KR20230101126A - Apparatus, system and method for diagnosing glaucoma - Google Patents

Abstract

A glaucoma diagnosis device, system and method are disclosed. An apparatus for diagnosing glaucoma according to an embodiment includes a data input unit receiving clinical information of a subject; and a diagnostic unit diagnosing whether the test subject has glaucoma from the clinical information using a machine learning-based glaucoma diagnosis model and generating a graphic chart to explain the reason for the diagnosis result; The clinical information includes the retinal nerve fiber layer upper thickness, the retinal nerve fiber layer lower thickness, the retinal nerve fiber layer temporal thickness, and the pattern standard deviation obtained through intraocular pressure and visual field tests.

Description

Apparatus, system and method for diagnosing glaucoma}

It is related to the technology for diagnosing glaucoma.

Glaucoma is the leading cause of irreversible blindness worldwide and progressively affects the optic nerve. Currently, it is possible to detect elevated intraocular pressure, evaluate optic disc damage by calculating cup-to-disc ratio (CDR), identify reduced retinal nerve fiber layer thickness, and characterize visual field defects. It is diagnosed through four tests, such as detection. Structural and functional clinical modalities such as optical coherence tomography (OCT) and visual field tests provide indicators or values that can be used for glaucoma diagnosis. However, since these existing glaucoma diagnosis methods have a possibility of misdiagnosis, it is necessary to develop technology to prevent misdiagnosis.

Korean Patent Publication No. 10-2019-0087272 (2019.07.24.)

An object of the present invention is to provide a device, system, and method for diagnosing glaucoma based on machine learning.

An apparatus for diagnosing glaucoma according to an aspect includes: a data input unit for receiving clinical information of a subject; and a diagnostic unit diagnosing whether the test subject has glaucoma from the clinical information using a machine learning-based glaucoma diagnosis model and generating a graphic chart to explain the reason for the diagnosis result; The clinical information may include the upper retinal nerve fiber layer thickness, the retinal nerve fiber layer lower thickness, the retinal nerve fiber layer temporal thickness, and the pattern standard deviation obtained through intraocular pressure and visual field tests.

The diagnosis unit may generate a graphic chart representing a position of a value of clinical information of the subject within the distribution of clinical information used to generate the glaucoma diagnosis model or expressing a role of each clinical information in the diagnosis process. there is.

The graphic chart may include a gauge chart, a radial chart, and a Shapley Additive Explanations (SHAP) chart.

The glaucoma diagnosis model may be generated using XGboost.

The apparatus for diagnosing glaucoma may include: a feedback input unit receiving a final diagnosis result for the test subject as feedback; and a model updater configured to update the glaucoma diagnosis model based on the final diagnosis result and the clinical information when a final diagnosis result input as the feedback is different from a diagnosis result of the diagnosis unit; may further include.

A method for diagnosing glaucoma according to another aspect includes receiving clinical information of a subject; diagnosing whether or not the subject has glaucoma from the clinical information using a machine learning-based glaucoma diagnosis model; and generating a graphic chart to explain the reason for the diagnosis result. The clinical information may include the upper retinal nerve fiber layer thickness, the retinal nerve fiber layer lower thickness, the retinal nerve fiber layer temporal thickness, and the pattern standard deviation obtained through intraocular pressure and visual field tests.

The generating of the graphic chart may include a graphic representing a position of a value of clinical information of the subject within the distribution of clinical information used to generate the glaucoma diagnosis model or a role of each clinical information in the diagnosis process. You can create charts.

The graphic chart may include a gauge chart, a radial chart, and a Shapley Additive Explanations (SHAP) chart.

The glaucoma diagnosis model may be generated using XGboost.

The method for diagnosing glaucoma may include receiving a final diagnosis result of the test subject as feedback; and updating the glaucoma diagnosis model based on the final diagnosis result and the clinical information when a final diagnosis result received as the feedback and a diagnosis result using the glaucoma diagnosis model are different; may further include.

A glaucoma diagnosis system according to another aspect collects clinical information about a plurality of patients and corresponding glaucoma diagnosis results as learning data, and generates a glaucoma diagnosis model by learning a machine learning model based on the collected learning data. a glaucoma diagnostic model generating device; and a glaucoma diagnosis device that receives clinical information of the subject, diagnoses whether or not the subject has glaucoma from the clinical information of the subject using the glaucoma diagnosis model, and generates a graphic chart to explain the reason for the diagnosis result; The clinical information may include the upper retinal nerve fiber layer thickness, the retinal nerve fiber layer lower thickness, the retinal nerve fiber layer temporal thickness, and the pattern standard deviation obtained through intraocular pressure and visual field tests.

By generating and using a glaucoma diagnosis model based on machine learning, it is possible to diagnose glaucoma easily and accurately.

In addition, by generating and providing a graphic chart to explain the reason for the diagnosis result, it is possible to help the clinician understand the diagnosis result.

1 is a diagram illustrating a system for diagnosing glaucoma according to an exemplary embodiment.
2 is a diagram illustrating an apparatus for generating a glaucoma diagnostic model according to an exemplary embodiment.
3 is a diagram illustrating an apparatus for diagnosing glaucoma according to an exemplary embodiment.
4 is a diagram illustrating a gauge type chart according to an exemplary embodiment.
5 is a diagram illustrating a radial chart according to an exemplary embodiment.
6 is a diagram illustrating a SHAP chart according to an embodiment.
7 is a diagram illustrating an apparatus for diagnosing glaucoma according to an exemplary embodiment.
8 is a diagram illustrating a method for generating a glaucoma diagnosis model according to an embodiment.
9 is a diagram illustrating a method for diagnosing glaucoma according to an exemplary embodiment.
10 is a diagram illustrating a method for diagnosing glaucoma according to an exemplary embodiment.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description will be omitted.

Meanwhile, in each step, each step may occur in a different order from the specified order unless a specific order is clearly described in context. That is, each step may be performed in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

Terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a user or operator. Therefore, the definition should be made based on the contents throughout this specification.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. Terms are only used to distinguish one component from another. Singular expressions include plural expressions unless the context clearly dictates otherwise, and terms such as 'include' or 'have' refer to features, numbers, steps, operations, components, parts, or combinations thereof described in the specification. It is intended to specify that something exists, but it should be understood that it does not preclude the possibility of the existence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

In addition, the division of components in the present specification is merely a classification for each main function in charge of each component. That is, two or more components may be combined into one component, or one component may be divided into two or more for each more subdivided function. In addition, each component may additionally perform some or all of the functions of other components in addition to its main function, and some of the main functions of each component are dedicated to other components. may be performed. Each component may be implemented in hardware (eg, memory, processor, etc.) or software, or a combination of hardware and software.

1 is a diagram illustrating a glaucoma diagnosis system according to an embodiment, FIG. 2 is a diagram illustrating a glaucoma diagnosis model generation device according to an embodiment, and FIG. 3 illustrates a glaucoma diagnosis device according to an embodiment. it is a drawing

Referring to FIG. 1 , a glaucoma diagnosis system 100 according to an exemplary embodiment may include a glaucoma diagnosis model generation device 110 and a glaucoma diagnosis device 120 .

The glaucoma diagnosis model generating device 110 may generate a glaucoma diagnosis model based on a machine learning algorithm. At this time, the machine learning algorithm is a tree-based machine learning algorithm (eg, random forest, XGboost (Extreme Gradient Boost), ADAboost (Adaptive Boost), Light GBM (Light Gradient Boost Machine), C5.0, ID3, C4.5, CART etc.), deep learning algorithm, K-Nearest Neighbor Algorithm, Naïve Bayes Classification algorithm, Neural Networks algorithm, Support Vector Machines, etc., but is not limited thereto. .

As shown in FIG. 2 , the glaucoma diagnosis model generation apparatus 110 may include a learning data collection unit 210 and a model generation unit 220 .

The learning data collection unit 210 may collect learning data to be used for generating a glaucoma diagnosis model.

More specifically, the learning data collection unit 210 may collect clinical information on a plurality of patients and a glaucoma diagnosis result corresponding thereto from an external device as learning data. Here, the clinical information is information obtained through a visual field test, a retinal nerve fiber layer (RNFL) optical coherence tomography (OCT) test, an intraocular pressure test, etc., and the pattern standard deviation ( pattern standard deviation), retinal nerve fiber layer (RNFL) superior thickness, retinal nerve fiber layer inferior thickness, retinal nerve fiber layer temporal thickness, and intraocular pressure. can Here, the pattern standard deviation may be obtained through a visual field test, and may be an index showing how much sensitivity difference each test point has compared to adjacent test points. The upper retinal nerve fiber layer thickness, the retinal nerve fiber layer inferior thickness, and the retinal nerve fiber layer temporal thickness may be obtained through a retinal nerve fiber layer optical coherence tomography test.

At this time, the learning data collection unit 210 may use various wired/wireless communication technologies to obtain learning data from an external device. The external device may be a device or server that stores electronic health records (eg, electronic health record (EHR) or electronic medical record (EMR)).

If a missing value exists in the collected clinical information, the learning data collection unit 210 may process the missing value. According to an embodiment, the learning data collection unit 210 processes missing values using various methods, such as a method of deleting missing values, a method of replacing missing values, and a method of predicting missing values with a data set composed of variables without missing values. can do.

The model generation unit 220 may generate a glaucoma diagnosis model based on the collected learning data.

For example, the model generating unit 220 may generate pattern standard deviation, upper retinal nerve fiber layer thickness, lower retinal nerve fiber layer thickness, temporal retinal nerve fiber layer thickness, and intraocular pressure for a plurality of patients, and corresponding glaucoma diagnosis results. Based on this, a glaucoma diagnosis model can be created by learning a machine learning model. At this time, the machine learning model is a tree-based machine learning model (e.g., random forest, XGboost, ADAboost, Light GBM, C5.0, ID3, C4.5, CART, etc.), deep learning, K-nearest neighbors, and Naive Bayes classification , neural networks, support vector machines, and the like. More preferably, the machine learning model may be a tree-based machine learning model, for example, XGboost.

Meanwhile, according to an embodiment, the model generating unit 220 may optimize hyperparameters of the machine learning model. For example, the model generation unit 220 may optimize the hyperparameters of the machine learning model through a Bayesian optimization algorithm.

The glaucoma diagnosis apparatus 120 may diagnose whether a test subject has glaucoma by using the glaucoma diagnosis model generated by the glaucoma diagnosis model generation apparatus 110 .

As shown in FIG. 3 , the glaucoma diagnosis apparatus 120 may include a data acquisition unit 310 , a storage unit 320 and a risk calculation unit 330 .

The data input unit 310 may receive clinical information of the test subject used for glaucoma diagnosis from a user. As described above, the clinical information may include pattern standard deviation, upper retinal nerve fiber layer thickness, lower retinal nerve fiber layer thickness, retinal nerve fiber layer temporal thickness, intraocular pressure, and the like.

According to an embodiment, the data input unit 310 includes a key pad, a dome switch, a touch pad, a jog wheel, a jog switch, and H/W. buttons and the like. In particular, when a touch pad forms a mutual layer structure with a display, it may be referred to as a touch screen.

The storage unit 320 may store the glaucoma diagnosis model generated by the glaucoma diagnosis model generating device 110 . The storage unit 320 is a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (eg, SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read Only Memory), PROM (Programmable Read Only Memory), magnetic memory, magnetic disk, optical disk At least one type of storage medium may be included.

The diagnosis unit 330 may diagnose whether or not the subject has glaucoma by using the clinical information of the subject input through the data input unit 310 and the glaucoma diagnosis model stored in the storage unit 320 . More specifically, the diagnosis unit 330 may input clinical information of the test subject into a glaucoma diagnosis model to diagnose glaucoma.

The diagnosis unit 330 may generate and output a graphic chart through an output means to explain the reason for the diagnosis result. In this case, the graphic chart may represent the position of the value of the subject's clinical information in the clinical information distribution of the learning data and/or the role of each clinical information in the diagnosis process of the glaucoma diagnosis model. For example, the graphic chart may include a gauge chart, a radial chart, a Shapley Additive Explanations (SHAP) chart, and the like.

4 is a diagram illustrating a gauge type chart according to an exemplary embodiment. Specifically, FIG. 3 is a gauge chart showing the upper retinal nerve fiber layer thickness (RNFL_S) among clinical information.

Referring to FIG. 4, 0 and 170 on the dial represent the minimum and maximum values of RNFL_S in the distribution of training data used to generate the glaucoma diagnosis model, and 64 and 155 represent the overlapping range between glaucoma patients and normal subjects in the distribution of learning data. boundary values can be indicated. The value 86 at the bottom center of the gauge chart represents the RNFL_S value for the subject and can determine the angle of the needle on the dial. The range of 0 to 64 on the dial represents the range of glaucoma patients, the range of 155 to 170 represents the range of normal people, and the range of 64 to 155 represents the overlapping range of glaucoma patients and normal people.

According to an embodiment, the glaucoma diagnosis apparatus 120 expresses each clinical information in a gauge-type chart, thereby intuitively confirming to which range the value of each clinical information of the subject belongs and understanding the diagnosis result of the glaucoma diagnosis model. It can help clinicians who are users.

5 is a diagram illustrating a radial chart according to an exemplary embodiment. In FIG. 5 , a chart on the left may be for a glaucoma patient, and a chart on the right may be for a normal person.

As shown in FIG. 5, a radar chart is a chart visualizing multivariate data in the form of a two-dimensional chart of three or more quantitative variables represented by axes starting from the same point.

Comparing the left chart (glaucoma patients) and the right chart (normal people) of FIG. 5, the glaucoma patients had higher pattern standard deviation (PSD) and intraocular pressure (IOP) than normal people, and the upper retinal nerve fiber layer thickness (RNFL_S) and retinal nerve It can be seen that the lower fiber layer thickness (RNFL_I) and the retinal nerve fiber layer temporal thickness (RNFL_T) are lower.

According to an embodiment, the glaucoma diagnosis apparatus 120 expresses five types of clinical information of a subject in a radial chart, thereby helping a clinician, a user, to make a diagnosis of glaucoma and to understand the diagnosis result of the glaucoma diagnosis model. can

6 is a diagram illustrating a SHAP chart according to an embodiment.

The SHAP chart of FIG. 6 is based on the Shapley value, the vertical axis represents each clinical information for the subject and the corresponding value, and the horizontal axis represents the contribution to glaucoma. A positive contribution means that the corresponding clinical information contributes to the diagnosis of glaucoma, and a negative contribution means that the corresponding clinical information contributes to the diagnosis of a normal person. It may mean that the longer the bar representing the contribution, that is, the greater the absolute value of the contribution, the stronger the contribution.

At the top of the SHAP chart, the diagnosis result and degree of certainty can be displayed. In the case of the example of FIG. 6 , the diagnosis result for the subject is glaucoma, and the degree of certainty is 0.96 (96%).

Referring to FIG. 6, the pattern standard deviation (PSD) value of 10.51 strongly contributes to the diagnosis of glaucoma, as does the retinal nerve fiber layer temporal thickness (RNFL_T) value of 47, and the retinal nerve fiber layer inferior thickness (RNFL_I) value of 119 contributes to the diagnosis of normal individuals. It can be seen that the retinal nerve fiber layer superior thickness (RNFL_S) value of 86 and intraocular pressure (IOP) value of 18 contribute weakly to the diagnosis of glaucoma.

According to an embodiment, the glaucoma diagnosis apparatus 120 may show the reason why the glaucoma diagnosis model diagnosed the subject as a glaucoma patient to the user, the clinician, by expressing the contribution of each clinical information to the diagnosis in a SHAP chart.

Meanwhile, in order to generate a SHAP chart, the diagnosis unit 330 may calculate a Shapley value of each clinical information.

7 is a diagram illustrating an apparatus for diagnosing glaucoma according to an exemplary embodiment.

The glaucoma diagnosis apparatus 700 of FIG. 7 may be another embodiment of the glaucoma diagnosis apparatus 12 of FIG. 1 .

Referring to FIG. 7 , the glaucoma diagnosis apparatus 700 includes a data input unit 310, a storage unit 320, a diagnosis unit 330, a feedback input unit 710, a model updating unit 720, a communication unit 730, and thrust. A portion 740 may be included. Since the data input unit 310, the storage unit 320, and the diagnosis unit 330 are the same as those described above with reference to FIG. 3, detailed descriptions thereof will be omitted.

The feedback input unit 710 may receive the clinician's final diagnosis result for the test subject as feedback. According to one embodiment, the feedback input unit 710 includes a key pad, a dome switch, a touch pad, a jog wheel, a jog switch, H/W buttons and the like.

When the diagnosis result of the diagnosis unit 330 and the final diagnosis result input from the feedback input unit 710 are different, the model updating unit 720 uses the clinical information of the subject input through the data input unit 310 and the feedback input unit 710. ), the pre-stored glaucoma diagnosis model may be updated based on the final diagnosis result input. For example, the model updating unit 720 uses the clinical information of the subject input through the data input unit 310 and the final diagnosis result input through the feedback input unit 710 as new learning data, and adds a pre-stored glaucoma diagnosis model. can be learned Accordingly, the diagnostic accuracy of the glaucoma diagnosis model can be increased.

The communication unit 730 may communicate with an external device. For example, the communication unit 730 may transmit data input to the glaucoma diagnosis apparatus 700, stored data, and processed data to an external device, or may receive various data useful for glaucoma diagnosis from the external device.

At this time, the external device may be the glaucoma diagnosis model generating device 110, medical equipment using the data input to the glaucoma diagnosis device 700, stored data, processed data, etc., and a printing or display device for outputting results. It could be. In addition, the external device may be a digital TV, desktop computer, mobile phone, smart phone, tablet, laptop, PDA (Personal Digital Assistants), PMP (Portable Multimedia Player), navigation device, MP3 player, digital camera, wearable device, etc. Not limited.

The communication unit 730 may communicate with an external device using wired/wireless communication technology. At this time, the wireless communication technology includes Bluetooth communication, BLE (Bluetooth Low Energy) communication, Near Field Communication (NFC), WLAN communication, Zigbee communication, Infrared Data Association (IrDA) communication, and WFD. (Wi-Fi Direct) communication, UWB (ultra-wideband) communication, Ant+ communication, WIFI communication, RFID (Radio Frequency Identification) communication, 3G communication, 4G communication, and 5G communication, etc., but this is only an example. , but is not limited thereto.

The output unit 740 may output data input to the glaucoma diagnosis apparatus 700, stored data, processed data, and the like. For example, the output unit 740 may output data input through the data input unit 310 and the feedback input unit 710, diagnosis results of the diagnosis unit 740, and graphic charts generated by the diagnosis unit 740. .

According to an embodiment, the output unit 740 may output data input to the glaucoma diagnosis apparatus 700, stored data, processed data, and the like using at least one of an auditory method, a visual method, and a tactile method. there is. To this end, the output unit 740 may include a display, a speaker, a vibrator, and the like.

Meanwhile, although FIG. 7 shows the data input unit 310 and the feedback input unit 710 as separate components, it is not limited thereto, and the data input unit 310 and the feedback input unit 710 may be integrated into one component.

Also, depending on embodiments, the function of the model updater 720 may be performed by the glaucoma diagnostic model generating apparatus 110 .

8 is a diagram illustrating a method for generating a glaucoma diagnosis model according to an embodiment.

The glaucoma diagnostic model generating method of FIG. 8 may be performed by the glaucoma diagnostic model generating apparatus 110 of FIG. 2 .

Referring to FIG. 8 , the apparatus for generating a glaucoma diagnostic model may collect learning data to be used for generating a glaucoma diagnostic model (810). For example, the glaucoma diagnosis model generation device may collect clinical information on a plurality of patients and a corresponding glaucoma diagnosis result as learning data from an external device. Here, the clinical information is information obtained through a visual field test, a retinal nerve fiber layer (RNFL) optical coherence tomography (OCT) test, an intraocular pressure test, etc., and the pattern standard deviation ( pattern standard deviation), retinal nerve fiber layer (RNFL) superior thickness, retinal nerve fiber layer inferior thickness, retinal nerve fiber layer temporal thickness, and intraocular pressure. can

The glaucoma diagnosis model generating apparatus may generate a glaucoma diagnosis model by learning a machine learning model based on the collected learning data (820). At this time, the machine learning model is a tree-based machine learning model (e.g., random forest, XGboost, ADAboost, Light GBM, C5.0, ID3, C4.5, CART, etc.), deep learning, K-nearest neighbors, and Naive Bayes classification , neural networks, support vector machines, and the like. More preferably, the machine learning model may be a tree-based machine learning model, for example, XGboost.

Meanwhile, according to an embodiment, the glaucoma diagnostic model generating apparatus may optimize the hyperparameters of the machine learning model in the glaucoma diagnostic model generating process 820 . For example, the apparatus for generating a glaucoma diagnosis model may optimize hyperparameters of a machine learning model through a Bayesian optimization algorithm.

According to an embodiment, the apparatus for generating a diagnostic model for glaucoma may process a missing value if a missing value exists in the collected clinical information. For example, the apparatus for generating a diagnostic model for glaucoma may process missing values using various methods, such as a method of deleting missing values, a method of replacing missing values, and a method of predicting missing values with a data set composed of variables without missing values.

9 is a diagram illustrating a method for diagnosing glaucoma according to an exemplary embodiment.

The glaucoma diagnosis method of FIG. 9 may be performed by the glaucoma diagnosis apparatus of FIG. 3 .

Referring to FIG. 9 , the apparatus for diagnosing glaucoma may receive clinical information of a subject to be used for glaucoma diagnosis from a user (910). As described above, the clinical information may include pattern standard deviation, upper retinal nerve fiber layer thickness, lower retinal nerve fiber layer thickness, retinal nerve fiber layer temporal thickness, intraocular pressure, and the like.

The glaucoma diagnosis apparatus may diagnose whether or not the subject has glaucoma by using the input clinical information of the subject and the pre-stored glaucoma diagnosis model (820). More specifically, the glaucoma diagnosis apparatus may diagnose glaucoma by inputting clinical information of a test subject into a glaucoma diagnosis model.

The apparatus for diagnosing glaucoma may generate and output a graphic chart through an output means to explain the reason for the diagnosis result (820). In this case, the graphic chart may represent the position of the value of the subject's clinical information in the clinical information distribution of the learning data and/or the role of each clinical information in the diagnosis process of the glaucoma diagnosis model. For example, the graphic chart may include a gauge chart, a radial chart, a Shapley Additive Explanations (SHAP) chart, and the like.

10 is a diagram illustrating a method for diagnosing glaucoma according to an exemplary embodiment.

The glaucoma diagnosis method of FIG. 10 may be performed by the glaucoma diagnosis apparatus of FIG. 7 .

Referring to FIG. 10 , the apparatus for diagnosing glaucoma may receive clinical information of a subject to be used for glaucoma diagnosis from a user (1010).

The glaucoma diagnosis apparatus may diagnose whether or not the subject has glaucoma by using the input clinical information of the subject and the pre-stored glaucoma diagnosis model (1020).

The apparatus for diagnosing glaucoma may generate and output a graphic chart through an output means to explain the reason for the diagnosis result (1030).

The glaucoma diagnosis apparatus may receive the clinician's final diagnosis result for the subject as feedback (1040).

When the diagnosis result in step 1020 and the final diagnosis result input in step 1030 are different, the glaucoma diagnosis apparatus uses a pre-stored glaucoma diagnosis model based on the clinical information of the subject input in step 1010 and the final diagnosis result input in step 1030. Can be updated (10580). For example, the glaucoma diagnosis apparatus may additionally learn a pre-stored glaucoma diagnosis model using clinical information of the subject and a final diagnosis result as new learning data.

An aspect of the present invention may be implemented as computer readable code on a computer readable recording medium. Codes and code segments implementing the above program can be easily inferred by a computer programmer in the art. A computer-readable recording medium may include all types of recording devices storing data that can be read by a computer system. Examples of computer-readable recording media may include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, and the like. In addition, the computer-readable recording medium may be distributed among computer systems connected through a network, and may be written and executed as computer-readable codes in a distributed manner.

So far, the present invention has been looked at mainly with its preferred embodiments. Those skilled in the art to which the present invention pertains will be able to understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the scope of the present invention should be construed to include various embodiments within the scope equivalent to those described in the claims without being limited to the above-described embodiments.

[Example]

Clinical information was collected for a total of 1624 cases from 430 patients with glaucoma and 377 normal subjects. Clinical information was acquired through visual field examination, RNFL OCT examination, intraocular pressure examination, and fundus imaging examination.

The collected clinical information included gender, age, glaucoma hemifield test, visual field index, mean deviation, pattern standard deviation, upper retinal nerve fiber layer thickness, lower retinal nerve fiber layer thickness, retinal nerve fiber layer temporal thickness, retinal nerve fiber layer nasal thickness, retinal nerve fiber layer average thickness, intraocular pressure, corneal thickness, best-corrected visual acuity, spherical equivalent, axial length length), neuro-retinal rim thickness, cup diameter, disc diameter, mean cup/disc ratio, vertical cup/disc ratio , CNN (convolutional neural network) degree, etc. A total of 22 features were included. Here, the CNN degree is the degree of glaucoma predicted from the fundus image by generating a diagnosis model based on CNN and using the generated diagnosis model.

Using the chi-square feature selection technique, 10 beneficial features were selected from a total of 22 features, and a combination test was performed to determine the 5 features that produced the best accuracy (pattern standard deviation, retinal nerve fiber layer). superior thickness, retinal nerve fiber layer inferior thickness, retinal nerve fiber layer temporal thickness and intraocular pressure) were obtained.

Clinical information of a total of 1624 cases was randomly divided, 1306 cases were used as a learning set to generate a glaucoma diagnosis model, and 318 cases were used as a verification set to verify the performance of the glaucoma diagnosis model. Four glaucoma diagnosis models were created and tested using a support vector machine, C5.0, random forest, and XGboost, and the following table was obtained.

Referring to the table above, it can be seen that when a glaucoma diagnosis model is generated using a machine learning model based on the five features, very high performance is shown.

In particular, in the case of the XGboost model, AUC was 0.945, sensitivity was 0.941, accuracy was 0.947, and specificity was 0.950, confirming that the diagnostic performance was better than other models.

100: glaucoma diagnosis system
110: glaucoma diagnostic model generating device
120, 700: glaucoma diagnosis device
210: learning data collection unit
220: model generating unit
310: data input unit
320: storage unit
330: diagnosis unit
710: feedback input unit
720: model update unit
730: communication department
740: output unit

Claims (11)

a data input unit that receives clinical information of the subject; and
a diagnostic unit for diagnosing whether or not the subject has glaucoma from the clinical information using a machine learning-based glaucoma diagnosis model and generating a graphic chart to explain the reason for the diagnosis result; including,
The clinical information includes a pattern standard deviation obtained through an upper retinal nerve fiber layer thickness, a lower retinal nerve fiber layer thickness, a temporal retinal nerve fiber layer thickness, an intraocular pressure, and a visual field test.
Glaucoma diagnostic device. According to claim 1,
The diagnosis unit,
Generating a graphic chart expressing the position of the value of the clinical information of the subject within the distribution of clinical information used in the generation of the glaucoma diagnostic model or expressing the role of each clinical information in the diagnosis process,
Glaucoma diagnostic device. According to claim 2,
The graphic chart includes a gauge chart, a radar chart, and a Shapley Additive Explanations (SHAP) chart.
Glaucoma diagnostic device. According to claim 1,
The glaucoma diagnosis model is generated using XGboost,
Glaucoma diagnostic device. According to claim 1,
a feedback input unit that receives a final diagnosis result for the test subject as feedback; and
a model updater configured to update the glaucoma diagnosis model based on the final diagnosis result and the clinical information when a final diagnosis result received as the feedback is different from a diagnosis result of the diagnosis unit; Including more,
Glaucoma diagnostic device. receiving clinical information of the subject;
diagnosing whether or not the subject has glaucoma from the clinical information using a machine learning-based glaucoma diagnosis model; and
generating a graphic chart to explain the reason for the diagnosis result; including,
The clinical information includes a pattern standard deviation obtained through an upper retinal nerve fiber layer thickness, a lower retinal nerve fiber layer thickness, a temporal retinal nerve fiber layer thickness, an intraocular pressure, and a visual field test.
Methods for diagnosing glaucoma. According to claim 6,
Generating the graphic chart,
Generating a graphic chart expressing the position of the value of the clinical information of the subject within the distribution of clinical information used in the generation of the glaucoma diagnostic model or expressing the role of each clinical information in the diagnosis process,
Methods for diagnosing glaucoma. According to claim 7,
The graphic chart includes a gauge chart, a radar chart, and a Shapley Additive Explanations (SHAP) chart.
Methods for diagnosing glaucoma. According to claim 6,
The glaucoma diagnosis model is generated using XGboost,
Methods for diagnosing glaucoma. According to claim 6,
receiving a final diagnosis result for the subject as feedback; and
updating the glaucoma diagnosis model based on the final diagnosis result and the clinical information when a final diagnosis result received as the feedback and a diagnosis result using the glaucoma diagnosis model are different; Including more,
Methods for diagnosing glaucoma. a glaucoma diagnosis model generating device that collects clinical information about multiple patients and corresponding glaucoma diagnosis results as learning data, and generates a glaucoma diagnosis model by learning a machine learning model based on the collected learning data; and
A glaucoma diagnosis device that receives clinical information of a subject, diagnoses whether or not the subject has glaucoma from the clinical information of the subject using the glaucoma diagnosis model, and generates a graphic chart to explain the reason for the diagnosis result; including,
The clinical information includes a pattern standard deviation obtained through an upper retinal nerve fiber layer thickness, a lower retinal nerve fiber layer thickness, a temporal retinal nerve fiber layer thickness, an intraocular pressure, and a visual field test.
Glaucoma Diagnostic System.

Images