnostic assistance information. Diagnostic assistance information may be obtained using the trained diagnostic assistance neural network model. Diagnostic assistance information may be determined based on the predicted label.

For example, the diagnostic assisted neural network model may predict diagnostic information (ie, information about the presence of a disease) or finding information (ie, information about the presence or absence of abnormalities) of an eye disease or systemic disease of a subject. In this case, the diagnostic information or findings information may be output in a probability form. For example, the probability that the subject has a specific disease or the probability that there is a specific abnormality of the fundus image of the subject may be output. In the case of using the diagnostic assistance neural network model provided in the classifier form, the predicted label may be determined in consideration of whether the output probability value (or prediction score) exceeds the threshold value.

As a specific example, the diagnosis-assisted neural network model may output the presence or absence of diabetic retinopathy of a subject as a probability value using the fundus photograph of the subject as a diagnosis target image. In the case of using a diagnostic assistive neural network model of a classifier type having a normal value of 1, a fundus photograph of a subject is input to the diagnostic assistance neural network model, and a probability value of normal: abnormal is determined as to whether the subject has diabetic retinopathy. It can be obtained in the form of.

Herein, the data is classified based on a classifier-type diagnostic auxiliary neural network model. However, the present invention is not limited thereto, and specific diagnostic auxiliary values may be defined using the diagnostic auxiliary neural network model implemented as a regression model. For example, blood pressure).

According to another embodiment of the present invention, suitability information of an image may be obtained. The suitability information may indicate whether the diagnosis target image is suitable for acquiring the diagnosis assistance information using the diagnosis assistance neural network model.

The suitability information of the image may be quality information of the image. The quality information or suitability information may indicate whether the diagnosis target image reaches the reference level.

For example, when the diagnosis target image has a defect due to a defect of the photographing equipment or the influence of illumination during shooting, an inappropriate result may be output as the suitability information with respect to the diagnosis target image. If the diagnosis target image includes noise at least a certain level, the diagnosis target image may be determined to be inappropriate.

The suitability information may be a value predicted using a neural network model. Alternatively, the suitability information may be information obtained through a separate image analysis process.

According to an embodiment, even when an image is classified as inappropriate, diagnostic assistance information acquired based on the inappropriate image may be obtained.

According to one embodiment, the image classified as inappropriate may be reviewed by a diagnostic aid neural network model.

In this case, the diagnostic assistance neural network model for performing the review may be different from the diagnostic assistance neural network model for performing the initial review. For example, the diagnostic apparatus may store the first diagnostic auxiliary neural network model and the second diagnostic auxiliary neural network model, and an image classified as inappropriate through the first diagnostic auxiliary neural network model may be reviewed through the second diagnostic auxiliary neural network model. .

According to another embodiment of the present invention, a Class Activation Map (CAM) may be obtained from a trained neural network model. Diagnostic assistant information may include CAM. The CAM can be obtained along with other diagnostic assistance information.

In the case of a CAM, it may optionally be obtained. For example, in the case of CAM, when the diagnostic information or finding information obtained by the diagnostic assistance model is classified into an abnormal class, the CAM may be extracted and / or output.

3.5 Output of diagnostic assistant information

Diagnostic assistance information may be determined based on labels predicted from the diagnostic assistance neural network model.

The output of the diagnostic assistance information may be performed by the above-described output module of the diagnostic unit. The diagnostic assistance information may be output from the diagnostic apparatus to the client apparatus. The diagnostic assistance information may be output from the diagnostic apparatus to the server apparatus. The diagnostic assistance information may be stored in a diagnostic device or a diagnostic server. The diagnostic assistance information may be stored in a server device or the like provided separately.

Diagnostic assistant information can be managed in a database. For example, the acquired diagnostic assistance information may be stored and managed along with the diagnosis target image of the subject according to the identification number of the subject. In this case, the diagnosis target image and the diagnosis assistance information of the subject may be managed in a time sequence. By managing the diagnosis assistant information and the diagnosis target image in time series, tracking and history management of individual diagnosis information may be facilitated.

Diagnostic assistance information may be provided to the user. The diagnostic assistance information may be provided to the user through the output means of the diagnostic apparatus or the client apparatus. The diagnostic assistance information may be output for the user to recognize through visual or audio output means provided in the diagnostic apparatus or the client apparatus.

According to an embodiment of the present invention, an interface for effectively providing diagnostic assistance information to a user may be provided. This user interface is described in more detail in section 5. User Interface which will be described later.

When the CAM is obtained by the neural network model, an image of the CAM may be provided together. In the case of a CAM image it may optionally be provided. For example, when the diagnostic information acquired through the diagnostic assisted neural network model is normal finding information or normal diagnostic information, the CAM image is not provided, and when the obtained diagnostic information is abnormal finding information or abnormal diagnostic information, for more accurate clinical diagnosis. CAM images may be provided together.

If the image is classified as inappropriate, the suitability information of the image may be provided together. For example, when the image is classified as inappropriate, the diagnostic assistance information and the non-conformance determination information acquired according to the image may be provided together.

The diagnosis subject image that is determined to be inappropriate may be classified as a retake subject image. In this case, a reshooting guide for the target object of the image classified as the reshooting target may be provided along with the suitability information.

Meanwhile, in response to providing the diagnostic assistance information obtained through the neural network model, feedback related to the training of the neural network model may be obtained. For example, feedback may be obtained for adjusting parameters or hyperparameters related to training of the neural network model. Feedback may be obtained through a user input provided in the diagnostic apparatus or the client apparatus.

According to an embodiment of the present invention, the diagnostic assistance information corresponding to the diagnosis target image may include grade information. Grade information may be selected from a plurality of grades. The grade information may be determined based on diagnosis information and / or finding information obtained through the neural network model. The grade information may be determined in consideration of suitability information or quality information of the diagnosis target image. If the neural network model is a classifier model that performs multiple classifications, the grade information may be determined in consideration of the class of the diagnosis target image classified by the neural network model. If the neural network model is a regression model that outputs a value associated with a particular disease, the grade information may be determined in consideration of the output value.

For example, the diagnostic assistance information acquired corresponding to the diagnosis target image may include any one grade information selected from the first grade information or the second grade information. The grade information may be selected as the first grade information when abnormal finding information or abnormal diagnosis information is obtained through the neural network model. The grade information may be selected as the second grade information when abnormal finding information or abnormal diagnosis information is not obtained through the neural network model. Alternatively, the grade information may be selected as the first grade information when the value obtained through the neural network model exceeds the reference value, and may be selected as the second grade information when the obtained value is less than the reference value. The first grade information may indicate that strong abnormality information exists in the diagnosis target image, compared to the second grade information.

Meanwhile, the grade information may be selected as the third grade information when it is determined that the quality of the diagnosis target image is lower than a reference by using an image analysis or a neural network model. Alternatively, the diagnostic assistance information may include third grade information together with the first or second grade information.

When the diagnostic assistance information includes the first grade information, the first user guide may be output through the output means. The first user guide may indicate that a more precise examination is required for the subject (ie, the patient) corresponding to the diagnosis assistance information. For example, the first user guidance may indicate that a secondary diagnosis (eg, a separate medical institution or a power supply procedure) for the subject is required. Alternatively, the first user guide may instruct treatment required for the subject. As a specific example, when abnormal information about macular degeneration of a subject is obtained by the diagnostic assistant information, the first user guide may provide instructions on the injection prescription and the power supply procedure for the subject (eg, List).

When the diagnostic assistance information includes the second grade information, the second user guide may be output through the output means. The second user guide may include a later management method for the subject corresponding to the diagnosis assistance information. For example, the second user guide may indicate a next medical treatment time, a next medical treatment subject, or the like.

When the diagnosis target information includes the third grade information, the third user guide may be output through the output means. The third user guide may indicate that re-photographing is required for the diagnosis target image. The third user guide may include information about the quality of the diagnosis target image. For example, the third user guide may include information of defects (eg, whether bright artifacts or dark artifacts or the like) present in the diagnosis target image.

4. Multiple Labels for Diagnostic Aid System

According to an embodiment of the present invention, a diagnostic assistance system for predicting a plurality of labels (eg, a plurality of diagnostic assistance information) may be provided. To this end, the diagnostic assistance neural network of the above-described diagnostic assistance system may be designed to predict a plurality of labels.

Alternatively, in the above-described diagnostic assistance system, a plurality of diagnostic assistance neural networks that predict different labels may be used in parallel. Hereinafter, such a parallel diagnostic assistance system will be described.

4.1 Parallel Diagnostic Assist System Configuration

According to an embodiment of the present invention, a parallel diagnostic assistance system for acquiring a plurality of diagnostic assistance information may be provided. The parallel diagnostic assistance system may learn a plurality of neural network models for acquiring a plurality of diagnostic assistance information, and obtain a plurality of diagnostic assistance information using the trained plurality of neural network models.

For example, the parallel diagnosis assistance system acquires a first neural network model obtaining first diagnosis assistance information related to the presence or absence of an eye disease in a subject and second diagnosis assistance information related to the presence or absence of a systemic disease in a subject based on the fundus image. The second neural network model may be trained, and diagnostic assistance information regarding the presence or absence of eye diseases and systemic diseases of the subject may be output using the learned first neural network model and the second neural network model.

21 and 22 are diagrams for describing a parallel diagnostic assistance system according to some embodiments of the present invention. 21 and 22, the parallel diagnosis assistance system may include a plurality of learning units.

Referring to FIG. 21, a parallel diagnosis assistance system 30 according to an embodiment of the present invention may include a learning apparatus 1000, a diagnosis apparatus 2000, and a client apparatus 3000. In this case, the learning apparatus 1000 may include a plurality of learning units. For example, the learning apparatus 1000 may include a first learner 100a and a second learner 100b.

Referring to FIG. 22, the parallel diagnosis assistance system 40 according to an embodiment of the present invention may include a first learning device 1000a, a second learning device 1000b, a diagnosis device 2000, and a client device 3000. It may include. The first learning apparatus 1000a may include a first learning unit 100a. The second learning apparatus 1000b may include a second learning unit 100b.

21 and 22, the first learner 100a may obtain a first data set and output a first parameter set of the first neural network model obtained as a result of learning with respect to the first neural network model. The second learner 100b may obtain a second data set and output a second parameter set of the second neural network model obtained as a result of learning with respect to the second neural network model.

The diagnosis apparatus 2000 may include a diagnosis unit 200. The description described with reference to FIG. 1 may be similarly applied to the diagnosis apparatus 2000 and the diagnosis unit 200. The diagnosis unit 200 obtains the first diagnosis assistance information and the second diagnosis assistance information by using the first neural network model and the second neural network model trained from the first learner 100a and the second learner 100b, respectively. can do. The diagnosis unit 2000 may store the parameters of the trained first neural network model and the parameters of the trained second neural network model acquired from the first learner 100a and the second learner 100b.

The client device 3000 may include a data acquirer, for example, an imaging unit 300. However, the imaging unit 300 may be replaced by an acquiring means for acquiring other diagnostic assistance information. The client device may transmit a diagnostic request and diagnosis target data (eg, an eye fundus image obtained from the image pickup unit) to the diagnostic device. In response to transmitting the diagnostic request, the client device 3000 may obtain a plurality of diagnostic assistance information according to the transmitted diagnosis target data from the diagnostic apparatus.

Meanwhile, in FIGS. 21 and 22, the diagnosis assistance system 40 has been described based on the case in which the diagnosis assistance system 40 includes the first learner 100a and the second learner 100b. However, the present disclosure is not limited thereto. According to another embodiment of the present invention, the learning apparatus may include a learning unit for acquiring three or more different diagnostic assistance information. Alternatively, the diagnostic assistance system may include a plurality of learning apparatuses for obtaining different diagnostic assistance information.

More specific operations of the learning apparatus, the diagnostic apparatus, and the client apparatus will be described in more detail below.

4.2 Parallel training process

According to an embodiment of the present invention, a plurality of neural network models may be learned. The training process for learning each neural network model can be performed in parallel.

4.2.1 Parallel learning

The training process may be performed by a plurality of learning units. Each training process may be performed independently of each other. The plurality of learning units may be provided in one learning device or in each of the plurality of learning devices.

FIG. 23 is a diagram illustrating a configuration of a learning apparatus including a plurality of learning units according to an exemplary embodiment. Configuration and operation of each of the first learner 100a and the second learner 100b may be implemented similarly to those described above with reference to FIG. 9.

Referring to FIG. 23, a process of a neural network model according to an embodiment of the present disclosure may include obtaining a first data processing module 110a, a first queue module 130a, a first learning module 150a, and a first learning result. The first learning unit 100a and the second data processing module 110b including the module 170a, the second cue module 130b, the second learning module 150b, and the second learning result obtaining module 170b may be used. It may be performed by the learning apparatus 1000 including the second learning unit 100b.

Referring to FIG. 23, a training process of a neural network model according to an embodiment of the present invention may be performed by the first learner 100a and the second learner 100b, respectively. The first learner 100a and the second learner 100b may independently learn the first neural network model and the second neural network model. Referring to FIG. 23, the first learner 100a and the second learner 100b may be provided in the above-described learning apparatus. Alternatively, the first learner and the second learner may be provided in different learning devices.

4.2.2 Parallel data acquisition

According to an embodiment of the present invention, the plurality of learners may acquire data. The plurality of learners may acquire different data sets. Alternatively, the plurality of learners may obtain the same data set. In some cases, the plurality of learners may acquire a data set, some of which are common. The data set may be a fundus image data set.

The first learner may obtain a first data set, and the second learner may obtain a second data set. The first data set and the second data set can be distinguished. The first data set and the second data set may be some common. The first data set and the second data set may be labeled fundus image data sets.

The first data set may include data labeled as normal for the first feature and data labeled as abnormal for the first feature. For example, the first data set may include a fundus image labeled as normal and a fundus image labeled as abnormal with respect to lens clouding.

The second data set may include data normally labeled for the second feature (distinguish from the first feature) and data labeled abnormally for the second feature. For example, the second data set may include a fundus image labeled as normal and a fundus image labeled as abnormal with respect to diabetic retinopathy.

The data set labeled normally for the first feature included in each of the first data set and the second data set and the data labeled normally for the second feature may be common. For example, the first data set includes an eye fundus image labeled as normal with respect to lens cloudiness and an ocular fundus image labeled as abnormal, and the second data set contains an eye fundus image and abnormal labeled as normal with diabetic retinopathy. A fundus image labeled as being included, wherein the fundus image labeled as normal with respect to the lens cloudiness included in the first data set and the fundus image labeled as normal with respect to diabetic retinopathy included in the second data set will be common. Can be.

Alternatively, data abnormally labeled with respect to the first feature included in each of the first data set and the second data set and data abnormally labeled with respect to the second feature may be common. That is, data labeled for multiple features may be used to train neural network models for multiple features.

Meanwhile, the first data set may be an eye fundus image data set photographed by the first method, and the second data set may be an eye fundus image data set photographed by the second method. The first method and the second method may be any one selected from red-free photographing, panorama photographing, autofluorescence photographing, infrared photographing, and the like.

The data set used in each learning unit may be determined in consideration of diagnostic assistance information obtained by the neural network to be learned. For example, when the first learner trains a first neural network model to acquire diagnostic assistance information related to retinal abnormalities (eg, microvascular perfusion, exudate, etc.), the first fundus image data that is red-free photographed. You can obtain a set. Alternatively, when the second learner learns a second neural network model for acquiring diagnosis assistance information related to macular degeneration, the second learner may acquire a second fundus image data set by autofluorescence.

4.2.3 Parallel data processing

The plurality of learners may process the obtained data, respectively. As described in 2.2 Data Processing Process described above, each learner may process the data by applying one or more of an image resizing, a preprocessing filter, an image enhancement, and an image serialization process. The first data processing module of the first learning unit may process the first data set, and the second data processing module of the second learning unit may process the second data set.

The first learner and the second learner included in the plurality of learners may differently process the obtained data set in consideration of diagnostic assistance information obtained from the neural network model learned by the learner. For example, the first learner may perform preprocessing so that the blood vessels are emphasized with respect to the fundus images included in the first fundus image data set in order to train the first neural network model for acquiring first diagnostic assistance information related to hypertension. Can be done. Alternatively, the second learner may learn about the fundus images included in the second fundus image data set to learn a second neural network model for acquiring second diagnostic assistance information related to abnormal findings such as exudates of the retina and microvascular vessels. You can also perform preprocessing to convert to a red-free image.

4.2.4 Parallel queue

The plurality of learners may store data in a queue. As described in 2.2.6 Queues above, each learner can store the processed data in a queue and deliver it to the learning module. For example, the first learner may store the first data set in the first cue module and provide the first learning module sequentially or randomly. The second learning module may store the second data set in the second cue module and provide the second learning module sequentially or randomly.

4.2.5 Parallel learning process

The plurality of learners may train the neural network model. Each learning module may independently train a diagnostic assisted neural network model that predicts for different labels using a training data set. The first learning module of the first learning unit trains the first neural network model, and the second learning module of the second learning unit trains the second neural network module.

The plurality of diagnostic assistance neural network models can be learned in parallel and / or independently. By learning the model to predict different labels through the plurality of neural network models, the prediction accuracy for each label can be improved and the efficiency of the prediction operation can be increased.

Each diagnostic assisted neural network model can be prepared similarly to that described in 2.3.2 Model Design. Each sub-learning process may be performed similar to that described above in 2.3.1 to 2.3.5.

The parallel learning process according to an embodiment of the present invention may include training a diagnostic assisted neural network model predicting different labels. The first learner may train the first diagnostic assistance neural network model predicting the first label. The second learner may train a second diagnostic assistance neural network model predicting the second label.

The first learner may train a first diagnostic assisted neural network model that obtains a first data set and predicts a first label. For example, the first learner may train a first diagnostic assisted neural network model predicting macular degeneration of a subject from an ocular fundus image by using a fundus image training data set labeled for macular degeneration.

The second learner can train the first diagnostic assistance neural network model to obtain a second data set and predict the second label. For example, the second learner may train the second diagnostic assisted neural network model predicting whether the subject is diabetic retinopathy from the fundus image, using the fundus image training data set that is labeled for diabetic retinopathy.

The learning process of the neural network model will be described in more detail with reference to FIGS. 24 and 25 below.

24 illustrates a parallel learning process according to an embodiment of the present invention. The parallel learning process may be applied to both the case where the parallel diagnosis assistance system is implemented as shown in FIG. 21, the case where it is implemented as shown in FIG. 22, and other forms. However, for convenience of description, the following description will be made based on the parallel diagnosis assistance system implemented as shown in FIG. 21.

Referring to FIG. 24, the parallel learning process may include a plurality of sub-learning processes each learning a plurality of diagnostic auxiliary neural network models predicting different labels. The parallel learning process may include a first sub learning process for learning the first neural network model and a second sub learning process for learning the second neural network model.

For example, the first sub-learning process obtains first data (S1010a), uses a first neural network model (S1030a), verifies a first model (ie, a first diagnostic assistance neural network model) (S1050a) This may be performed by obtaining parameters of the neural network model (S1070a). The second sub-learning process acquires second data (S1010b), uses the second neural network model (S1030b), verifies the second neural network model (ie, the second diagnostic auxiliary neural network model) (S1050b), and the second neural network. This may be performed by obtaining parameters of the model (S1070b).

The sub-learning process may include training the neural network model by inputting training data into the sub-neural network model, verifying the model by comparing the label value obtained as the output with the input training data, and reflecting the verification result back into the sub-neural network model. have.

Each sub-learning process obtains a result value using a neural network model to which arbitrary weight values are assigned, compares the obtained result value with a label value of training data, and performs backpropagation according to the error. May include optimizing them.

In each sub-learning process, the diagnostic assistance neural network model may be verified through a validation data set that is distinct from the training data set. Validation data sets for verifying the first neural network model and the second neural network model may be distinguished.

The plurality of learning units may acquire a learning result. Each learning result obtaining module may obtain information about the learned neural network module from the learning module. Each learning result obtaining module may obtain parameter values of the neural network module learned from the learning unit. The first learning result obtaining module of the first learning unit may obtain a first parameter set of the first neural network model learned from the first learning module. The second learning result obtaining module of the second learning unit may obtain a second parameter set of the second neural network model learned from the second learning module.

By each sub-learning process, optimized parameter values, i.e., a set of parameters, of the trained neural network model can be obtained. As learning proceeds with more training data sets, more appropriate parameter values can be obtained.

A first parameter set of the first diagnostic assistance neural network model learned by the first sub learning process may be obtained. A second parameter set of the second diagnostic assistance neural network model learned by the second sub learning process can be obtained. As the learning proceeds sufficiently, an optimized value of the weight and / or bias of the first diagnostic assistance neural network model and the second diagnostic assistance neural network model can be obtained.

The acquired parameter set of each neural network model may be stored in a learning device and / or a diagnostic device (or server). The first parameter set of the first diagnostic assistance neural network model and the second parameter set of the second diagnostic assistance neural network model may be stored together or separately. The parameter set of each neural network model learned may be updated by feedback obtained from a diagnostic device or a client device.

4.2.6 Parallel Ensemble Learning Process

Even when training a plurality of neural network models in parallel, the above-described ensemble type model learning may be used. Each sub-learning process may include training a plurality of sub-neural network models. The plurality of submodels may have different hierarchical structures. In the following, the content described in 2.3.7 may be similarly applied unless otherwise stated.

When a plurality of diagnostic assistive neural network models are trained in parallel, some of the sub-learning processes that train each diagnostic assistive neural network model train a single model, and some sub-learning processes share multiple submodels together. It may be implemented in the form of learning.

As the model is trained using the ensemble in each sub-learning process, the shape of a more optimized neural network model is obtained in each sub-process, and the error of prediction may be reduced.

25 illustrates a parallel learning process according to another embodiment of the present invention. Referring to FIG. 25, each learning process may include training a plurality of sub-neural network models.

Referring to FIG. 25, the first sub-learning process obtains first data (S1011a), uses a 1-1 neural network model and a 1-2 neural network model (S1031a, S1033a) and a 1-1 neural network model. The second neural network model may be verified (S1051a), and the final shape and its parameters of the first neural network model may be determined (S1071a). The second sub-learning process acquires second data (S1011b), uses the 2-1 neural network model and the 2-2 neural network model (S1031b, S1033b), the 2-1 neural network model and the 2-2 neural network model. In operation S1051b, the final shape of the first model (ie, the first diagnostic assistance neural network model) and its parameters may be determined (S1071b).

The first neural network learned in the first sub-learning process may include a 1-1 neural network model and a 1-2 neural network model. The 1-1 neural network model and the 1-2 neural network model may be provided in different hierarchical structures. The first-first neural network model and the first-two neural network model may obtain the first data set and output the predicted labels, respectively. Alternatively, the predicted label by the ensemble of the 1-1 neural network model and the 1-2 neural network model may be determined as the final predictive label.

In this case, the 1-1 neural network model and the 1-2 neural network model may be verified using the verification data set, and a highly accurate neural network model may be determined as the final neural network model. Alternatively, the ensemble of the 1-1 neural network model, the 1-2 neural network model, the 1-1 neural network model, and the 1-2 neural network model is verified, and the neural network model form of the case where the accuracy is high is the final first neural network. You can also decide by model.

Similarly with respect to the second sub-learning process, a highly accurate neural network of the ensemble of the 2-1 neural network model, the 2-2 neural network model, the 2-1 neural network model, and the 2-2 neural network model is final. Model (ie, a second diagnostic assistance neural network model).

Meanwhile, in FIG. 25, for convenience, each sub-learning process includes two sub-models. However, this is only an example and the present invention is not limited thereto. The neural network model trained in each sub-learning process may include only one neural network model or three or more submodels.

4.3 Parallel diagnostic process

According to an embodiment of the present invention, a diagnostic process for acquiring a plurality of diagnostic assistance information may be provided. The diagnostic process for acquiring the plurality of diagnostic assistance information may be implemented in the form of a parallel diagnostic assistance process including a plurality of diagnostic processes independent of each other.

4.3.1 Parallel diagnostics

According to an embodiment of the present invention, the diagnostic assistance process may be performed by a plurality of diagnostic modules. Each diagnostic assistance process can be performed independently.

26 is a block diagram illustrating a diagnosis unit 200 according to an embodiment of the present invention.

Referring to FIG. 26, the diagnosis unit 200 according to an embodiment of the present invention may include a diagnostic request obtaining module 211, a data processing module 231, a first diagnostic module 251, and a second diagnostic module 253. And an output module 271. Each module of the diagnostic unit 200 may operate similarly to the diagnostic module of the diagnostic unit illustrated in FIG. 18 unless otherwise noted.

In FIG. 26, even when the diagnostic unit 200 includes a plurality of diagnostic modules, the diagnostic request obtaining module 211, the data processing module 231, and the output module 271 are illustrated as being common. It is not limited to this configuration, a plurality of diagnostic request acquisition module, data processing module and / or output module may also be provided. A plurality of diagnostic request acquisition modules, data processing modules and / or output modules may also operate in parallel.

For example, the diagnosis unit 200 may include a first data processing module for performing a first processing on an input diagnosis target image and a second processing module performing second data processing on a diagnosis target image. The first diagnostic module may acquire first diagnostic assistance information based on the first processed diagnosis target image, and the second diagnostic module may acquire second diagnostic assistance information based on the second processed diagnosis target image. The first and / or second processing may be any one selected from image resizing, color modulation of the image, blur filter application, blood vessel highlighting, red-free transformation, partial region cropping, and some element extraction.

The plurality of diagnostic modules may acquire different diagnostic assistance information. The plurality of diagnostic modules may obtain diagnostic assistance information using different diagnostic assistance neural network models. For example, the first diagnostic module obtains first diagnostic assistance information related to the subject's eye disease using a first neural network model that predicts the subject's eye disease, and the second diagnostic module detects the eye disease. By using a second neural network model for predicting whether the subject is a systemic disease, second diagnostic assistance information related to the subject's systemic disease may be obtained.

More specifically, the first diagnostic module acquires first diagnostic assistance information on whether the subject is diabetic retinopathy by using a first diagnostic assistive neural network model that predicts whether the subject is diabetic retinopathy based on the fundus image. The second diagnostic module may obtain second diagnostic assistance information related to hypertension of the subject using a second diagnostic assistive neural network model that predicts whether the subject is hypertension based on the fundus image.

4.3.2 Parallel diagnostic process

The diagnostic assistance process according to an embodiment of the present invention may include a plurality of subdiagnostic processes. Each sub-diagnostic process may be performed using a different diagnostic assistance neural network model. Each sub-diagnostic process may be performed in a different diagnostic module. For example, the first diagnostic module may perform a first sub-diagnosis process of acquiring first diagnostic assistance information through the first diagnostic assistance neural network model. Alternatively, the second diagnostic module may perform a second sub-diagnostic process of obtaining second diagnostic assistance information through the second diagnostic assistance neural network model.

The trained plurality of neural network models may input diagnostic target data and output predicted labels or probabilities. Each neural network model is provided in the form of a classifier, and the input diagnosis target data may be classified with respect to a predetermined label. In this case, the plurality of neural network models may be provided in the form of a classifier learned about different characteristics. Each neural network model can classify the data to be diagnosed as described in 3.4.2.

On the other hand, from each diagnostic assistance neural network model, a CAM can be obtained, and the CAM can be obtained selectively. The CAM can be extracted if a predetermined condition is satisfied. For example, when the first diagnostic assistance information indicates that the subject is abnormal with respect to the first characteristic, the first CAM may be obtained from the first diagnostic assistance neural network model.

27 is a diagram for describing a diagnostic assistance process according to an embodiment of the present invention.

Referring to FIG. 27, in the diagnosis assisting process according to an embodiment of the present invention, the diagnosis target data is acquired (S2011), and the diagnosis is performed using the first diagnosis assistance neural network model and the second diagnosis assistance neural network model (S2031a, S2031b). It may include acquiring (S2051) diagnostic assistance information according to the target data. The data to be diagnosed may be processed data.

According to an embodiment of the present invention, the diagnostic assistance process may include obtaining first diagnostic assistance information through a trained first diagnostic assistance neural network model, and obtaining second diagnostic assistance information through a learned second diagnostic assistance neural network model. It may include. The first diagnostic assistance neural network model and the second diagnostic assistance neural network model may obtain first diagnostic assistance information and second diagnostic assistance information based on the same diagnosis target data.

For example, the first diagnosis aid neural network model and the second diagnosis aid neural network model are based on the diagnosis target fundus image, the first diagnosis aid information on whether the subject is macular degeneration and the subject on whether diabetic retinopathy of the subject 2 diagnostic assistance information can be obtained respectively.

In addition, unless otherwise indicated, the diagnostic assistance process described with reference to FIG. 27 may be implemented similarly to the diagnostic assistance process described above with reference to FIG. 20.

4.3.3 Output of diagnostic assistant information

According to an embodiment of the present invention, diagnostic assistance information obtained by the parallel diagnostic assistance process may be obtained. The obtained diagnostic assistance information may be stored in the diagnostic apparatus, the server apparatus, and / or the client apparatus. The obtained diagnostic assistance information may be delivered to an external device.

The plurality of diagnostic assistance information may indicate a plurality of labels predicted by the plurality of diagnostic assistance neural network models, respectively. The plurality of diagnostic assistance information may correspond to a plurality of labels predicted by the plurality of diagnostic assistance neural network models, respectively. Alternatively, the diagnostic assistance information may be information determined based on a plurality of labels predicted by the plurality of diagnostic assistance neural network models. The diagnostic assistance information may correspond to a plurality of labels predicted by the plurality of diagnostic assistance neural network models.

In other words, the first diagnostic assistance information may be diagnostic assistance information corresponding to the first label predicted through the first diagnostic assistance neural network model. Alternatively, the first diagnostic assistance information may be diagnostic assistance information determined by considering the first label predicted through the first diagnostic assistance neural network model and the second label predicted through the second diagnostic assistance neural network model.

Meanwhile, an image of the CAM obtained from the plurality of diagnostic assistance neural network models may be output. The CAM image may be output when a predetermined condition is satisfied. For example, when the first diagnostic assistance information indicates that the subject is abnormal with respect to the first characteristic or when the second diagnostic assistance information indicates that the subject is abnormal with respect to the second characteristic, The CAM image obtained from the diagnostic assistance neural network model to which the diagnostic assistance information indicated as abnormal may be output.

The plurality of diagnostic assistance information and / or CAM images may be provided to the user. The plurality of diagnostic assistance information and the like may be provided to the user through output means of the diagnostic apparatus or the client apparatus. Diagnostic assistance information may be visually output. In this regard, it is described in more detail in 5. User interface.

According to an embodiment of the present invention, the diagnostic assistance information corresponding to the diagnosis target image may include grade information. Grade information may be selected from a plurality of grades. The grade information may be determined based on the plurality of diagnostic information and / or finding information obtained through the neural network model. The grade information may be determined in consideration of suitability information or quality information of the diagnosis target image. The grade information may be determined in consideration of a class in which a diagnosis target image is classified by a plurality of neural network models. The grade information may be determined in consideration of numerical values output from the plurality of neural network models.

For example, the diagnostic assistance information acquired corresponding to the diagnosis target image may include any one grade information selected from the first grade information or the second grade information. The grade information may be selected as the first grade information when at least one abnormality finding information or abnormal diagnosis information is obtained from the diagnosis information acquired through the plurality of neural network models. The grade information may be selected as the second grade information when abnormal finding information or abnormal diagnosis information is not obtained from the diagnosis information acquired through the neural network model.

The grade information is selected as the first grade information when at least one of the values obtained through the neural network model exceeds the reference value, and the grade information is selected as the second grade information when all of the acquired values fall below the reference value. May be The first grade information may indicate that strong abnormality information exists in the diagnosis target image, compared to the second grade information.

The grade information may be selected as the third grade information when it is determined that the quality of the diagnosis target image is equal to or lower than a reference value by using an image analysis or a neural network model. Alternatively, the diagnostic assistance information may include third grade information together with the first or second grade information.

When the diagnostic assistance information includes the first grade information, the first user guide may be output through the output means. The first user guide may include at least one abnormal finding information included in the diagnosis assistance information or a matter corresponding to the abnormal diagnostic information. For example, the first user guide may indicate that a more precise examination is required for a subject (ie, a patient) corresponding to abnormal information included in the diagnostic assistance information. For example, the first user guidance may indicate that a secondary diagnosis (eg, a separate medical institution or a power supply procedure) for the subject is required. Alternatively, the first user guide may instruct treatment required for the subject. As a specific example, when abnormal information about macular degeneration of a subject is obtained by the diagnostic assistant information, the first user guide may provide instructions on injection prescription and power supply procedure for the subject (eg, List).

When the diagnostic assistance information includes the second grade information, the second user guide may be output through the output means. The second user guide may include a later management method for the subject corresponding to the diagnosis assistance information. For example, the second user guide may indicate a next medical treatment time, a next medical treatment subject, or the like.

When the diagnosis target information includes the third grade information, the third user guide may be output through the output means. The third user guide may indicate that re-photographing is required for the diagnosis target image. The third user guide may include information about the quality of the diagnosis target image. For example, the third user guide may include information of defects (eg, whether bright artifacts or dark artifacts or the like) present in the diagnosis target image.

The first to third grade information may be output by an output unit of the client device or the diagnostic device. Specifically, it may be output through a user interface described later.

4.4 Example 2-Diagnostic Assistance Systems

The diagnostic assistance system according to an embodiment of the present invention may include a fundus image obtaining unit, a first processing unit, a second processing unit, a third processing unit, and a diagnostic information output unit.

According to an embodiment of the present invention, the diagnostic assistance system may include a diagnostic device. The diagnostic apparatus may include a fundus image acquirer, a first processor, a second processor, a third processor, and / or a diagnostic information output unit. However, the present invention is not limited thereto, and each unit included in the diagnosis assistance system may be located at appropriate locations on the learning device, the diagnosis device, the learning diagnosis server, and / or the client device. Hereinafter, for convenience, a diagnosis apparatus of the diagnosis assistance system will be described based on a case in which the fundus image obtaining unit, the first processing unit, the second processing unit, the third processing unit, and the diagnostic information output unit are included.

28 is a diagram for describing a diagnosis assistance system according to an embodiment of the present invention. Referring to FIG. 28, the diagnostic assistance system may include a diagnostic apparatus, and the diagnostic apparatus may include a fundus image obtaining unit, a first processing unit, a second processing unit, a third processing unit, and a diagnostic information output unit.

According to an embodiment of the present invention, a diagnosis aiding system for assisting diagnosis of a plurality of diseases based on the fundus image may include a fundus image obtaining unit configured to obtain a target fundus image as a basis for obtaining diagnosis aid information for a subject. And obtaining a first result associated with a first finding on the subject using a first neural network model with respect to the subject fundus image, wherein the first neural network model is machine learned based on the first set of fundus images. The subject using a first processing unit, a second neural network model with respect to the target fundus image, wherein the second neural network model is machine learned based on a second set of fundus images different from at least a portion of the first fundus image set; A second processing unit for obtaining a second result related to a second finding about the, based on the first result and the second result, for the subject The processing unit may comprise 3, and the user parts of the diagnostic information output for providing the determined diagnostic information for determining the diagnostic information. In this case, the first finding and the second finding may be used to diagnose different diseases.

The first neural network model is trained to classify the input fundus image into any one of a normal label and an abnormal label in relation to the first finding, and the first processing unit uses the first neural network model to classify the target fundus image into the normal label. Alternatively, the first result may be obtained by classifying into one of the abnormal labels.

The third processing unit may determine whether the diagnostic information according to the target fundus image is normal information or abnormal information by considering the first result and the second result together.

The third processing unit may determine the diagnostic information on the subject by giving priority to the abnormal label so that diagnostic accuracy is improved.

 The third processing unit determines that the diagnostic information is normal when the first label is a normal label for the first finding, and the second label is a normal label for the second finding, and the first label is the first label. If the label is not a normal label for the finding or the second label is not a label for the second finding, the diagnostic information may be abnormally determined.

The first finding is associated with an ocular disease, and the first result may indicate whether the subject is normal for the ocular disease. The second finding is associated with systemic disease, and the second result may indicate whether the subject is normal for the systemic disease.

A first eye is associated with a first eye disease, the first result indicates whether the subject is normal for the first eye disease, and the second eye is a second eye that is distinct from the first eye disease Associated with a disease, the second result may indicate whether the subject is normal for the second eye disease.

 The first finding is a finding for diagnosing a first eye disease, the first result indicates whether the subject is normal to the first ocular disease, and the second finding is for diagnosing the first eye disease. The findings are distinguished from the first findings, and the second result may indicate whether the subject is normal to the second eye disease.

The first neural network model includes a first sub neural network model and a second sub neural network model, and the first result is a first prediction value predicted by the first sub neural network model and a second predicted by the second sub neural network model. It may be determined by considering the prediction value together.

The first processor may acquire a Class Activation Map (CAM) associated with the first label through the first neural network model, and the diagnostic information output unit may output an image of the CAM.

The diagnostic information output unit may output an image of the CAM when the diagnostic information acquired by the third processing unit is abnormal diagnostic information.

The diagnosis assistance system may further include a fourth processor configured to acquire quality information of the target fundus image, and the diagnostic information output unit may output quality information of the target fundus image obtained by the fourth processor.

 If it is determined by the fourth processor that the quality information of the target fundus image is equal to or less than a predetermined quality level, the diagnostic information output unit is configured such that the quality information of the target fundus image is determined by the user together with the determined diagnostic information. Information indicating that the following can be provided together.

5. User interface

According to one embodiment of the invention, the above-described client device or diagnostic device may have a display unit for providing the diagnostic assistance information to the user. In this case, the display unit may be provided to clearly transmit the diagnostic assistance information to the user, and to easily obtain feedback from the user.

As an example of the display unit, a display for providing visual information to a user may be provided. In this case, a graphic user interface for visually delivering the diagnostic assistance information to the user may be used. For example, in the fundus diagnosis assistance system for acquiring the diagnosis assistance information based on the fundus image, a graphic user interface may be provided to effectively display the acquired diagnosis assistance information and help the user understand.

29 and 30 are diagrams for describing a graphical user interface for providing diagnostic information to a user, according to some embodiments of the present disclosure. In the following, with reference to Figures 29 and 30, some embodiments of the user interface that can be used in the fundus diagnosis assistance system will be described.

Referring to FIG. 29, a user interface according to an embodiment of the present invention may display identification information of a subject corresponding to a diagnosis fundus image. The user interface may include a target image identification information display unit 401 that displays identification information of a subject (ie, a patient) and / or photographing information (eg, a photographing date) of a diagnosis fundus image.

The user interface according to an embodiment of the present invention may include a fundus image display unit 405 for displaying a fundus image of a left eye and a right eye of a same subject, respectively. The fundus image display unit 405 may display a CAM image.

The user interface according to an embodiment of the present invention indicates that the image of the left eye or the right eye for each of the left eye fundus image and the right eye fundus image, and a diagnostic information indicator indicating whether each image is diagnosed and whether the user is confirmed The diagnostic information indicating unit 403 may be included.

The color of the diagnostic information indicator may be determined in consideration of diagnostic assistance information obtained based on the target fundus image. The diagnostic information indicator may be displayed in a first color or a second color according to the diagnostic assistance information. For example, when the first to third diagnostic assistance information is acquired in one target fundus image, when even one diagnostic assistance information includes abnormal (that is, abnormal) information, the diagnostic information indicator is displayed in red. In the case where all the diagnostic assistant information includes normal (ie, no abnormality) information, the diagnostic information indicator may be displayed in green.

The type of diagnostic information indicator may be determined according to whether the user confirms it. The diagnostic information indicator may be displayed in a first form or a second form depending on whether the user confirms the check. For example, referring to FIG. 25, the diagnostic information indicator corresponding to the target fundus image reviewed by the user is indicated by a filled circle, and the diagnostic information indicator corresponding to the target fundus image unreviewed by the user is filled with a semicircle. Can be displayed.

The user interface according to an embodiment of the present invention may include a diagnostic information indicating unit 407 for indicating diagnostic assistance information. The diagnostic assistance information indicator may be located in the left eye and right eye images, respectively. The diagnostic assistance information indicating unit may indicate a plurality of findings or diagnostic information.

The diagnostic assistance information indicator may include at least one diagnostic assistance information indicator. The diagnostic assistant information indicator may indicate corresponding diagnostic assistant information through color change.

For example, the first diagnosis aid information indicating the presence or absence of lens turbidity through the first diagnosis aid neural network model, and the second diagnosis aid information indicating the presence or absence of glaucoma findings through the second diagnosis aid neural network model. When the third diagnostic assistance information indicating whether there is a retinal abnormality is obtained through the third diagnostic assistance neural network model, the diagnostic information indicating unit indicates the first diagnostic assistance information, the second diagnostic assistance, and the third diagnostic assistance information, respectively. It may include the first to third diagnostic assistant information indicator.

As a specific example, referring to FIG. 29, in relation to the left eye fundus image, the diagnostic information indicating unit 407 may include first diagnostic assistance information indicating that the diagnosis aid information acquired based on the left eye fundus image of the subject is abnormal lens turbidity. The first diagnostic assistant information indicator having the first color when the second diagnostic assistant information indicating normal glaucoma (no glaucoma) and the third diagnostic assistant information indicating retinal abnormality (with abnormal findings) are obtained; The first diagnostic assistance information indicator having the second color and the third diagnostic assistance information indicator having the first color may be displayed.

The user interface according to an embodiment of the present invention may obtain a user comment on the diagnosis target fundus image from the user. The user interface may include a user comment object 409 and display a user input window in response to a user selection for the user comment object. The comments obtained from the user may be used for updating the diagnostic assistance neural network model. For example, the user input window displayed in response to the selection of the user comment object may obtain a user evaluation of the diagnostic assistance information through the neural network, and the obtained user evaluation may be used to update the neural network model.

The user interface according to an embodiment of the present invention may include a review instruction object 411 indicating whether the user reviews the respective fundus image. The review instruction object may receive a user input indicating that the user review of each diagnosis target image is completed, and the display may be changed from the first state to the second state. For example, referring to FIGS. 29 and 30, the review instruction object may be changed to a second state indicating that confirmation is obtained when a user input is obtained in a first state displaying a confirmation request text.

A list 413 of the diagnosis fundus images may be displayed. In the list, the identification information of the subject, the image photographing date, and the indicator for reviewing the binocular image may be displayed together.

In the list 413 of the diagnosis target fundus images, a review completion indicator 415 indicating the reviewed diagnosis fundus image may be displayed. The review completion indicator 415 may be displayed when user selection has occurred for all of the review indicating object 411 for both eyes of the corresponding image.

Referring to FIG. 30, when it is determined that there is a quality abnormality in the diagnosis fundus image, the user graphic interface may include a low quality warning object 417 which indicates to the user that there is an abnormality in the quality of the target fundus image. have. The low quality warning object 417 is displayed when the diagnosis fundus image is judged by the diagnosis unit to be less than the quality (that is, the reference quality level) of the level at which the appropriate diagnosis aid information can be predicted from the diagnosis aid neural network model. Can be.

Also, referring to FIG. 28, the low quality warning object 419 may be displayed in the list 413 of the diagnosis fundus images.

Although the embodiments have been described by the limited embodiments and the drawings as described above, various modifications and variations are possible to those skilled in the art from the above description. For example, the described techniques may be performed in a different order than the described method, and / or components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different form than the described method, or other components. Or, even if replaced or substituted by equivalents, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are within the scope of the claims that follow.

Claims (28)

In the diagnostic assistance system for assisting in the diagnosis of a plurality of diseases based on the fundus image,
A fundus image acquisition unit acquiring a target fundus image as a basis for acquiring diagnosis assistance information about a subject;
Obtaining a first result associated with a first finding for the subject using a first neural network model for the subject fundus image, wherein the first neural network model is machine learned based on the first set of fundus images 1 processing unit;
A second neural network model for the subject fundus image, wherein the second neural network model is machine trained based on a second set of fundus images that are at least partially different from the first fundus image set; A second processing unit for obtaining a second result related to the findings;
A third processor configured to determine diagnostic information on the subject based on the first result and the second result; And
A diagnostic information output unit providing the determined diagnostic information to a user; Containing
The first and second findings are used to diagnose different diseases,
Diagnostic Assistant System.
The method of claim 1,
The first neural network model is trained to classify the input fundus image into any one of a normal label and an abnormal label in relation to the first finding,
The first processor may classify the target fundus image into either the normal label or the abnormal label by using the first neural network model to obtain the first result.
Diagnostic Assistant System.
The method of claim 1,
The third processor is configured to determine whether the diagnostic information according to the target fundus image is normal information or abnormal information in consideration of the first result and the second result.
Diagnostic Assistant System.
The method of claim 3,
The third processing unit determines the diagnostic information on the subject by giving priority to the abnormal label so that diagnostic accuracy is improved.
Diagnostic Assistant System.
The method of claim 3,
The third processor determines the diagnostic information to be normal when the first label is a normal label for the first finding and the second label is a normal label for the second finding,
Determining that the diagnostic information is abnormal when the first label is not a normal label for the first finding or the second label is not a normal label for the second finding.
Diagnostic Assistant System.
The method of claim 1,
The first finding is associated with an ocular disease, the first result indicates whether the subject is normal for the ocular disease,
Said second finding relates to systemic disease, and said second result indicates whether said subject is normal for said systemic disease,
Diagnostic Assistant System.
The method of claim 1,
The first finding is associated with a first eye disease, and the first result indicates whether the subject is normal to the first eye disease,
The second finding relates to a second eye disease that is distinct from the first eye disease, and wherein the second result indicates whether the subject is normal to the second eye disease,
Diagnostic Assistant System.
The method of claim 1,
The first neural network model includes a first sub neural network model and a second sub neural network model,
The first result is determined in consideration of the first prediction value predicted by the first sub-neural network model and the second prediction value predicted by the second sub-neural network model.
Diagnostic Assistant System.
The method of claim 1,
The first processor acquires a class activation map (CAM) associated with the first label through the first neural network model,
The diagnostic information output unit outputs an image of the CAM,
Diagnostic Assistant System.
The method of claim 1,
The diagnostic information output unit outputs an image of the CAM when the diagnostic information acquired by the third processing unit is abnormal diagnostic information.
Diagnostic Assistant System.
The method of claim 1,
The diagnostic assistance system further includes a fourth processing unit for obtaining quality information of the target fundus image,
The diagnostic information output unit outputs quality information of the target fundus image obtained by the fourth processing unit.
Diagnostic Assistant System.
The method of claim 1,
If it is determined by the fourth processor that the quality information of the target fundus image is less than or equal to a predetermined quality level,
The diagnostic information output unit,
Providing the user with the determined diagnostic information together with information indicating that the quality information of the target fundus image is below the predetermined quality level,
Diagnostic Assistant System.
Learning to obtain a first set of training data comprising a plurality of fundus images, to process the fundus image included in the first training data set, and to train a first neural network model using the first training data set; Device;
A diagnostic apparatus for acquiring an object fundus image for acquiring diagnosis assistance information, and acquiring the diagnosis aid information based on the object fundus image using the learned first neural network model; In the control method of the learning apparatus included in the system comprising a,
Performing preprocessing of the first fundus image such that the first fundus image included in the first training data set is transformed into a form suitable for learning the first neural network model;
Serializing the preprocessed first fundus image; And
Training the first neural network model to classify the target fundus image into a first label or a second label using the serialized first fundus image; Containing
How to control the learning device.
The method of claim 13,
The learning device obtains a second training data set comprising a plurality of fundus images and at least partially different from the first training data set, using the second training data set, training a second neural network model,
The control method of the learning device is
Performing preprocessing of the second fundus image such that the second fundus image included in the second training set is suitable for learning of the second neural network model;
Serializing the preprocessed second fundus image;
Training the second neural network model to classify the target fundus image into a third label or a fourth label using the serialized second fundus image; Containing more
How to control the learning device.
The method of claim 14,
The first preprocessing performed on the fundus image included in the first training data set is distinct from the second preprocessing performed on the fundus image included in the second training data set,
How to control the learning device.
The method of claim 14,
Validating the first neural network model by evaluating the accuracy of the learned first neural network model using a first verification data set that is at least partially distinct from the first training data set; And
Validating the second neural network model by evaluating the accuracy of the learned first neural network model using a second verification data set that is at least partially distinct from the second training data set; Include more,
Verification of the first neural network model and the second neural network model is performed independently,
How to control the learning device.
The method of claim 14,
The serialized first fundus image is sequentially stored in a first cue, the serialized fundus image stored in the first cue is used for learning the first neural network model in a predetermined capacity unit from the first cue,
The serialized second fundus image is sequentially stored in a second queue that is distinct from the first queue, and the serialized fundus image stored in the second queue is the first neural network model in a predetermined capacity unit from the second queue. Used for learning
How to control the learning device.
The method of claim 14,
The first neural network model includes a first sub neural network model and a second sub neural network model,
Classifying the target fundus image into the first label or the second label is performed by considering a first prediction value predicted by the first sub-neural network model and a second prediction value predicted by the second sub-neural network model. Become,
The second neural network model includes a third sub neural network model and a fourth sub neural network model,
The object fundus image is classified into the third label or the fourth label by considering a third prediction value predicted by the third sub-neural network model and a fourth prediction value predicted by the fourth sub-neural network model. felled,
How to control the learning device.
The method of claim 14,
The first training data set includes at least a portion of the fundus image labeled with a first label, and the second training data set includes at least a portion of the fundus image labeled with a third label,
The fundus image labeled with the first label and the at least some fundus image labeled with the third label are partially common,
How to control the learning device.
The method of claim 13,
The first label is a normal label indicating that the subject corresponding to the subject fundus image is normal with respect to the first finding, and the second label is an abnormal label indicating that the subject is abnormal with respect to the second finding. sign,
How to control the learning device.
The method of claim 13,
The pre-processing of the first fundus image may include cropping the first fundus image to satisfy a reference aspect ratio and changing the size of the first fundus image.
How to control the learning device.
The method of claim 13,
The performing of the pretreatment of the first fundus image may further include applying a blood vessel emphasis filter to the fundus image such that the processing unit highlights blood vessels included in the first fundus image.
How to control the learning device.
The method of claim 13,
The serialized first fundus image is sequentially stored in a queue, and the serialized first fundus image stored in the queue is used for learning the first neural network model in a predetermined capacity unit from the queue,
The queue requests replenishment of the serialized first fundus image when the serialized first fundus image that is not used in the first training is reduced below a reference capacity;
How to control the learning device.
The method of claim 13,
The first finding is any one of retinal bleeding findings, retinal exudation findings, lens turbidity findings, diabetic retinopathy findings,
How to control the learning device.
The method of claim 13,
Validating the first neural network model by evaluating the accuracy of the learned first neural network model using a first verification data set that is at least partially distinct from the first training data set; Containing more
How to control the learning device.
The method of claim 25,
Updating the first neural network model based on the verification result;
How to control the learning device.
The method of claim 13,
The first neural network model includes a first sub neural network model and a second sub neural network model,
Training the first neural network model
Verifying the first sub-neural network model using a first verification data set to obtain an accuracy of the first sub-neural network model,
Verifying the second sub-neural network model using the first verification data set to obtain an accuracy of the second sub-neural network model,
Comparing the accuracy of the first sub-neural network model with the accuracy of the second sub-neural network model to determine a more accurate sub-neural network model as a final neural network model.
How to control the learning device.
A computer-readable recording medium having recorded thereon a program for performing any one of the control method of the learning apparatus of claim 13 to 27.

Images