KR102648322B1 - System for aiding keratoconus diagnosis and method using the system - Google Patents

Abstract

A keratoconus diagnostic assistance system and a method of using the same are disclosed. According to one embodiment, an assistive device for diagnosing keratoconus using artificial intelligence performed by a computing device includes: a memory that stores cornea-related image data based on digitized test data of a subject; and a processor, wherein the processor acquires the stored cornea-related image data of the subject from the memory, inputs the obtained cornea-related image data into a diagnostic model, and determines whether keratoconus occurs in the subject's eye. Detects the occurrence of keratoconus and can be classified as keratoconus (KC), suspected keratoconus (KCS), or normal cornea (NM).

Description

Keratoconus diagnostic assistance system and method of using the same {SYSTEM FOR AIDING KERATOCONUS DIAGNOSIS AND METHOD USING THE SYSTEM}

The following embodiments relate to an assistive system for diagnosing keratoconus and a method of using the same. More specifically, it relates to a system that assists in diagnosing keratoconus using an artificial neural network and a method of using the same.

Keratoconus is a corneal disease in which the corneal stroma gradually becomes thinner and cone-shaped corneal protrusions occur. In the early stages of keratoconus, no symptoms are felt, but as it progresses, the myopia and astigmatism that started earlier continue to increase.

In the prior art, a formula for diagnosing keratoconus is installed in the test equipment, and the presence or absence of keratoconus can be determined from the test data results. However, this has a problem in that the false positive rate in determining whether or not the patient has keratoconus is high, resulting in misdiagnosis where the patient is judged to have keratoconus when it is not.

Recently, various research and equipment are being developed to make a more accurate diagnosis related to keratoconus.

One task relates to providing information on whether keratoconus occurs in a subject's cornea using an artificial neural network.

One task relates to providing information about whether keratoconus occurs in a subject's cornea from cornea-related image data.

The problem to be solved by the present invention is not limited to the above-mentioned problems, and problems not mentioned can be clearly understood by those skilled in the art from this specification and the attached drawings. .

According to one embodiment, an assistive device for diagnosing keratoconus using artificial intelligence performed by a computing device includes a memory and a processor that stores cornea-related image data based on digitized test data of a subject, wherein the processor is configured to: Obtain image data related to the subject's cornea stored from memory, input the acquired image data related to the cornea into a diagnostic model to detect whether keratoconus occurs in the subject's eye, and determine whether keratoconus occurs in the subject's eye (KC) ), suspected keratoconus (KCS), or normal cornea (NM). It can provide an assistance system for diagnosing keratoconus.

The solution to the problem of the present invention is not limited to the above-mentioned solution, and the solution not mentioned can be clearly understood by those skilled in the art from this specification and the attached drawings. You will be able to.

According to one embodiment, the accurate diagnosis rate of keratoconus can be increased through a keratoconus diagnosis assistance system and process.

According to one embodiment, by accurately diagnosing keratoconus through a keratoconus diagnosis assistance system and process, it is possible to prevent future corneal ectasia that may occur by performing vision correction surgery.

The effects of the present invention are not limited to the effects described above, and effects not mentioned can be clearly understood by those skilled in the art from this specification and the attached drawings.

Figure 1 is a diagram for explaining keratoconus.
Figure 2 is a diagram illustrating a keratoconus diagnosis assistance system according to an embodiment.
Figure 3 is a diagram illustrating a keratoconus diagnosis assistance system according to another embodiment.
Figure 4 is a diagram illustrating a keratoconus diagnosis assistance system according to another embodiment.
Figure 5 is a diagram illustrating a keratoconus diagnosis assistance system according to another embodiment.
Figure 6 is a diagram of a keratoconus diagnosis assistance process according to an embodiment.
Figure 7 is a diagram illustrating a diagnostic assistance model according to an embodiment.
Figure 8 is a diagram illustrating a diagnostic assistance model according to an embodiment.
Figure 9 is a diagram illustrating a diagnostic assistance device according to an embodiment.
FIG. 10 is a diagram illustrating cornea-related image data according to an embodiment.
FIG. 11 is a diagram illustrating cornea-related image data according to an embodiment.
FIG. 12 is a diagram illustrating cornea-related image data according to another embodiment.
FIG. 13 is a diagram illustrating cornea-related image data according to another embodiment.
Figure 14 is a diagram related to standardization of input data according to an embodiment.
Figure 15 is a diagram related to standardization of input data according to an embodiment.
Figure 16 is a flowchart of steps for assisting in diagnosing keratoconus according to an embodiment.
Figure 17 is a flowchart of steps for assisting in diagnosing keratoconus according to an embodiment.
Figure 18 is a diagram related to keratoconus diagnosis assistance according to an embodiment.
Figure 19 is a diagram related to keratoconus diagnosis assistance according to another embodiment.
Figure 20 is a diagram related to keratoconus diagnosis assistance according to another embodiment.

According to one embodiment, an assistive device for diagnosing keratoconus using artificial intelligence performed by a computing device includes: a memory that stores cornea-related image data based on digitized test data of a subject; and a processor; wherein the processor acquires the stored image data related to the cornea of the subject from the memory, and inputs the acquired image data related to the cornea into a diagnostic model to determine whether keratoconus occurs in the eye of the subject. Detects the occurrence of keratoconus and can be classified as keratoconus (KC), suspected keratoconus (KCS), or normal cornea (NM).

The diagnostic model may be learned based on image data including any one or more of test data of a person who was previously diagnosed with keratoconus, test data of a person who was previously diagnosed with suspected keratoconus, and test data of a person who was previously diagnosed with normal cornea. .

The cornea-related image data includes a first group parameter, the digitized inspection data includes a second group parameter, the first group parameter is generated based on the second group parameter, and the first group parameter represents a parameter that is not defined in the test equipment that measures the numerical test data, and the accuracy of the result of whether keratoconus occurs in the eye of the subject is determined by data including the first group parameter being input to the diagnostic model. The case may be higher than the case where data including the second group parameter is input to the diagnostic model.

The processor may detect whether keratoconus occurs in the subject's eye by inputting the quantified test data together with the acquired cornea-related image data into the diagnostic model.

According to one embodiment, an assistive device for diagnosing keratoconus using artificial intelligence performed by a computing device includes: a memory that stores cornea-related image data based on digitized test data of a subject; and a processor, wherein the processor acquires the stored cornea-related image data of the subject from the memory, and inputs first input data obtained from the acquired cornea-related image data into a first diagnostic model, A first output value - the first output value outputs that the subject's cornea is either a normal cornea (NM) or keratoconus (KC) - is derived, and when the first output value is a normal cornea (NM), the Inputting cornea-related image data into a second diagnostic model to derive a second output value - the second output value indicating that the subject's cornea is either normal cornea (NM) or suspected keratoconus (KCS), When the first output value is keratoconus (KC), the cornea-related image data is input into a third diagnostic model, and the third output value - the third output value is determined to indicate that the subject's cornea is keratoconus (KC) and corneal ectasia ( CE), which outputs one of - can be derived.

The first diagnostic model is learned based on test data of a person who was previously diagnosed with a normal cornea and test data of a person who was previously diagnosed with keratoconus, and the second diagnostic model is learned based on test data of a person who was previously diagnosed with a normal cornea and past suspicious cones. It is learned based on the examination data of a person who was previously diagnosed with corneal ectasia, and the third diagnostic model may be learned based on the examination data of a person who was previously diagnosed with suspected keratoconus and the examination data of a person who was previously diagnosed with corneal ectasia.

According to one embodiment, an assistive device for diagnosing keratoconus using artificial intelligence performed by a computing device includes: a memory that stores cornea-related image data based on digitized test data of a subject; and a processor, wherein the processor acquires image data related to the subject's cornea stored from the memory, and applies first input data and second input data based on the image data related to the subject's cornea to a first diagnostic model. Input, a first output value indicating whether keratoconus occurs in the subject's eye and a second output value indicating the expected probability of future progression of keratoconus are output, and the first output value indicates that the subject's cornea is suspected of having keratoconus (KC). Either keratoconus (KCS) or normal keratoconus (NM) may be output, and the second output value may output an expected probability of future keratoconus progression corresponding to each of the first output values.

The second input data may include at least one of the subject's questionnaire data, gene-related information, and the digitized test data.

At least one of the cornea-related image data and the second input data can be input into a second diagnostic model to output whether the subject has keratoconus and the corresponding expected probability of future keratoconus progression.

The quantified test data of the subject may include at least one of the curvature values of the front and back sides of the cornea, the thickness of the cornea, and the height of the cornea.

The cornea-related image data is generated based on digitized test data of the subject, and may include at least one of information about corneal topography, corneal thickness, and corneal elevation.

A method of assisting in diagnosing keratoconus using artificial intelligence performed by a computing device according to an embodiment includes cornea-related data of a subject, wherein the cornea-related data includes at least one of the subject's quantified test data and cornea-related image data. Obtaining a; And a step of detecting whether keratoconus has occurred in the eye of the subject by inputting the cornea-related data into a diagnostic model. The step of detecting whether keratoconus has occurred includes: detecting whether keratoconus has occurred in the eye of the subject; Based on relevant data, it can be classified as keratoconus (KC), suspected keratoconus (KCS), or normal cornea (NM).

When the cornea-related data includes cornea-related image data, the diagnostic model may be based on a CNN algorithm.

The above-described objects, features and advantages of the present invention will become more apparent through the following detailed description in conjunction with the accompanying drawings. However, since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail below.

In the drawings, the thicknesses of layers and regions are exaggerated for clarity, and elements or layers are referred to as “on” or “on” other elements or layers. This includes not only those directly on top of other components or layers, but also cases in which other layers or other components are interposed. Like reference numerals throughout the specification in principle refer to the same elements. In addition, components having the same function 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 a detailed description of a known function or configuration related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. In addition, numbers (eg, first, second, etc.) used in the description of this specification are merely identification symbols to distinguish one component from another component.

In addition, the suffixes “module” and “part” for components used in the following description are given or used interchangeably only for the ease of writing the specification, and do not have distinct meanings or roles in themselves.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

1 Keratoconus diagnostic assistance system and diagnostic assistance process

1.1 Definition of Terms

Below, a diagnostic assistance system and method for assisting in determining whether keratoconus occurs or whether there is an abnormality that serves as the basis for the determination based on cornea-related images will be described. In particular, a neural network model for diagnosing keratoconus is constructed using artificial intelligence, and a diagnostic assistance system or method that assists in detecting the presence of keratoconus or abnormal findings using the constructed model is described.

(1) Keratoconus (KC)

Keratoconus is a congenital corneal disease in which part of the cornea gradually becomes thinner, cannot maintain its gentle, round shape, and gradually protrudes into a cone shape. In addition to congenital causes, it can also be caused by acquired causes, such as the habit of rubbing the eyes frequently, wearing contact lenses, or existing diseases. Additionally, compared to normal corneas, keratoconus may show asymmetry between both eyes, sharp corneal curvature, and thin corneal thickness in cornea-related image data.

Figure 1 is a diagram for explaining keratoconus. Figure 1 (a) shows a normal cornea, and (b) shows keratoconus. I stands for iris, L stands for lens, and Co stands for cornea. In Figure 1 (a), a normal cornea has a constant corneal thickness and may have a thickness that can withstand a constant intraocular pressure. In Figure 1 (b), keratoconus has a relatively thin cornea, and the cornea may protrude in a cone shape due to intraocular pressure or other causes.

In this specification, keratoconus is generally described as a corneal disease caused by congenital factors, but it is not limited thereto and may also include acquired disease.

(2) Keratoconus suspect (KCS)

Although it is not keratoconus, it contains factors that cause keratoconus, and includes a stage that may later progress to keratoconus due to lifestyle habits, genetic factors, etc. In this specification, this is defined as suspected keratoconus.

Compared to normal corneas, suspected keratoconus may show asymmetry, sharp corneal curvature, and thin cornea in only one eye of both eyes in cornea-related image data. Additionally, corneal asymmetry in both eyes, somewhat gentle corneal curvature, and thinning cornea thickness may appear.

For example, when suspected keratoconus is judged based on both eyes, one of the two eyes may have a normal cornea and the other eye may have keratoconus. Alternatively, when suspected keratoconus is judged based on one eye, it may be a case where one eye has keratoconus.

(3) corneal ectasia (CE)

Although the cause of keratoconus has not been clearly identified, if the surgery is performed excessively without leaving enough residual corneal thickness after laser vision correction, the thinned cornea may not be able to withstand the pressure of the eye, leading to corneal expansion. In this specification, this is called corneal ectasia. define.

Corneal ectasia is often caused by expansion of the cornea after laser vision correction surgery in suspected keratoconus conditions with keratoconus-causing factors.

In some instances herein, corneal ectasia may be included in keratoconus. Corneal ectasia and keratoconus have the same symptoms, but there may be differences in the timing of occurrence.

(4) Keratoconus diagnostic assistance model

The keratoconus diagnostic assistance model refers to an algorithm and/or model that assists in diagnosing whether keratoconus occurs in a subject's cornea using an artificial neural network.

Of course, the keratoconus diagnostic assistance model described below refers to an algorithm and/or model that assists in diagnosing whether keratoconus occurs in the subject's cornea using various algorithms (e.g., machine learning algorithms) in addition to artificial neural networks. You can. For example, the keratoconus diagnostic assistance model uses various algorithms such as logistic regression, K-Nearest Neighbors, Support Vector Machine, and Decision Tree. This can assist in diagnosing the occurrence of keratoconus in the subject's cornea.

In addition, the keratoconus diagnosis assistance model described below is used when the subject's cornea-related data is input as input data through the model, and output data corresponding to the input data is used to determine whether keratoconus occurs in the subject's cornea (e.g., This is a model that derives information about keratoconus, suspected keratoconus, normal cornea, etc.).

Here, cornea-related data may refer to data that can indicate information about the subject's cornea. As an example, the cornea-related data may include the subject's examination data and cornea-related image data. The subject's examination data may output information about the subject's cornea obtained from various examination equipment. As an example, the subject's test data may include raw corneal imaging data. Corneal imaging raw data is the data itself output from the test equipment when the cornea is measured using various test equipment, and may refer to unprocessed data. As another example, the subject's examination data may include data obtained from various equipment such as tomography, topography, optical coherence tomography (OCT), ultrasound biomicroscopy (UBM) such as Pentacam, CASIA2, AL-Scan, and OQAS. The subject's test data may include raw data output from the test equipment and derived data derived from the raw data. As a more specific example, when the test data of the subject represents data output from Pentacam, the test data of the subject may include a file with the u21 extension of Pentacam. Of course, it is not limited to this, and the subject's test data may include information that can be obtained through the above equipment and similar equipment and information derived therefrom.

Additionally, cornea-related image data may represent image data generated based on the subject's examination data. For example, cornea-related image data is not output from various examination equipment, but may represent image data generated by processing the subject's examination data output from the various examination equipment. Of course, cornea-related image data may include image data output from various inspection equipment. Acquisition of cornea-related image data is described in detail below.

Hereinafter, the keratoconus diagnostic assistance model will be expressed as a diagnostic assistance model and a keratoconus diagnostic assistance model.

(5) Learning

Learning refers to the process of learning a keratoconus diagnostic assistance model based on learning data and labeled data or unlabeled data, so that the diagnostic assistance model can determine output data for the input data. In other words, the diagnostic assistance model forms rules for the data and makes decisions.

The diagnostic assistance model can be learned through training data. Learning a diagnostic assistance model means adjusting the weights of the model.

There are various learning methods, such as supervised learning, unsupervised learning, reinforcement learning, and imitation learning.

1.2 Keratoconus diagnostic assistance system

Figure 2 is a diagram showing the configuration of the keratoconus diagnosis assistance system 1. In one embodiment, the keratoconus diagnostic assistance system 1 may include a learning device 100 and a diagnostic assistance device 200.

The learning device 100 can learn a diagnostic assistance model. Specifically, the learning device 100 can learn a diagnostic assistance model based on learning data.

For example, the learning device 100 may use cornea-related data as learning data. For example, the learning device 100 may use image data related to the subject's cornea as learning data, and may also use examination data of the subject along with image data related to the cornea as learning data. Additionally, in some cases, the learning device 100 may use only the subject's test data as learning data.

Additionally, the learning device 100 can learn a diagnostic assistance model using various learning methods. For example, the learning device 100 can learn a diagnostic assistance model using methods such as supervised learning, unsupervised learning, reinforcement learning, and imitation learning. The learning device 100 can learn a diagnostic assistance model by providing labeled training data. However, labeled data is not necessarily used, and unlabeled data may be used.

The diagnostic assistance device 200 may receive and use the diagnostic assistance model learned from the learning device 100. Specifically, the diagnostic assistance device 200 may assist in diagnosing whether keratoconus occurs in the subject's eye using a learned diagnostic assistance model. Specifically, when the diagnostic assistance device 200 receives input data, such as data related to the subject's cornea, it may output information on whether keratoconus occurs in the subject's eye. Through the output of the diagnostic assistance device 200, the user can know that the subject's eye corresponds to any one of keratoconus (KC), suspected keratoconus (KCS), normal cornea (NM), and corneal ectasia (CE). .

The diagnostic assistance model learned using the learning data in the learning device 100 may be transmitted to the diagnostic assistance device 200. Of course, in Figure 2, the learning device 100 and the diagnostic assistance device 200 are shown as separate, but this is not limited, and in some cases, the learning device 100 and the diagnostic assistance device 200 are implemented separately. It can be implemented as one, without being separated. As an example, the diagnostic assistance device may be the same device as the learning device or may be a separate device.

3 is a diagram showing the configuration of a learning device and/or a diagnostic assistance device. Referring to FIG. 3, the learning device and/or diagnostic assistance device may include a memory 301, a control unit 303, and a communication unit 305.

The learning device and/or diagnostic assistance device may include a control unit 303. The control unit 303 may control the operation of the learning device and/or the diagnostic assistance device. The control unit 303 can read system programs and various processing programs stored in the memory unit 301.

The control unit 303 is one or more of a CPU (Central Processing Unit), RAM (Random Access Memory), GPU (Graphics Processing Unit), one or more microprocessors, and other electronic components capable of processing input data according to predetermined logic. may include.

The learning device and/or diagnostic assistance device may include a memory unit 301. The memory unit 301 can store data and learning models required for learning. Additionally, the memory unit 301 may store cornea-related data, that is, the subject's digitized test data and cornea-related image data.

The memory unit 301 may store learning data, labeled data, unlabeled data, input data, output data, etc.

The memory unit 301 may be a non-volatile semiconductor memory, hard disk, flash memory, RAM, ROM (Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), or other tangible non-volatile recording media. It can be implemented as:

The memory unit 301 may store various processing programs, parameters for processing the programs, or data resulting from such processing.

The learning device and/or diagnostic assistance device may further include a communication unit 305. The communication unit 305 can communicate with external devices. The communication unit 305 can perform wired or wireless communication. The communication unit 305 can perform two-way or one-way communication.

The learning device and/or diagnostic assistance device may include a processor, volatile memory, non-volatile memory, mass storage, and a communication interface. The processor may perform learning of the diagnostic assistance model through a learning device and/or a diagnostic assistance device.

Figure 4 is a schematic diagram of a keratoconus diagnosis assistance system using a server. Referring to FIG. 4, the keratoconus diagnosis assistance system may include a plurality of client devices and a server device. Below, the first client device 510 among the plurality of client devices will be described as an example, but the same operation can be performed on the second client device 520 as well.

The first client device 510 may request information from the server device 400 and obtain information about keratoconus diagnostic assistance transmitted in response, and may request keratoconus diagnostic assistance information from the server device 400. there is.

The first client device 510 may obtain data necessary to assist in diagnosing keratoconus and transmit the data obtained from the assisting diagnosis device.

The first client device 510 may be a portable device such as a smartphone or tablet PC.

The server device 400 may store and/or drive a diagnostic assistance model. The server device 400 may store weights that constitute the learned diagnostic assistance model. The server device 400 may collect and/or store data used to assist in diagnosing keratoconus.

The server device 400 may output the results of a diagnostic assistance process using a diagnostic assistance model to the first client device 510. The server device 400 may obtain feedback from the first client device 510.

In one embodiment, the first client device 510 may obtain a diagnostic assistance model from the server device 400 and drive the obtained diagnostic assistance model. In this case, the first client device 510 can obtain information about keratoconus diagnostic assistance by running a diagnostic assistance model without providing input data to the server device 400.

The server device 400 may communicate with a first client device 510 that obtains first diagnostic assistance information and/or a second client device 520 that obtains second diagnostic assistance information.

FIG. 5 is a diagram illustrating the relationship between the server device 400 and the client device 500. Referring to FIG. 5, the server device 400 may communicate with the client device 500 through a communication unit. The communication unit may perform wired or wireless communication. The communication unit can perform two-way or one-way communication. The client device 500 can also communicate with the server device 400 through a communication unit.

In one embodiment, when the client device 500 transmits the subject's input data to the server device, information regarding keratoconus diagnosis assistance of the subject's eye is received using the diagnostic assistance model learned in the server device 400. You can.

In one embodiment, the control unit 503 of the client device 500 obtains input data from the memory unit 501, and the obtained input data is sent to the communication unit 405 of the server device 400 through the communication unit 505. can be transmitted. In addition, the control unit 403 of the server device 400 obtains a result value by inputting input data into the diagnostic assistance model stored in the memory unit 401, and the obtained result value is obtained by using the communication unit 405 on the client device ( It may be transmitted to the communication unit 505 of 500).

Figure 6 is a diagram of a keratoconus diagnosis assistance process. Referring to FIG. 6, the keratoconus diagnostic assistance process is largely comprised of a learning step (S100) of learning a diagnostic assistance model and a diagnostic assistance process of performing diagnostic assistance on whether keratoconus occurs in the subject's cornea using the learned diagnostic assistance model. It can be considered divided into steps (S200).

Referring to FIG. 6, the learning step (S100) may be a process of learning a diagnostic assistance model using training data. Additionally, the learning step (S100) may be performed by a learning device.

According to one embodiment, the learning step (S100) may acquire training data and learn a diagnostic assistance model from the acquired data. That is, the learning step (S100) is a process of generating a diagnostic assistance model, and according to the creation of the diagnostic assistance model, model parameters constituting the diagnostic assistance model can be obtained. As an example, model parameters may include weights adjusted while training the diagnostic assistance model.

In one embodiment, the learning data may include cornea-related data of a person who was previously diagnosed with keratoconus, cornea-related data of a person who was previously diagnosed with suspected keratoconus, and cornea-related data of a person who was previously diagnosed with normal cornea. For example, the learning data may include image data including any one or more of test data of a person who was previously diagnosed with keratoconus, test data of a person who was previously diagnosed with suspected keratoconus, and test data of a person who was previously diagnosed with normal cornea. there is.

The diagnostic assistance model may be a model that outputs keratoconus diagnostic assistance information based on learning data. As the diagnostic assistance model, at least one of a plurality of learning algorithms that calculate keratoconus diagnostic assistance information may be selected. For example, the algorithm may be logistic regression, K-Nearest Neighbors, Support Vector Machine, Decision Tree, etc.

The diagnostic assistance model may use multiple learning algorithms among a plurality of learning algorithms that calculate predicted values. For example, an ensemble method may be used in a diagnostic assistance model, and compared to using learning algorithms separately, in one embodiment, the learning step (S100) uses a model to which random weights are assigned to produce results. This can be performed by obtaining a value (output data), comparing the obtained result value (output data) with the labeling data of the learning data, performing backpropagation according to the error, and optimizing the weights.

Although not shown, the learning step (S100) may include an evaluation step that evaluates the performance of the learned diagnostic assistance model. In the evaluation phase, the diagnostic assistance model may be evaluated using the evaluation data set. Evaluation of the diagnostic assistance model may be a step of evaluating the diagnostic assistance model learned through the learning step and making predictions about new data using the diagnostic assistance model. Specifically, the evaluation step may be a step to measure whether the learned diagnostic assistance model can generalize to new data.

Additionally, referring to FIG. 6, the diagnostic assistance step (S200) may use a diagnostic assistance model learned by obtaining model parameters in the learning step. Specifically, in the diagnostic assistance step (S200), after acquiring input data such as data related to the subject's cornea, information on whether keratoconus occurs in the subject's eye may be obtained using a learned diagnostic assistance model. Additionally, the diagnostic assistance step (S200) may be performed by a diagnostic assistance device.

In one embodiment, in the diagnosis assistance step (S200), the input data may include data related to the subject's cornea, that is, digitized test data of the subject and/or cornea-related image data based on the numerical test data. The result may include information on whether keratoconus occurs in the subject's eye. For example, it may include keratoconus (KC), suspected keratoconus (KCS), normal cornea (NM), etc.

Specific embodiments of the diagnostic assistance step (S200) and the diagnostic assistance process will be described in more detail with reference to FIGS. 16 and 17.

Figures 7 and 8 are diagrams showing an example of a diagnostic assistance model. Referring to FIG. 7, the diagnostic assistance model can output whether keratoconus occurs from input data.

Referring to Figures 7 (a) and (b), in one embodiment, the diagnostic assistance model may be implemented including a classifier and/or regression model.

In one embodiment, the diagnostic assistance model may be implemented in the form of a classifier that generates diagnostic assistance information. A classifier can perform binary classification or multiple classification. For example, the diagnostic assistance model may be a binary classification model that classifies input data as normal cornea or keratoconus with respect to diagnostic assistance information on whether keratoconus occurs. Alternatively, the diagnostic assistance model may be a multi-classification model that classifies input data into more specific classes. For example, it may be a multi-classification model that classifies the cornea of the subject's eye into normal cornea (NM), keratoconus (KC), suspected keratoconus (KCS), etc.

In one embodiment, the classifier may use a decision tree, support vector machine, random forest, or other type of algorithm. This is only an example and is not limited thereto.

In one embodiment, the diagnostic assistance model may be implemented in the form of a regression model that generates diagnostic assistance information. The regression model can derive the presence or absence of keratoconus in the form of probability. For example, results such as keratoconus (KC) 5%, suspected keratoconus (KCS) 85%, and normal cornea 10% may be derived. The regression model may use various types of algorithms such as linear regression, regression tree, support vector regression, and kernel regression. This is only an example and is not limited thereto.

In one embodiment, the diagnostic assistance model may represent a diagnostic assistance model that outputs diagnostic assistance information on whether keratoconus occurs based on cornea-related data. As an example, the diagnostic assistance model may represent a diagnostic assistance model that outputs diagnostic assistance information on whether keratoconus occurs based on cornea-related image data. The structure of a diagnostic assistance model for obtaining diagnostic assistance information may have a predetermined form. The diagnostic assistance model may include multiple layers or layers.

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

Referring to FIG. 8, the diagnostic assistance model may include a convolutional layer and an artificial neural network. The diagnostic assistance model can use cornea-related data, especially cornea-related image data, as input data, go through a convolutional layer and an artificial neural network, and output information on whether keratoconus occurs as a result.

In one embodiment, a convolutional layer may extract features from input data. Additionally, the convolutional layer can be comprised of a filter that extracts features and an activation function that changes the value of the filter that extracts features into a non-linear value. The filter may include a function that detects whether the input data includes the feature.

In one embodiment, the image data of the input data is converted into a matrix and then multiplied by the filter, so that if the input data has the feature, the filter produces a large result value, and if the input data does not have the feature, the result value is large. A value close to 0 appears, allowing you to know whether the input data has the above characteristics.

In one embodiment, because the input data has multiple characteristics, multiple filters, rather than one filter, can be applied together. If features are extracted from the convolutional layer, the extracted feature values can be input into an existing artificial neural network for classification. In other words, the diagnosis of keratoconus can be classified through an artificial neural network.

In one embodiment, the diagnostic assistance model may be implemented to include multiple VGGNet blocks. As a more specific example, the diagnostic assistance model includes a first structure in which a CNN layer with 64 filters of 3x3 size, a BN (Batch Normalization) layer, and a ReLu layer are sequentially combined, a CNN layer with 128 filters of 3x3 size, and a ReLu layer. And a second block in which BN layers are sequentially combined may be prepared by combining them.

In one embodiment, the diagnostic assistance model includes a max pooling layer following each CNN block, and at the end, a Global Average pooling (GAP) layer, a Fully Connected (FC) layer, and an activation layer (e.g., sigmoid, softmax). etc.) may be included.

Figure 9 is a diagram showing the configuration of a diagnostic assistance device. Referring to FIG. 9 , the diagnostic assistance device 200 may include a data input unit 201, a data generation unit 203, and a diagnostic unit 205.

The data input unit 201 can obtain the subject's test data as a numerical value from the test equipment.

The test equipment may be a plurality of test equipment capable of measuring cornea-related data. For example, multiple tests that can measure cornea-related data may include corneal topography, corneal tomography, etc.

The subject's examination data may include corneal thickness, corneal height, corneal curvature data, etc. The subject's test data can be derived as numerical values from the test equipment.

The data generator 203 may generate cornea-related image data using the cornea-related numerical data acquired by the data input unit 201.

In one embodiment, data such as a plurality of corneal curvature values, corneal thickness, and corneal height may be acquired through the data input unit 201.

Corneal-related numerical data may include data such as corneal curvature value, corneal thickness, and corneal height.

Corneal-related image data may include information on corneal topography, corneal thickness, and corneal elevation. A detailed description of cornea-related image data is given in FIGS. 10 to 13.

In addition, in FIG. 9, the data input unit 201 and the data generating unit 203 are shown separately, but the present invention is not limited thereto, and the data generating unit 203 may be included in the data input unit 201. Additionally, the data generation unit 203 may be omitted. For example, the data input unit 201 may receive cornea-related data from external examination equipment.

The diagnosis unit 205 may receive cornea-related image data obtained from the data input unit 201 or the data generation unit 203 and diagnose whether the subject has keratoconus. The diagnostic unit 205 may include a plurality of diagnostic assistance models. As an example, the diagnosis unit 205 may include one or more of a first diagnosis model, a second diagnosis model, a third diagnosis model, and a fourth diagnosis model. Embodiments of each diagnostic model will be described in detail in FIGS. 19 to 21.

In addition, in one embodiment, the diagnostic unit 205 receives cornea-related data, that is, the subject's quantified test data and/or cornea-related image data generated based on the quantified test data, and receives the cornea-related image data to form a cone of the subject's cornea. Information on whether the cornea has occurred can be derived. In one embodiment, cornea-related image data may have higher accuracy in the results of whether keratoconus occurs than quantified test data. The cornea-related image data may include a first group parameter, and the digitized test data may include a second group parameter. The first group parameters may represent parameters that are not defined in the test equipment that measures the numerical test data. In other words, the accuracy of the result of whether keratoconus occurs in the subject's eye is higher when data including group 1 parameters are input into the diagnostic assistance model than when data including group 2 parameters are input into the diagnostic assistance model. It can be high.

Additionally, in one embodiment, the diagnostic unit 205 outputs that the subject has either keratoconus or a normal cornea using the digitized test data and/or the cornea-related image data generated based on the numerical test data. can do. Alternatively, the diagnostic unit 205 may output that the subject corresponds to any one of keratoconus, suspected keratoconus, and normal cornea from image data related to the subject's cornea. Alternatively, the diagnosis unit 205 may output that the subject corresponds to any one of keratoconus, suspected keratoconus, normal cornea, and corneal ectasia from image data related to the subject's cornea. This may result in different output values from different diagnostic assistance models.

In one embodiment, the diagnostic unit 205 may output the probability of developing keratoconus in the future along with whether keratoconus occurs in the subject's cornea. For example, the diagnostic unit 205 uses the subject's quantified test data and/or cornea-related image data generated based on the quantified test data to output suspected keratoconus and the probability of developing keratoconus in the future. Alternatively, the probability of developing keratoconus in the future can also be output. In this way, by looking at the results of whether keratoconus has occurred and the probability of developing it in the future, the user can decide whether to proceed with laser vision correction surgery. For example, if the subject's result is an 80% probability of developing keratoconus, it may be determined that the subject cannot proceed with laser vision correction surgery. This is because it can expand into corneal ectasia.

10 to 13 are diagrams showing an example of cornea-related image data. Corneal-related image data can be used as input data for a diagnostic assistance model. Corneal-related image data may be generated based on the subject's quantified test data. This can be generated from the data generation unit 203 in step S2000 of FIG. 16, and can also be obtained from an external device. That is, cornea-related image data may be generated by the diagnostic assistance device 200 based on the subject's numerical examination data, or may be generated by an external device using the subject's numerical examination data.

10 and 11 are corneal topography maps and may include information about corneal curvature. The corneal topography may include information about corneal curvature (eg, front corneal curvature image, posterior corneal curvature image). The cornea is aspherical, not spherical, and a corneal topography map that connects points with the same curvature allows you to intuitively understand the shape and shape of the cornea.

Figure 10 is a diagram showing the corneal topography of keratoconus, and Figure 11 is a diagram showing the cornea topography of a normal cornea. This is only an example, and this limited corneal topography does not represent keratoconus and/or a normal cornea.

In one embodiment, corneal topography may vary in shape from region to region. The center of a normal cornea has radial symmetry and may include a corneal topography that becomes flatter towards the periphery. Keratoconus does not have a symmetrical structure compared to a normal cornea, and the slope may become steeper from the center to the periphery. Additionally, keratoconus may be suspected if there is localized steepening of the corneal topography.

In one embodiment, the degree of convexity of the cornea can be determined through corneal topography. In areas where the corneal curvature is large, the degree of convexity of the cornea may be large, and in areas where the corneal curvature is small, the degree of convexity of the cornea may be small.

Figure 12 is a diagram showing the thickness of the cornea. Referring to Figure 12, a normal cornea has a constant corneal thickness and can be expressed symmetrically, whereas keratoconus has an irregular corneal thickness, and the thickness of the cornea at a specific location is thinner than the thickness of the surrounding cornea. It can be expressed asymmetrically.

Figure 13 is a diagram showing the height of the cornea. Referring to Figure 13, since the cornea is not spherical, the height of the cornea may be different. In a normal cornea, the range of change in the height of the cornea may not be large, but in the case of keratoconus, the range of change in the height of the cornea may be large.

Figures 14 and 15 are diagrams showing the standardization process of input data. Specifically, Figures 14 and 15 can generate input data by standardizing cornea-related numerical data or image data.

Referring to FIG. 14, in one embodiment, image data may be normalized to generate input data.

In one embodiment, cornea-related image data may be generated based on quantified test data of the subject. Numerical data related to the subject's cornea may be a plurality of data. Due to the diversity of quantified test data related to the subject's cornea, cornea-related image data may also vary. Standardization or normalization may be necessary to use various cornea-related image data as input data. Normalization of image data may include cornea-related image resizing, image preprocessing, etc.

In one embodiment, the size of cornea-related image data may be adjusted. That is, images can be resized. Image resizing may be performed by the data input unit 201 or the data generation unit 203 described above.

Additionally, the size or aspect ratio of the cornea-related image may be adjusted. The size of the acquired plurality of cornea-related images may be adjusted to have a constant size. Alternatively, cornea-related images may be resized to have a constant aspect ratio. As an example, resizing of a cornea-related image may be performed by applying an image conversion filter to the cornea-related image.

If the size or resolution of the acquired individual images is excessively large or small, the size or resolution of the cornea-related image can be adjusted to convert it to an appropriate size and resolution. Alternatively, when the size or resolution of individual images varies, the size or resolution can be matched through resizing.

In one embodiment, the resolution of the cornea-related images may be adjusted. For example, if the capacity of the cornea-related image exceeds an appropriate range, the resolution of the cornea-related image can be reduced through down sampling. Alternatively, in cases where the capacity of the cornea-related image falls short of an appropriate range, the resolution of the image can be increased through upsampling or interpolating.

In one embodiment, when there is only some cornea-related image data, the cornea-related image data can be generated by estimating or interpolating it. For example, if there is only 3 image data related to the subject's cornea among 10 cornea-related image data, cornea-related image data can be generated through estimation and interpolation using cornea-related quantified test data.

In one embodiment, standardization of cornea-related images may be performed. If cornea-related images are used for learning as they are input, overfitting of learning results for unnecessary characteristics may occur, and learning efficiency may also decrease. In addition, when cornea-related images are used in a diagnostic assistance model learned as input, the accuracy of the results may decrease and errors in the result values may increase.

In order to prevent this, cornea-related image data can be appropriately standardized and used to meet the purpose of learning and using the learned model, thereby improving learning efficiency, improving accuracy of results, and reducing errors in result values.

Referring to FIG. 15 , in one embodiment, the test data of the subject, that is, cornea-related numerical data, may be standardized, and cornea-related image data may be generated based on the standardized cornea-related numerical data.

Normalization of cornea-related numerical data may be intended to generate standardized cornea-related image sizes. When inputting standardized cornea-related image data into a diagnostic assistance model, it can serve as a reference value for size and resolution that can represent maximum efficiency in using it.

In one embodiment, cornea-related numerical data may be standardized through adjustment of the size of the cornea-related numerical data. For example, the cornea-related numerical data may be adjusted so that the value of each parameter of the cornea-related numerical data falls within a predetermined range. For example, the size of the cornea-related numerical data may be adjusted by the data input unit 201 or the data generation unit 203 described above.

In one embodiment, cornea-related image data may be standardized based on a specific parameter among cornea-related numerical data. That is, the values of other parameters of the cornea-related numerical data may be adjusted based on the value of a specific parameter among the cornea-related numerical data. As an example, the specific parameter may be intraocular pressure. In this case, the values of other parameters of the cornea-related numerical data may be standardized based on the value of intraocular pressure among the cornea-related numerical data. This is because, in some cases, other parameters of cornea-related numerical data, such as corneal thickness, may appear differently depending on intraocular pressure. Accordingly, as the values of other parameters of the cornea-related numerical data are adjusted based on intraocular pressure, cornea-related numerical data that is not affected by intraocular pressure can be generated. In this way, standardized cornea-related image data can be generated based on cornea-related data in which the values of other parameters are adjusted based on the values of a specific parameter.

In this way, by appropriately standardizing and using cornea-related image data to meet the purpose of learning and using the learned model, learning efficiency can be improved, accuracy of results can be improved, and errors in result values can be reduced.

In addition, although not shown in Figures 14 and 15, it is possible to use the standardized cornea-related numerical data directly as input data for learning and diagnosis without changing the standardized cornea-related numerical data to cornea-related image data. Of course.

1.3 Keratoconus diagnostic assistance process

Figures 16 and 17 are flowcharts of keratoconus diagnosis assistance steps. Referring to FIG. 16, the keratoconus diagnosis assistance step includes acquiring test data (numerical values) (S1000), generating cornea-related image data based on the acquired test data (numerical values) (S2000), and cornea It may include a step (S3000) of inputting related image data into a diagnosis model and obtaining a diagnosis result. The keratoconus diagnosis assistance step may be performed by the diagnosis assistance device 200 described above in FIG. 9 .

In one embodiment, the step of acquiring inspection data (S1000) may acquire quantified inspection data. The step of acquiring test data (S1000) may be performed by the diagnostic assistance device 200 described above in FIG. 9. However, this is only an example and is not limited to this, and inspection data can be obtained and transmitted from an external device.

Numerical test data may be result values obtained from test equipment. For example, quantified inspection data may include data obtained from various equipment such as tomography, topography, optical coherence tomography (OCT), ultrasound biomicroscopy (UBM) such as Pentacam, CASIA2, AL-Scan, and OQAS. Of course, it is not limited to this, and the numerical inspection data may include information that can be obtained through the above equipment and similar equipment and information derived therefrom.

More specifically, the numerical test data may include information about the physical shape of the eye. As an example, the quantified test data includes white-to-white (WTW) distance, angle-to-angle (ATA) distance, internal anterior chamber depth (ACD), sulcus-to-sulcus (STS) distance, and pupil. This may include size, etc.

Additionally, the numerical test data may include information about the shape of the cornea, such as corneal shape factors. Here, the corneal shape factor is a value that represents the physical shape of the cornea, including corneal curvature value, corneal thickness, corneal height, index of surface variance (ISV), index of vertical asymmetry (IVA), keratoconus index (KI), and central keratoconus index. (CKI), minimum radius of curvature (Rmin), index of height asymmetry (IHA), index of height decentration (IHD), central cornea thickness, etc.

In one embodiment, the step of generating cornea-related image data based on the acquired inspection data (S2000) may be performed by the data generator 203 described above in FIG. 9.

In one embodiment, the step (S3000) of obtaining a diagnosis result by inputting cornea-related image data into a diagnosis model may be performed by the diagnosis unit 205 described above in FIG. 9.

The diagnostic assistance result may be information about whether keratoconus occurs in the subject's cornea. For example, if image data related to the subject's cornea is input into a diagnostic model, any one of the following results can be derived: keratoconus, suspected keratoconus, normal cornea, and corneal ectasia.

Additionally, in one embodiment, in step S3000, the diagnostic assistance device 200 may obtain a diagnostic result by inputting numerical data, along with cornea-related image data, into a diagnostic model. For example, the diagnostic assistance device 200 can input corneal formation factors such as IHD and ISV along with cornea-related image data into a diagnostic model, and the diagnostic model can be used as a diagnosis result, such as keratoconus, suspected keratoconus, and normal cornea. and corneal ectasia can be derived. In some cases, the accuracy of diagnostic results can be improved by inputting not only cornea-related image data but also numerical data into the diagnostic model.

In addition, although not shown in FIGS. 16 and 17, the diagnostic assistance device 200 can input only numerical data into the diagnostic model instead of cornea-related image data, and the diagnostic model includes diagnostic results such as keratoconus, suspected keratoconus, The result of either normal cornea or corneal ectasia can be derived.

Referring to FIG. 17, the keratoconus diagnosis assistance step includes acquiring cornea-related image data based on numerical test data (S1100) and inputting the cornea-related image data into a diagnostic model to obtain a diagnosis result (S3000). may include. The keratoconus diagnosis assistance step may be performed by the diagnosis assistance device 200 described above in FIG. 9 .

In one embodiment, cornea-related image data based on the subject's digitized test data may be acquired by the data input unit 201 or the data generation unit 203 described above in FIG. 9 . This is just an example, and can also be acquired and transmitted by an external device.

Compared to FIG. 16 , the keratoconus diagnosis assistance step may omit the step (S1000) of acquiring numerical test data.

The step of obtaining the diagnosis result (S3000) is omitted as the description overlaps with that of FIG. 16.

Example 2

18 to 20 are diagrams according to an embodiment of keratoconus diagnosis assistance.

Referring to FIG. 18, the first diagnostic model may receive first input data and output a first output value. The first diagnostic model may be a binary classification or multi-classification form of the diagnostic assistance model.

In one embodiment, the first diagnostic model may be learned based on image data related to the cornea of a subject who has been diagnosed in the past. The learning data of the first diagnostic model may include cornea-related data of a person who was previously diagnosed with keratoconus, cornea-related data of a person who was previously diagnosed with suspected keratoconus, and cornea-related data of a person who was previously diagnosed with normal cornea. For example, the learning data includes cornea-related image data of a person who was previously diagnosed with keratoconus, cornea-related image data of a person who was previously diagnosed with suspected keratoconus, cornea-related image data of a person who was previously diagnosed with normal cornea, and cornea-related image data of a person who was previously diagnosed with corneal ectasia. It may include one or more cornea-related image data. As for the learning step of the first diagnostic model, the learning step (S100) described in FIG. 6 may be applied as is.

Referring to FIG. 19, the first diagnostic model may receive first input data and output a first output value. And, depending on what value the first output value has, a second diagnosis model or a third diagnosis model may be driven. Here, the first output value may be two different result values.

In one embodiment, the second diagnostic model and the third diagnostic model may not operate simultaneously. For example, depending on what value the first output value has, either the second diagnosis model or the third diagnosis model may be operated.

In one embodiment, the first diagnostic model may classify the cornea of the subject's eye as a normal cornea or keratoconus based on cornea-related data of the subject. That is, the first output value may be normal cornea (NM) or keratoconus (KC).

In one embodiment, when the first output value is a normal cornea (NM), the second diagnostic model may operate. In this case, the subject's cornea-related data is input to the second diagnostic model, and the second diagnostic model is based on the subject's cornea-related data, and as a second output value, the subject's cornea is classified as normal cornea (NM) and suspected keratoconus ( KCS) can be output. That is, the second diagnostic model can classify the subject's cornea as normal cornea (NM) or suspected keratoconus (KCS).

In one embodiment, when the first output value is keratoconus (KC), a third diagnostic model may operate. In this case, the subject's cornea-related data is input to the second diagnostic model, and the third diagnostic model is based on the subject's cornea-related data, and as a third output value, the subject's cornea is diagnosed with keratoconus (KC) and corneal ectasia ( CE) can be output. That is, the third diagnostic model can classify the subject's cornea as keratoconus (KC) or corneal ectasia (CE).

Additionally, the first diagnostic model may be learned through a process of inputting data related to the subject's cornea and outputting that the subject's cornea is a normal cornea or keratoconus. As for the learning step of the first diagnostic model, the learning step (S100) described in FIG. 6 may be applied as is.

The second diagnostic model may be learned using a process of inputting cornea-related data classified as normal corneas and outputting normal corneas or suspected keratoconus. That is, the second diagnostic model can classify the subject's corneal condition more specifically than the first diagnostic model. As for the learning step of the second diagnostic model, the learning step (S100) described in FIG. 6 may be applied as is.

The third diagnostic model may be learned using a process of inputting cornea-related image data classified as keratoconus and outputting keratoconus or corneal ectasia. That is, the third diagnostic model can classify the subject's corneal condition more specifically than the first diagnostic model. As for the learning step of the third diagnostic model, the learning step (S100) described in FIG. 6 may be applied as is.

Referring to FIG. 20, the first diagnostic model receives first input data and outputs a first output value, and the fourth diagnostic model receives the first output value and/or the first input data output from the first diagnostic model. By receiving the second input data, a result value regarding the probability of future occurrence can be output. The diagnostic model of FIG. 20 is intended to provide not only whether keratoconus occurs, which is output through the diagnostic model described in FIGS. 18 and 19, but also the probability of future onset. By providing a predicted probability of future onset, subjects can pay attention to corneal or eye care.

In one embodiment, the first diagnostic model may classify the cornea of the subject's eye into at least one of keratoconus, suspected keratoconus, normal cornea, and corneal ectasia. That is, the first output value may indicate any one of keratoconus, suspected keratoconus, normal cornea, and corneal ectasia.

In one embodiment, the fourth diagnostic model may receive second input data that is separate from the first output value output from the first diagnostic model and output a result value regarding the probability of future onset. For example, the first diagnostic model receives first input data and outputs a result indicating that the subject has a normal cornea as a first output value, and the fourth diagnostic model receives the first output value and/or data related to the subject's cornea. Together with the second input data, the probability of developing keratoconus in the future can be output.

In one embodiment, the second input data may include medical examination data of the subject. The subject's interview data may include the subject's interview information, for example, information related to lifestyle habits (whether there is a habit of rubbing the eyes, number of times of rubbing the eyes per day, etc.), and information related to diseases (atopy, skin disease, eye disease, etc.). You can. Additionally, the second input data may include information related to the subject's genes. Additionally, when the first input data is cornea-related image data, the second input data may be the numerical test data described above. As a specific example, the second input data may include data acquired from various equipment. For example, the second input data may include the above-described WTW distance, ATA distance, ACD, STS distance, pupil size, and the above-described corneal shape factor. (Corneal curvature value, corneal thickness, corneal height, ISV, IVA, KI, CKI, Rmin, IHA, IHD, central corneal thickness, etc.). Additionally, when the first input data is numerical test data of the subject, the second input data may be cornea-related image data.

In FIG. 20, the first diagnosis model and the fourth diagnosis model are shown separately, but the present invention is not limited thereto, and the fourth diagnosis model may be included in the first diagnosis model. When the first input data and the second input data are input to the first diagnosis model, the first output value and the corresponding probability of future onset can be output.

Claims (14)

In the keratoconus diagnostic assistance device using artificial intelligence performed by a computing device,
Including a processor;
The processor,
Obtaining quantified test data of the subject, wherein the test data includes information about the shape of the subject's cornea,
Converting the digitized inspection data into first cornea-related image data, each of the cornea-related image data having a predetermined aspect ratio,
If the number of first cornea-related image data is less than a specified number, generate second cornea-related image data through estimation and interpolation using the numerical inspection data,
Adjusting the predetermined aspect ratio of each of the first cornea-related image data and the second cornea-related image data to the same value,
Input the first cornea-related image data and the second cornea-related image data having the aspect ratio adjusted to the same value as input data to a first diagnostic model to detect whether keratoconus occurs in the subject's eye,
The occurrence of keratoconus is classified as keratoconus (KC), suspected keratoconus (KCS), or normal keratoconus (NM),
Based on the detection result of the occurrence of keratoconus, if the subject's eyeball is a normal cornea, the quantified test data of the subject and the subject's genetic data are input into a second diagnostic model to determine whether the subject's keratoconus is occurring. Print the probability,
The quantified test data include WTW distance, ATA distance, ACD, STS distance, pupil size, or corneal shape factor,
The corneal shape factor includes corneal curvature value, corneal thickness, corneal height, ISV, IVA, KI, CKI, Rmin, IHA, IHD, or central corneal thickness,
Keratoconus diagnostic aid device.
According to claim 1,
The first diagnostic model is,
Learned based on image data including one or more of the following: test data of a person who was previously diagnosed with keratoconus, test data of a person who was previously diagnosed with suspected keratoconus, and test data of a person who was previously diagnosed with normal cornea.
Keratoconus diagnostic aid device.
According to claim 1,
The first cornea-related image data includes a first group parameter,
The quantified test data include a second group parameter,
The first group parameter is generated based on the second group parameter,
The first group parameters represent parameters that are not defined in the inspection equipment that measures the numerical inspection data,
The accuracy of the results of whether keratoconus occurs in the subject's eye is,
The case where data including the first group parameter is input into the first diagnostic model is higher than when data including the second group parameter is input into the first diagnostic model.
Keratoconus diagnostic aid device.
According to paragraph 3,
The processor,
Inputting the numerical test data together with the first cornea-related image data into the first diagnostic model to detect whether keratoconus occurs in the subject's eye,
Keratoconus diagnostic aid device. delete delete delete delete delete delete delete delete delete delete