Ophthalmic device, program, and recording medium

Abstract

The glaucoma screening method according to an embodiment includes the following steps. In a calculation model preparation step, a calculation model is prepared, the calculation model including a linear combination of at least one papilla parameter that represents the thickness of a retinal tissue in an optic papilla area and a linear combination of at least one macula parameter that represents the thickness of a retinal tissue in a macula area. In an OCT data generation step, the ocular fundus of an eye to be tested is subjected to OCT scanning to generate OCT data. In a measurement value calculation step, a plurality of measurement values including a papilla measurement value corresponding to each papilla parameter and a macula measurement value corresponding to each macular parameter of the calculation model are calculated on the basis of the OCT data. In a measurement value input step, the calculated plurality of measurement values are input into the calculation model. In a calculation result provision step, a calculation result output from the calculation model in accordance with the inputting of the plurality of measurement values is provided.

Description

[Technical Field]

[0001] This disclosure ,eye Regarding scientific equipment, programs, and recording media. [Background technology]

[0002] Glaucoma is one of the leading causes of blindness. It is estimated that there are over 60 million people with glaucoma worldwide, and this number is projected to increase further. Several epidemiological studies have revealed that many glaucoma patients are undiagnosed and unaware that they have the disease.

[0003] To prevent limitations in daily life caused by visual field defects or blindness due to glaucoma, it is crucial to detect the disease early and begin treatment promptly. Providing widespread screening using ophthalmological examinations is considered an effective way to promote early detection. However, mass screening for glaucoma is not common in many countries today. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 5-184542 [Patent Document 2] U.S. Patent Application Publication No. 2018 / 0025112 [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] To implement glaucoma mass screening, several problems need to be solved. Two representative problems are described below.

[0006] The first problem is the reliability of glaucoma identification. For example, conventional glaucoma identification methods using image analysis often misdiagnosed retinal thinning in normal eyes due to the influence of artifacts and vascular images. In other words, conventional methods had the problem of a high rate of false positives. A specialist can easily distinguish false positives caused by artifacts, etc. However, since mass screening is usually conducted in an environment without specialists, there is a possibility that normal eyes may be classified as eyes suspected of having glaucoma. Such subjects will then undergo detailed examinations at specialized hospitals, but the results will naturally be negative. Such occurrences not only reduce the reliability of the mass screening facilities and screening methods, but also reduce the quality of statistical data created from the data collected through mass screening. Furthermore, it creates unnecessary psychological, laborious, financial, and time burdens for individual subjects. Therefore, one condition for successful glaucoma mass screening is the improvement of disease identification quality (accuracy, precision, etc.).

[0007] The second problem is the difficulty and complexity of identifying normal-tension glaucoma. Normal-tension glaucoma is a type of glaucoma in which the optic nerve is damaged despite normal intraocular pressure, unlike glaucoma in which elevated intraocular pressure damages the optic nerve. Conventional glaucoma screening methods often started with screening using intraocular pressure as an indicator (biomarker), making it impossible to identify normal-tension glaucoma, which does not exhibit elevated intraocular pressure. Identifying normal-tension glaucoma required a physician's judgment after performing fundus examination and visual field testing in addition to intraocular pressure testing. It is clear that such methods are unsuitable for mass screening.

[0008] One objective of this disclosure is to provide a glaucoma identification technology suitable for mass screening. [Means for solving the problem]

[0009] One exemplary aspect of the present disclosure is a method for glaucoma screening, comprising: preparing a computational model including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area; applying an optical coherence tomography scan to the fundus of the eye under examination to generate OCT data; calculating a plurality of measurements based on the OCT data, including one or more optic disc measurements corresponding to each of the one or more optic disc parameters and one or more macular measurements corresponding to each of the one or more macular parameters; inputting the plurality of measurements into the computational model; and providing a calculation result output from the computational model in response to the input of the plurality of measurements.

[0010] Another exemplary aspect of the present disclosure is an ophthalmic device usable for glaucoma screening, comprising: a storage unit for storing a pre-prepared calculation model including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area; an OCT data acquisition unit for acquiring optical coherence tomography (OCT) data of the fundus of an eye under examination; a measurement value calculation unit for calculating a plurality of measurements based on the OCT data, including one or more optic disc measurements corresponding to the one or more optic disc parameters and one or more macular measurements corresponding to the one or more macular parameters; and a risk information generation unit for generating glaucoma risk information of the eye under examination based on the calculation model and the plurality of measurements.

[0011] A further exemplary embodiment of the present disclosure is a program capable of causing a computer including a processor and memory to perform glaucoma screening, wherein the program causes the processor to store in the memory a computational model including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area; to receive optical coherence tomography (OCT) data of the fundus of the eye under examination; to calculate a plurality of measurements based on the OCT data, including one or more optic disc measurements corresponding to the one or more optic disc parameters and one or more macular measurements corresponding to the one or more macular parameters; to input the plurality of measurements into the computational model; and to provide a computational result output from the computational model in response to the input of the plurality of measurements.

[0012] A further exemplary aspect of the present disclosure is a computer-readable non-temporary recording medium on which a program capable of causing a computer including a processor and memory to perform glaucoma screening is recorded, the program causing the processor to store in the memory a computational model including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area; a control for receiving optical coherence tomography (OCT) data of the fundus of an eye under examination; a process for calculating a plurality of measurements based on the OCT data, including one or more optic disc measurements corresponding to each of the one or more optic disc parameters and one or more macular measurements corresponding to each of the one or more macular parameters; a process for inputting the plurality of measurements into the computational model; and a process for providing a computational result output from the computational model in response to the input of the plurality of measurements. [Effects of the Invention]

[0013] According to exemplary embodiments of this disclosure, a glaucoma identification technology suitable for mass screening can be provided. [Brief explanation of the drawing]

[0014] [Figure 1] 10 It is a flowchart showing the procedure of the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 2] 10 It is a flowchart showing the procedure of the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 3] 10 It is a flowchart for explaining the creation of a calculation model and the verification of its quality in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 4A] 10 It is a diagram for explaining OCT measurement prediction factors in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 4B] 10 It is a diagram for explaining OCT measurement prediction factors in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 4C] 10 It is a diagram for explaining OCT measurement prediction factors in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 5] 10 It is a table showing the characteristics of cases and controls in the development data used for the development of the calculation model in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 6] 10 It is a histogram showing the distribution of cases and the distribution of controls for OCT measurement prediction factors in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 7] 10 It is a table showing three calculation models in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 8] 10 It is a receiver operating characteristic curve of three calculation models in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 9] 10 It is a table for explaining the verification results of three calculation models in the glaucoma screening method according to an exemplary aspect of the embodiment. [Figure 10]This table illustrates the validation results of three computational models in a glaucoma screening method according to one exemplary embodiment of the model. [Figure 11] This table illustrates the validation results of three computational models in a glaucoma screening method according to one exemplary embodiment of the model. [Figure 12] This is a block diagram showing the configuration of an ophthalmic device according to one exemplary embodiment of the model. [Modes for carrying out the invention]

[0015] Several non-limiting embodiments relating to this disclosure will be described in detail with reference to the drawings.

[0016] Any known technology can be incorporated into the embodiments relating to this disclosure. For example, any subject matter freely selected from any of the documents referenced herein can be incorporated into the embodiments relating to this disclosure. Furthermore, any technology known in any art related to this disclosure can be incorporated into the embodiments relating to this disclosure. The subjects incorporated into the embodiments relating to this disclosure are not limited to these. For example, any technical subject matter disclosed by the applicant of this application (e.g., any technical subject matter disclosed by any means such as a patent application, a paper, or a website) can be incorporated into this disclosure by reference.

[0017] Two or more exemplary embodiments relating to this disclosure can be combined, at least partially.

[0018] Any of the various functions of the various elements described in the embodiments relating to this disclosure are implemented using a circuitry or processing circuitry. The circuitry or processing circuitry is a general-purpose processor, dedicated processor, integrated circuit, CPU (Central Processing Unit), GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), programmable logic device (e.g., SPLD (Simple Programmable Logic Device), CPLD (Complex Programmable Logic Device), FPGA (Field Programmable Gate) configured and / or programmed to perform a certain function. This includes any of the following: an array, a conventional circuit configuration, and any combination thereof. A processor is considered a processing circuit configuration or circuit configuration, including transistors and / or other circuit configurations. In this disclosure, a circuit configuration, unit, means, or similar terms means hardware that performs a certain function, or hardware programmed to perform a certain function. Hardware may be any of the various hardware described in the embodiments relating to this disclosure, or it may be hardware programmed and / or configured to perform a certain function. When a processor that can be considered a certain type of circuit configuration is employed as hardware, a circuit configuration, unit, means, or similar terms means represent an element that includes a combination of hardware and software, the software being used to constitute the hardware and / or processor.

[0019] <Overview of Embodiments> Several exemplary embodiments of this disclosure aim to propose glaucoma identification technologies suitable for mass screening. To this end, these exemplary embodiments of glaucoma screening technologies focus on the two aforementioned problems, namely, the reliability problem in glaucoma identification (the first problem) and the difficulty and complexity in identifying normal-tension glaucoma (the second problem), and aim to solve or mitigate these problems. Those skilled in the art will understand from the embodiments described later that several exemplary embodiments of this disclosure solve or mitigate the first and second problems.

[0020] The embodiments relating to this disclosure are not limited to those made with regard to both of these two problems, but may, for example, be made with regard to only one of these two problems, or with regard to one or both of these two problems plus another problem, or with regard to a problem other than these two problems. In other words, the embodiments relating to this disclosure may solve or mitigate only one of these two problems, or with regard to one or both of these two problems plus another problem, or with regard to a problem other than these two problems. Hereinafter, unless otherwise specified, "solving a problem" and "mitigating a problem" are synonymous.

[0021] Optical coherence tomography (OCT) is a non-contact, high-speed, and non-invasive imaging modality. Although OCT is a relatively new technology, its adoption in the field of ophthalmology has been remarkable, and it is now used in many medical institutions, including small clinics. Therefore, OCT can be said to be a suitable imaging modality for use in mass screening.

[0022] In the embodiments of this disclosure, OCT is used to generate cross-sectional images of the fundus. More specifically, glaucoma identification in the embodiments of this disclosure is essentially performed based solely on images obtained using OCT (referred to as OCT images). In contrast, various known glaucoma identification techniques essentially use not only images obtained using OCT (referred to as OCT images), but also background information of the subject, intraocular pressure data, visual field test data, and images obtained using ophthalmic modalities other than OCT (such as slit-lamp microscopes, fundus cameras, and scanning laser ophthalmoscopy). Even if a glaucoma identification technique using only OCT images exists, it is considered to differ from glaucoma identification in the embodiments of this disclosure in that it is not possible to solve the first and / or second problems, and that it does not have the configuration to solve these problems.

[0023] The type of OCT used in the embodiments of this disclosure may be arbitrary, but is typically spectral domain OCT (SD-OCT) or swept-source OCT (SS-OCT). In the embodiments described below, spectral domain OCT is used, but those skilled in the art will understand that similar configurations and effects can be obtained using other types of OCT.

[0024] In the embodiments described herein, an algorithm suitable for glaucoma mass screening using OCT is created and validated. To this end, in the embodiments described later, an algorithm (computational model) is created that functions to generate a glaucoma risk score from OCT images using important indicators selected from various indicators obtained from OCT data of the fundus.

[0025] Several embodiments of glaucoma screening methods that can be implemented using this algorithm are described below. These embodiments are illustrative and are not intended to limit the embodiments of this disclosure.

[0026] Figure 1 shows one embodiment of a glaucoma screening method (glaucoma identification method). Details of each step, including non-specific examples, will be described later.

[0027] In the method according to this embodiment, first, a calculation model for determining a glaucoma risk score from OCT images is created (step S11).

[0028] Step S11 is performed by a computer operating according to a computational model creation program. This program may include a program that utilizes machine learning techniques. For example, development data (training data, teacher data, training data) containing a large number of OCT data (or a large number of measurements calculated from a large number of OCT data) labeled with labels (glaucoma risk score values) created by a specialist or another computational model, and a mathematical model such as a neural network are prepared, and machine learning using this development data is applied to this mathematical model to construct a computational model. Several examples of how to create a computational model will be described later.

[0029] The calculation model created in step S11 includes parameters related to the optic nerve head (optic nerve head parameters) and parameters related to the macula (macular parameters). In other words, this calculation model is a mathematical model configured to receive measured values ​​corresponding to the optic nerve head parameters and measured values ​​corresponding to the macular parameters as inputs, and to output calculation results based on these measured values.

[0030] Optic disc parameters are arbitrary parameters measured or calculated in relation to the optic nerve head, such as parameters representing the retinal tissue thickness in a defined area (optic nerve head area) for the optic nerve head. The optic nerve head area may be, for example, at least a portion of the optic nerve head, the area surrounding the optic nerve head, or a combination of both. Retinal tissue thickness represents the dimensions (thickness) of one or more of the tissues that make up the retina. Histologically, the retina is divided into the following 10 layers: retinal pigment epithelium, photoreceptor layer, outer limiting membrane, outer granular layer, outer plexiform layer, inner granular layer, inner plexiform layer, ganglion cell layer, nerve fiber layer, and internal limiting membrane. Retinal tissue thickness may be the thickness of one or more of these layers. Optic disc parameters are not limited to these examples and may be any parameters that can be derived based on data collected by OCT scan from a fundus area defined based on the optic nerve head, such as cup dimensions (diameter, area, etc.), disk dimensions, rim dimensions, ratio of two dimensions, or optic disc tilt.

[0031] Macular parameters are arbitrary parameters measured or calculated in relation to the macula, such as parameters representing the retinal tissue thickness in a defined area of ​​the macula (macular area). The macular area may be, for example, at least a portion of the macula, the area surrounding the macula, or a combination of both. Macular parameters are not limited to these examples and may be any parameters that can be derived based on data collected by OCT scan from a fundus area defined based on the macula, such as macular dimensions (diameter, area, depth, etc.). The retinal tissue corresponding to the macular parameter may be the same as or different from the retinal tissue corresponding to the optic disc parameter.

[0032] The calculation model in this embodiment is expressed as a predetermined mathematical formula that includes one or more papillary parameters and one or more macular parameters as variables. The form of this formula may be arbitrary.

[0033] Some exemplary computational models include a linear combination of one or more nipple parameters and / or a linear combination of one or more macular parameters. In other words, some exemplary computational models include a linear equation with one or more nipple parameters and / or one or more macular parameters as variables.

[0034] As is well known, there are N parameters P1, P2, ..., P N The linear combination of the coefficients c1, c2, ..., c N Using this, it can be expressed as follows: c1P1+c2P2+···+c N P N Also, N variables x1, x2, ..., x N The linear equation y=f(x) containing a has coefficients a0, a1, a2, ..., a N Using this, it can be expressed as follows: y = a0 + a1x1 + a2x2 + ... + a N x N .

[0035] The mathematical formulas representing some exemplary computational models may include terms of forms other than zeroth and first-degree terms. For example, the mathematical formulas of some exemplary computational models may include polynomials of degree two or higher. Furthermore, the mathematical formulas of some exemplary computational models may include mathematical symbols other than those used in basic arithmetic operations (e.g., square roots, differentials, integrals, etc.).

[0036] Three examples of computational models (Y(Model 1), Y(Model 2), Y(Model 3)) defined as multivariable linear equations with both nipple and macular parameters as variables are shown below. Each variable (e.g., TSNITlower) will be described later.

[0037] [First computational model] Y (Model 1) = 14.4305 - 0.0404 × TSNITlower - 0.0303 × cpRNFLqS - 0.0304 × cpRNFLqIt - 0.0271 × cpRNFLqIn - 0.1424 × mGCIPL_IT + 1.1427 × log_mGCIPL_STvsIT + 0.6971 × log_mGCIPL_ITvsIN + 0.0602 × mGCC_IN

[0038] [Second computational model] Y (Model 2) = 12.6694 - 0.0329 × TSNITlower - 0.0344 × cpRNFLqS - 0.0318 × cpRNFLqIt - 0.0398 × cpRNFLqIn + 1.1646 × log_mGCIPL_STxvsITx + 0.712 × log_mGCC_ITxvsINx

[0039] [Third computational model] Y (Model 3) = 12.5935 - 0.066 × TSNITlower - 0.0296 × cpRNFLqIn - 0.0289 × mGCC + 0.9845 × log_mGCIPL_SNyvsINy + 0.5671 × log_mGCIPL_STyvsITy + 0.6613 × log_mGCC_ITyvsINy

[0040] In this embodiment, step S11 (creation of a glaucoma risk score calculation model) may be performed at any point prior to the evaluation of the eye using this calculation model (steps S12 to S15). The created calculation model is stored on the computer that performs the evaluation of the eye, and / or on a storage device accessible by this computer. This storage device may be, for example, a node in a computer network to which this computer belongs, such as a local area network (LAN), the Internet, or a wide area network (WAN). Alternatively, this storage device may be a peripheral device directly or indirectly connected to this computer.

[0041] Furthermore, as will be explained later with reference to Figure 2, the calculation model can also be updated using data used to evaluate the eye under examination (corresponding to the OCT data generated in step S12 of this embodiment).

[0042] The example calculation model described above is defined as a mathematical formula with only parameters derived from OCT data as variables. However, several other forms of calculation models may include other parameters as variables, for example, to improve the quality of the calculation (accuracy, precision, reproducibility, etc.). Examples of such additional or auxiliary parameters include: (1) parameters obtained from OCT data other than optic disc and macular parameters; (2) parameters relating to specific sub-tissues of the retina; (3) parameters relating to tissues other than the retina (e.g., parameters relating to the choroid, sclera, vitreous humor, iridocorneal angle, etc.); (4) parameters relating to fundus blood vessels (e.g., vascular density, etc.); (5) parameters relating to the dimensions and / or shape of the area where the lesion or impairment occurs; (6) parameters relating to fundus blood flow; (7) parameters relating to the subject's background information (e.g., age, sex, race, medical history, treatment history, medication history, family history, etc.); (8) parameters derived from data obtained from ophthalmic modalities other than OCT; (9) parameters relating to data obtained from ophthalmic examinations (e.g., intraocular pressure, visual field test parameters, etc.); (10) parameters relating to data obtained from examinations in medical departments other than ophthalmology (e.g., blood pressure, blood test parameters, genetic test parameters, etc.).

[0043] In step S12, an OCT scan is applied to the fundus of the eye being examined to generate OCT data. The OCT scan is performed, for example, by a spectral domain type OCT device (OCT scanner).

[0044] The type (type, format, etc.) of OCT data generated in step S12 is typically determined by the processing content in step S13. For example, this OCT data may include any of the following: data collected by OCT scanning (collected data), image data obtained by applying imaging processing such as Fourier transform to the collected data (OCT image data), data obtained during the imaging processing (pre-imaging data), and data obtained by applying a predetermined process to the OCT image data (processed OCT image data).

[0045] When the collected data is processed to generate OCT data, this processing is performed by a computer operating according to an OCT data generation program. The OCT data generation program may be part of a program that causes the computer to perform the series of processes in steps S12 to S15.

[0046] The number of OCT scans applied to the fundus of the eye under examination in step S12 may be one or more. In some cases, multiple OCT scans can be applied to the same region of the fundus of the eye under examination. In this case, based on multiple acquired data from the same region, it is possible to generate average images with reduced random noise (such as additive average images), OCT angiography images (OCTA images) depicting fundus blood vessels, and OCT hemodynamic data showing fundus hemodynamics.

[0047] In several other examples, OCT scans can be applied to two or more different regions of the fundus of the eye being examined, generating separate OCT data for these regions. For example, an OCT scan of the region containing the optic nerve head area (optic nerve head area scan) and an OCT scan of the region containing the macula (macular area scan) can be performed separately. In this case, OCT data for the optic nerve head area can be generated based on the optic nerve head area scan, and OCT data for the macular area can be generated based on the macular area scan.

[0048] Alternatively, a single OCT scan (wide-area scan) may be applied to a region that includes both the optic nerve head area and the macular area. In this case, both OCT data for the optic nerve head area and OCT data for the macular area can be generated based on the data collected by the wide-area scan.

[0049] In the next step, S13, data to be input into the calculation model created in step S11 is obtained based on the OCT data generated in step S12. As mentioned above, the calculation model in this embodiment includes one or more nipple parameters and one or more macular parameters as variables. Therefore, in step S13, nipple measurements corresponding to each nipple parameter and macular measurements corresponding to each macular parameter are calculated. In addition, other measurements may be calculated additionally or as an auxiliary measure.

[0050] Step S13 is performed by a computer operating according to a measurement value calculation program. The measurement value calculation program may be part of a program that causes the computer to perform the series of processes in steps S12 to S15.

[0051] In configurations where information other than the measurements calculated from OCT data (e.g., the additional or auxiliary parameters mentioned above) is also used, the computer performs a process to acquire that information. For example, the computer may perform a process to acquire information about the subject by accessing a medical information database (e.g., an electronic medical record system, a medical image archiving system, a medical information sharing system, etc.).

[0052] In the next step, S14, the nipple measurement values ​​and macular measurement values ​​calculated in step S13 are input into the calculation model created in step S11. Each nipple measurement value is substituted into the corresponding nipple parameter (variable), and each macular measurement value is substituted into the corresponding macular parameter (variable).

[0053] In some exemplary embodiments, additional information is input to the calculation model in addition to the measurement values ​​calculated in step S13. Also in some exemplary embodiments, the results output from the calculation model based on the input of the measurement values ​​calculated in step S13 are adjusted (corrected) based on the additional information.

[0054] Step S14 is performed by a computer operating according to a measurement input program. The measurement input program may be part of a program that causes the computer to perform the series of processes in steps S12 to S15.

[0055] The computer obtains a calculation result (glaucoma risk score) based on the calculation model created in step S11 and the measurements (and other information) entered into the calculation model in step S14. In step S15, the obtained glaucoma risk score is provided.

[0056] The recipients of the glaucoma risk score are predetermined or determined for each processing step. Examples of recipients include users, other computers, medical systems, ophthalmic devices, and storage devices. In some exemplary embodiments, the glaucoma risk score may be displayed on a display device, transmitted to a physician's terminal, a patient's terminal, a server, a database, a testing device, a treatment device, or stored in a storage device.

[0057] Step S15 is performed by a computer operating according to an information-providing program. The information-providing program may be part of a program that causes the computer to perform the series of processes in steps S12 to S15.

[0058] The embodiment shown in Figure 1 is configured to determine a glaucoma risk score from optic disc and macular measurements calculated from OCT data of the fundus of the eye being examined, and does not require the intraocular pressure of the eye being examined for glaucoma assessment. Therefore, this embodiment makes it possible to identify normal-tension glaucoma. Furthermore, as will be described in detail later, this embodiment makes it possible to identify glaucoma with high reliability. Therefore, this embodiment provides a glaucoma identification technology suitable for mass screening.

[0059] As mentioned above, some exemplary embodiments may be configured to update the computational model using data used to evaluate the eye under examination. One embodiment of such a glaucoma screening method (glaucoma identification method) will be described with reference to Figure 2.

[0060] Steps S21 (creation of calculation model), S22 (generation of OCT data), S23 (calculation of nipple and macular measurements), and S24 (input of measurements into calculation model) may be performed in the same manner as steps S11, S12, S13, and S14 in Figure 1, respectively. However, a different procedure may be applied.

[0061] In the next step, S25, the computer obtains a glaucoma risk score based on the calculation model created in step S21 and the measurements entered into the calculation model in step S24, and adjusts the coefficients of the calculation model based on this glaucoma risk score.

[0062] Typically, the computational model in this embodiment is defined as a multivariable linear equation with nipple parameters and macular parameters as variables. Each term in this multivariable linear equation is the product of a coefficient and a variable. In this embodiment, for example, the development data used in creating the computational model in step S21 is updated by adding the nipple and macular measurements calculated in step S23. By applying machine learning using the updated development data to the computational model, the coefficients of the computational model can be adjusted.

[0063] Step S25 is performed by a computer operating according to a calculation model update program. The calculation model update program may be part of a program that causes the computer to perform the series of processes in steps S22-S25 (or steps S21-S25). The calculation model update program may also be the same as the calculation model creation program used in step S21.

[0064] In the next step, S26, the nipple measurement values ​​and macular measurement values ​​calculated in step S23 are input into the calculation model updated in step S25.

[0065] The computer derives a glaucoma risk score based on the new calculation model obtained in step S25 and the measured values ​​(and other information) input into the new calculation model in step S26. In step S27, the computer provides the glaucoma risk score thus obtained. The provision of the glaucoma risk score is performed, for example, in the same manner as in step S15 in Figure 1. However, a different method may be applied.

[0066] The embodiment shown in Figure 2, like the embodiment in Figure 1, can identify normal-tension glaucoma and can identify glaucoma with high reliability, thus providing a glaucoma identification technology suitable for mass screening. In addition, this embodiment can use OCT data of the fundus of the eye being evaluated for glaucoma to update the computational model, making it possible to continuously improve the quality of the computational model.

[0067] Since this embodiment uses subject data to update a computational model, it is desirable to perform processing to protect the subject's personal information. In some exemplary embodiments, items that constitute personal information (personal information items) are defined in advance. Examples of personal information items include subject identifiers (subject ID), name, address, public personal identification number, and background information. When processing information for a specific subject, the computer removes information corresponding to each personal information item from all information about that subject, and uses only the remaining information (including OCT data and / or measurements) to update the computational model. This improves the level of personal information protection. Such processing is performed by a computer operating in accordance with a personal information protection program. The personal information protection program may be part of the computational model update program described above.

[0068] <Example of a particular configuration> The outline of the embodiments has been described above. Several exemplary aspects of such embodiments are described below.

[0069] First, the creation of a computational model used in a glaucoma screening method according to one exemplary embodiment will be described along with an example carried out by the inventors. The flowchart in Figure 3 shows an overview of the procedure for creating a computational model used in the glaucoma screening method according to this embodiment.

[0070] The first step, S31, is the creation of development data. The development data used in this example included data from 11,487 eyes of 7,572 participants. The average age of the participants was 51.3 years (±10.0 years), and the age range was 35 to 74 years. Each participant underwent a medical history review, fundus photography using a digital fundus camera, visual field testing using an FDT (Frequency Doubling Technology) perimeter, automated OCT measurement using a spectral domain OCT, and axial length measurement using an optical axial length measuring device. In other words, each participant's data includes medical history data, fundus photographs, visual field data, fundus OCT data, and axial length data. Note that, considering the detection of normal-tension glaucoma, the development data in this example does not include intraocular pressure values.

[0071] The next step, S32, is the selection of eligible cases and eligible controls. In the development of the glaucoma risk score in this study, a matched case-control design was employed. Each case was a participant who was diagnosed with glaucoma by an ophthalmologist based on visual field testing and fundus examination. Each control was a participant who had no history of glaucoma and showed no signs of glaucoma on fundus photography or FDT visual field testing.

[0072] The next step, S33, is background matching between cases and controls. In this example, one control was randomly selected for each case by matching for sex (male or female) and age (5-year age group). The final development dataset used for analysis included 191 cases (284 eyes) and 277 controls (287 eyes).

[0073] The next step, S34, is the development of a computational model using the development data obtained in step S33. Details of the computational model development methods (statistical analysis methods) used in this example will be described later, with some examples provided.

[0074] The next step, S35, is the creation of a validation dataset. The validation dataset is different from the development dataset. In this example, the development dataset was obtained from participants who underwent health checkups in 2016, while the validation dataset was obtained from participants who underwent health checkups in 2018. The validation dataset used in this example consisted of OCT data from 9720 eyes. Note that intraocular pressure values ​​were not included in the validation dataset in this example.

[0075] The next step, S36, is to validate the calculation model for the positive predictive value. The positive predictive value is defined as the probability that a positive result is actually positive, and in some embodiments, it is defined as the probability that a glaucoma risk score is actually positive when it is a "high" score as described later.

[0076] In this example, the calculation model obtained in step S34 was applied to each validation dataset (OCT data from each of the 9720 eyes). Furthermore, for validation purposes, the score range was divided into three levels: "low," "medium," and "high," and 723 eyes were randomly selected from each score level. The inventors then independently evaluated the OCT reports of each selected eye to determine which of the four options each eye fell into: "normal," "glaucoma," "eye disease other than glaucoma," and "undetermined (requires further detailed ophthalmological examination)."

[0077] The verification of this step was performed using the following five OCT parameters to determine glaucomatous changes: (1) the mean thickness of the retinal nerve fiber layer (RNFL) on a given circle (a circular region set around the optic disc); (2) the presence or absence of localized thinning of the RNFL and the difference in height between double humps in the TSNIT plot; (3) quadrant RNFL thickness chart and clock hour RNFL thickness chart; (4) retinal thickness map, ganglion cell layer++ thickness map (GCL++ thickness map), and GCL+ thickness map in the macular area, as well as the RNFL thickness map in the optic disc area; (5) GCL++ thickness deviation map, GCL+ thickness deviation map, and RNFL thickness deviation map. Here, GCL+ refers to the composite layer of the ganglion cell layer and the internal plexiform layer, and is also called GCIPL. Furthermore, GCL++ represents a complex layer consisting of the retinal nerve fiber layer, ganglion cell layer, and internal plexiform layer, and is also called the ganglion cell complex (GCC).

[0078] A TSNIT plot is data representing the distribution of RNFL thickness in a predetermined circular region around the optic nerve head. More specifically, a TSNIT plot is RNFL thickness distribution data represented using a two-dimensional coordinate system with a first coordinate axis representing the position (angle) in this circular region and a second coordinate axis representing the RNFL thickness. The reference position (angle zero) of the first coordinate axis of the TSNIT plot is temporal (T), and the positive direction of the first coordinate axis is a rotational direction that passes through temporal (T), upward (S), nasal (N), and downward (I) in that order and returns to temporal (T).

[0079] Glaucomatous RNFL defects often present as arcuate defects with temporal borders. OCT reports where inventors did not agree on the evaluation were repeatedly evaluated until all evaluations were consistent. All participants (ophthalmologists) who evaluated the validation data were blinded to the glaucoma score and performed the evaluation in an environment where they could not access information other than the OCT report.

[0080] The next step, S37, is to validate the computational models for sensitivity and specificity. Here, "sensitivity" is defined as the probability of correctly identifying cases that should be judged as positive as positive, and "specificity" is defined as the probability of correctly identifying cases that should be judged as negative as negative.

[0081] To validate this step, a follow-up study was conducted on participants in the development data creation in step S31 who were diagnosed with suspected glaucoma based on fundus images (fundus photographs). In this follow-up study, additional examinations were performed on 129 eyes of 66 participants using a Humphrey visual field analyzer (HFA®) with a standard program based on the 24-2 Swedish Interactive Threshold Algorithm. The inventors evaluated the results of the Humphrey visual field tests and verified the consistency of the evaluation results of the Humphrey visual field tests by applying a computational model to the dataset of 129 eyes in the follow-up study.

[0082] The verified calculation model for this example is obtained by following the steps (steps S31 to S37) of the flowchart in Figure 3, which is outlined above.

[0083] Next, we will explain the OCT measurement predictors used to create the computational model in this example. OCT measurement predictors are predictors generated from data obtained from OCT scans.

[0084] In this example, inner retinal layer thickness was determined using OCT scans based on a 3D wide scan protocol. The 3D wide scan protocol is a protocol for applying OCT scans to a wide 3D area that includes both the retinal area and the optic nerve head area. The inner retina refers to the area from the internal limiting membrane to the external limiting membrane. From the 3D dataset obtained from the OCT scan using the 3D wide scan protocol, layer thickness distribution data in the retinal area and layer thickness distribution data in the optic nerve head area are derived.

[0085] In another example, an OCT scan may be performed separately on a region including the retina and a region including the optic nerve head. The former may generate layer thickness distribution data for the retina, and the latter may generate layer thickness distribution data for the optic nerve head. The OCT scan for the region including the retina may be an OCT scan of the entire region, or a combination of separate OCT scans of two or more subregions of that region.

[0086] The inventors redefined the segmentation of the optic nerve head area and the grid of the macular area, and created variables as promising predictors. The promising predictors were selected from 312 variables defined as shown in Figures 4A, 4B, and 4C.

[0087] In the first step, thickness data for the optic nerve head area (perioptic nerve head RNFL (cpRNFL) thickness data) and several types of thickness data for the macular area are obtained.

[0088] Chart 41A, the left of the two diagrams shown in Figure 4A, shows 12 segments set in the optic disc area. The optic disc area is a circular area (circumferential area) centered on the optic disc center and having a predetermined radius. The optic disc area is divided into 12 segments (arc-shaped regions) having an equal central angle of 30 degrees. The dimensions (diameter) of the optic disc area may be, for example, a default value, or a value set or selected according to the dimensions (diameter) of the optic disc.

[0089] In this example, the average thickness of the RNFL was calculated for each of the 12 segments of the optic nerve head area based on a 3D dataset collected by OCT scans using a 3D wide scan protocol.

[0090] Those skilled in the art will understand that the names of the 12 segments of the optic nerve head area shown in Chart 41A are defined by superior (S), inferior (I), temporal (T), and nasal (N), as well as combinations thereof (the same applies hereafter).

[0091] Chart 42A, the leftmost of the three diagrams shown in Figure 4B, shows a grid that divides the macular area into 10 x 10 = 100 segments. The macular area is a rectangular area centered on the fovea and having predetermined dimensions, and is divided into 100 segments (rectangular regions) of equal dimensions. The dimensions (diameter) of the macular area may be, for example, a default value, or a value set or selected according to the eye being examined.

[0092] In this example, for each of the 100 segments of the macular area, the average thickness of the macular RNFL (mRNFL), the average thickness of the macular ganglion cell layer-internal plexiform layer (mGCIPL), and the average thickness of the macular ganglion cell complex (mGCC) were calculated based on a 3D dataset obtained from OCT scans using a 3D wide-scan protocol.

[0093] In the second step, variables related to the optic nerve head area are determined. In this example, based on the 12 segments of the optic nerve head area (cpRNFL), five regions (segments) qT, qS, qN, qIn, and qIt were defined as shown in chart 41B on the right side of Figure 4A. Furthermore, five variables corresponding to these five regions qT, qS, qN, qIn, and qIt were created: cpRNFLqT, cpRNFLqS, cpRNFLqN, cpRNFLqIn, and cpRNFLqIt.

[0094] Here, cpRNFLqT is the temporal quadrant, cpRNFLqS is the superior quadrant, and cpRNFLqN is the nasal quadrant. In contrast, the inferior quadrant is divided into two regions, qIn and qIt, based on the results of the analysis described later. This division of the inferior quadrant was adopted because the inferior temporal RNFL thickness is considered to be highly useful in detecting glaucoma.

[0095] In the third step, variables related to the macular area are determined. In this example, the segmentation (division into 10x10 segments) of the three layer thickness distributions described above (mRNFL thickness distribution, mGCIPL thickness distribution, and mGCC thickness distribution) is first reconstructed into four regions ST, SN, IT, and IN, as shown in Chart 42A on the left side of Figure 4B, and the variables for the macular area are defined based on these four regions. The variables in this example are used, for example, in the first example Y (Model 1) of the calculation model described above.

[0096] Furthermore, variables are defined by excluding the peripheral portion of the macular area from the 100 segments (10x10 segments) of the macular area. In this example, 64 segments belonging to a range of two segments from the outer edge (periphery) of the macular area are excluded from the 100 segments of the macular area, and the remaining 36 segments are focused on. Then, using these 36 segments, variables corresponding to the four regions STx, SNx, ITx, and INx shown in Chart 42B in the center of Figure 4B are defined. The variables in this example are used, for example, in the second example Y (Model 2) of the calculation model described above.

[0097] Furthermore, variables are defined by excluding both the peripheral and central parts of the macular area from the 100 segments of the macular area. In this example, 64 segments belonging to a range of two segments from the outer edge of the macular area and 4 segments belonging to a range of one segment from the center of the macular area are excluded, and the remaining 32 segments are focused on. Then, using these 32 segments, variables corresponding to the four regions STy, Sny, Ity, and Iny shown in Chart 42C on the right side of Figure 4B are defined. The variables in this example are used, for example, in the third example Y (Model 3) of the calculation model described above.

[0098] Each variable defined for the macular area is the average value of the thickness (in micrometers) of the corresponding segment. For example, the RNFL thickness in segment qS shown in chart 41B of Figure 4A (cpRNFLqS) is defined as the average value of the RNFL thickness in the three segments S, ST, and SN in chart 41A that correspond to segment qS: cpRNFLqS = (cpRNFL_S + cpRNFL_ST + cpRNFL_SN) / 3 (micrometers). Similarly, the GCC thickness in segment SN shown in chart 42A of Figure 4B (mGCC_SN) is defined as the average value of the GCC thickness in the 25 segments belonging to segment SN out of 100 segments separated by a grid: mGCC_SN = (mGCC 01_01 + 01_02 + ... + 05_04 + 05_05) / 25 (micrometers).

[0099] In the fourth step, variables are created that represent the difference in layer thickness between two regions. In this example, variables are created that represent the difference between the region located above and the region located below, and the difference between the region located on the ear side and the region located on the nose side. These variables do not follow a normal distribution. Also, in order to avoid the value calculated when the layer thicknesses of the two regions are equal from becoming zero, these variables are defined as the natural logarithm transformation applied to the real number obtained by adding 1 to the absolute value of the difference between the layer thickness values of the two regions. For example, the variable for the difference in mGCIPL thickness between the two regions ST and IT in FIG. 4B is defined as follows: log_mGCIPL_STvsIT = log e (|mGCIPL_ST - mGCIPL_IT| + 1).

[0100] In the final fifth step, considering the case where the difference in cpRNFL thickness between the region located above and the region located below is large, but the thickness value on the thinner side is not significant, a variable TSNITlower corresponding to the layer thickness corresponding to the lower peak of the double hump presented by the TSNIT plot is defined. An example of this variable TSNITlower is shown in FIG. 4C. Generally, the double hump is a spike near the upper (S) and a spike near the lower (I), and the variable TSNITlower is defined as the smaller of the two layer thickness values at the peak positions of these two spikes.

[0101] Next, the statistical analysis used to create the calculation model of this example will be described.

[0102] In this example, all variables obtained from the OCT measurement values are treated as continuous variables. In the statistical analysis of this example, backward elimination using a multivariate logistic regression model with a variable increment-decrement method with a P-value of 0.1 was performed to select the optimal prediction model in the development dataset. As a result, in this example, the three calculation models Y (Model 1), Y (Model 2), and Y (Model 3) described above were obtained.

[0103] The first calculation model Y (Model 1) includes all of the following as promising predictors: values ​​related to cpRNFL, variables related to mRNFL, variables related to mGCIPL, and variables related to mGCC. Here, each variable related to the macula is defined for the entire 10x10 macular grid (the entire macular area) (see Chart 42A in Figure 4B). For the variables related to the macula, the second calculation model Y (Model 2) uses variables defined in the region where the peripheral part of the macular area is excluded (see Chart 42B), and the third calculation model Y (Model 3) uses variables defined in the region where both the peripheral and central parts of the macular area are excluded (see Chart 42C).

[0104] All three of these computational models include the average layer thickness across the entire defined domain (cpRNFL, mRNFL, mGCIPL, and mGCC), a variable related to the difference in layer thickness between two domains (e.g., log_mGCIPL_STvsIT), and a variable related to the spikes in the TSNIT plot graph representing the layer thickness distribution around the optic nerve head (cpRNFL) (TSNITlower).

[0105] The effectiveness of the statistical analysis performed in this example was estimated and evaluated using the area under the receiver operating characteristic curve (AUC-ROC).

[0106] In the validation step of this example, the intercept and beta (slope, regression coefficient) obtained from the three regression models in the development step were applied, and the risk score was calculated using the following two equations (Equation 1 and Equation 2).

[0107] [Formula 1] Log e (p / (1-p))=Y ⇔ p=exp(Y) / (1+exp(Y)) [Formula 2] Glaucoma screening score=round(p×100) Here, round(X) is a function that rounds the fractional part of X, and represents the integer closest to X.

[0108] According to the glaucoma risk score defined in this way, the scores of all participants who underwent fundus OCT measurement fall within the range of 0 to 100. In other words, the range of the glaucoma risk score in this case is 0 to 100. Furthermore, a higher glaucoma risk score in this case indicates a higher probability of having glaucoma.

[0109] Based on the score distribution, the population was classified into three groups: a low-risk group with scores in the range of 0–49; a middle-risk group with scores in the range of 50–89; and a high-risk group with scores in the range of 90–100. Furthermore, for each group, a positive predictive value (PPV) was calculated to screen for the need for further detailed ophthalmic examination. In addition, sensitivity and specificity were evaluated using diagnostic results obtained from a subset further enhanced by Humphrey visual field testing.

[0110] To clarify the impact of axial length on glaucoma screening in this case, we examined the effect of axial length information on improving the risk score. Furthermore, to confirm the accuracy of differentiating between retinal thinning caused by glaucoma and retinal thinning caused by high myopia, we estimated axial length based on the angle (direction) of the double hump in the perioptic nerve head RNFL.

[0111] To achieve this, first, the raw value (measured value) of axial length was independently added to the third calculation model Y (Model 3), and then the axial length was added to all three calculation models Y (Model 1), Y (Model 2), and Y (Model 3), and the variable selection in the logistic regression described above was performed.

[0112] The results of this statistical analysis will be explained. Figure 5 shows the characteristics of the cases and controls in the development data used in this example.

[0113] In Figure 5, cpRNFL represents the peripapillary retinal nerve fiber layer, mRNFL represents the macular retinal nerve fiber layer, mGCIPL represents the macular ganglion cell layer-internal plexiform layer, and mGCC represents the macular ganglion cell layer complex (retinal nerve fiber layer-macular ganglion cell layer-internal plexiform layer in the macular area).

[0114] Furthermore, cpRNFLqIn represents the cpRNFL thickness in the inferior nasal quadrant, cpRNFLqIt represents the cpRNFL thickness in the inferior temporal quadrant, and cpRNFLqS represents the cpRNFL thickness in the superior quadrant.

[0115] Furthermore, log_mGCIPL_ITvsIN represents a logarithmic transformation (first calculation model) where the difference in mGCIPL thickness between the inferior temporal quadrant and the inferior nasal quadrant is the real number, log_mGCIPL_SNyvsINy represents a logarithmic transformation (third calculation model) where the difference in mGCIPL thickness between the superior nasal quadrant and the inferior nasal quadrant is the real number, log_mGCIPL_STxvsITx represents a logarithmic transformation (second calculation model) where the difference in mGCIPL thickness between the superior temporal quadrant and the inferior temporal quadrant is the real number, and log_mGCIPL_STvsIT represents the superior temporal quadrant and the inferior temporal quadrant The first calculation model shows a logarithmic transformation where the difference in mGCIPL thickness between and is the true value. log_mGCIPL_STyvsITy shows a logarithmic transformation where the difference in mGCIPL thickness between the superior temporal quadrant and the inferior temporal quadrant is the true value. log_mGCC_ITxvsINx shows a logarithmic transformation where the difference in mGCC thickness between the inferior temporal quadrant and the inferior nasal quadrant is the true value. log_mGCC_ITyvsINy shows a logarithmic transformation where the difference in mGCC thickness between the inferior temporal quadrant and the inferior nasal quadrant is the true value.

[0116] Furthermore, mGCC_IN represents the mGCC thickness in the inferior nasal quadrant (first calculation model), and mGCIPL_IT represents the mGCIPL thickness in the inferior temporal quadrant.

[0117] Additionally, TSNITlower indicates the cpRNFL thickness at the lower spike (peak) of the double hump in the TSNIT plot.

[0118] The values ​​shown in Figure 5 represent numerical values ​​(percentages in parentheses) or mean ± standard deviation. Cases are participants confirmed by an ophthalmologist to have glaucoma based on detailed ophthalmological examination, while controls are participants confirmed by an ophthalmologist to have no signs of glaucoma based on both fundus photography and FDT visual field testing. P-values ​​for age are calculated using a chi-square test, and for continuous variables, they are calculated using a t-test.

[0119] In the case group, cpRNFL was significantly thinner than in the normal control group, and the overall mean values ​​of the 10x10 grid for mRNFL, mGCIPL, and mGCC were also significantly thinner than in the normal control group. The Spearman rank correlation coefficient for cpRNFL thickness between the left and right eyes was 0.85 (P<0.01) in the control group and 0.24 (P<0.01) in the case group.

[0120] Figure 6 shows the distribution of cases and controls for each OCT measurement predictor in Figure 5. For each OCT measurement predictor, the histogram with the higher peak corresponds to the distribution of the normal control group, and the histogram with the lower peak corresponds to the distribution of the case group.

[0121] Figure 7 shows the diagnostic model for glaucoma screening developed in this example. Figure 7 shows the intercept information and retinal thickness-related predictor information for each of the three computational models Y(Model 1), Y(Model 2), and Y(Model 3). In Figure 7, B represents the slope (beta, regression coefficient), SE represents the standard error, OR represents the odds ratio, CI represents the confidence interval, and AUC-ROC represents the area under the receiver-operated characteristic curve.

[0122] After selecting the variables for the logistic regression model, negative regression coefficients were generally obtained for the predictors of retinal thickness in each region (cpRNFLqS, cpRNFLqIt, cpRNFLqIn, mGCIPL_IT, mGCC, and TSNITlower). This result suggests that a thicker retinal layer in the selected region is associated with a lower likelihood of developing glaucoma.

[0123] In contrast, positive regression coefficients were obtained for the predictors of the difference between the upper and lower regions (log_mGCIPL_STvsIT, log_mGCIPL_STxvsITx, log_mGCIPL_SNyvsINy, log_mGCIPL_STyvsITy), and for the predictors of the difference in retinal thickness between the temporal and nasal regions (log_mGCIPL_ITvsIN, log_mGCC_ITxvsINx, log_mGCC_ITyvsINy). This result suggests that a large difference in retinal thickness between the upper and lower regions and / or left and right regions indicates a high probability of glaucoma.

[0124] Furthermore, the predicted area under the receiver operation characteristic curve for the three computational models Y(Model 1), Y(Model 2), and Y(Model 3) was approximately 0.97 for all of them (see also the three ROC curves shown in Figure 8). This result indicates that there is no significant difference among these three computational models.

[0125] The inventors applied the glaucoma risk score constructed in this manner to a validation dataset (9720 eyes from 6006 individuals). Equations [1] and [2] above are used to calculate the glaucoma risk score. Specifically, the glaucoma risk score can be calculated using three calculation models: Y(Model 1), Y(Model 2), and Y(Model 3).

[0126] The results of the validation of the three calculation models for positive predictive value are explained with reference to Figure 9. Figure 9 shows the distribution of glaucoma risk scores obtained by the three calculation models and the positive predictive value (PPV) for the need for further detailed ophthalmic examination.

[0127] Figure 9 shows the results of classifying the 9,720 risk scores obtained by applying each computational model (Model 1; Model 2; Model 3) to 9,720 eyes in the validation dataset into the three levels mentioned above: "Low," "Middle," and "High." According to the results, the high-risk group (risk score of 90 or higher) corresponds to 6.1 percent of the population for the first computational model Y (Model 1), 6.2 percent for the second computational model Y (Model 2), and 6.2 percent for the third computational model Y (Model 3).

[0128] To validate the positive predictive value (and negative predictive value), two-thirds of the high-risk group, one-fifth of the intermediate-risk group, and 2 percent of the low-risk group were randomly selected from all subjects. Four ophthalmologists then assessed each selected subject based on their OCT reports. The results showed that, according to the first, second, and third calculation models, the positive predictive values ​​for screening in the high-risk group were 80.7%, 83.3%, and 90.8%, respectively. The negative predictive values ​​for the low-risk group were 87.9%, 88.4%, and 88.2%, respectively. These results indicate that the third calculation model, Y (Model 3), had the best accuracy for glaucoma screening among the three calculation models.

[0129] Next, the results of the validation of the third computational model Y (Model 3) regarding sensitivity and specificity will be explained with reference to Figures 10 and 11. Figure 10 shows the results of SD-OCT examinations and Humphrey visual field tests performed on participants (129 eyes) who were judged to have suspected glaucoma based on the interpretation of fundus photographs among the participants in the development data creation, as well as a comparison of these test results with the glaucoma risk score of the third computational model. Figure 11 shows the sensitivity and specificity of the risk score in glaucoma screening using the third computational model.

[0130] As shown in Figure 10, according to the third calculation model, 67 eyes were classified as having a high-risk score (90 or higher). Of these 67 eyes, 53 were diagnosed with glaucoma, 10 were deemed indeterminate by OCT and Humphrey visual field testing, and 4 were found to be normal.

[0131] Furthermore, as shown in Figure 11, when subjects with diseases other than glaucoma were excluded, the sensitivity of the risk score using the third calculation model was 85 percent (85.4 percent), and the specificity was 91 percent (91.3 percent).

[0132] In the false-positive group, two cases suggested preperimetric glaucoma (PPG), and in two cases, the optic nerve could not be tracked during OCT scanning. In the false-negative group, two cases presented with mild retinal optic nerve fiber layer defects (NFLD), two cases presented with narrow NFLD, and three cases presented with diffuse thinning of the retina due to myopia. Two cases were diagnosed with glaucoma, but their scores were 89 and 88, respectively, both very close to the threshold of 90 between high-risk and medium-risk scores.

[0133] The Spearman rank correlation coefficients between the risk score of the third calculation model and the mean deviation and pattern standard deviation of Humphrey visual field testing were -0.503 (P<0.001) and 0.618 (P<0.001), respectively. In cases with high risk scores and confirmed glaucoma, the mean deviation and pattern standard deviation (mean ± standard deviation) were -6.80 ± 5.91 and 6.42 ± 4.64, respectively. In the nine cases with low or medium risk scores (<90) and confirmed glaucoma (i.e., false negative cases), the mean deviation and pattern standard deviation (mean ± standard deviation) were -1.96 ± 1.53 and 3.02 ± 0.76, respectively. A significant difference between these two groups was confirmed by a t-test.

[0134] The inventors also investigated the effect of improving axial length on the risk score. First, they independently added the raw value (measured value) of axial length to the third calculation model. As a result, the odds ratio (95 percent confidence interval) for axial length was 1.30 (95 percent confidence interval: 1.02 to 1.65). However, no improvement in accuracy was observed.

[0135] Furthermore, the value of axial length was added to the OCT-related variables, and variable selection in logistic regression was performed for the first, second, and third calculation models. As a result, the odds ratios for axial length for the first, second, and third calculation models were 1.52 (95 percent confidence interval: 1.21~1.91), 1.56 (95 percent confidence interval: 1.32~2.10), and 1.52 (95 percent confidence interval: 1.20~1.93), respectively. In this case as well, no improvement in accuracy was observed.

[0136] In addition, in all three computational models, the estimated axial length angle of the double hump exhibited by the perioptic nerve head RNFL was not selected as a significant variable.

[0137] The glaucoma risk score of this embodiment, as described above, was obtained using SD-OCT and can be used for population-based glaucoma mass screening.

[0138] The development of the glaucoma risk score in this embodiment references the diagnostic logic of glaucoma specialists. This diagnostic logic includes the following seven items: (1) the absolute value of the thickness of each layer in the entire region, or the absolute value of the thickness of each layer in multiple separate regions; (2) differences in the thickness of the perioptic nerve fiber layer defect (RNFL) in the vertical direction; (3) differences in the thickness of each layer in the macular area in the vertical direction, and differences in the thickness of each layer in the macular area in the horizontal direction; (4) localization of a wedge-shaped optic nerve fiber layer defect (NFLD) around the optic nerve head; (5) differences in the double hump pattern of the thickness distribution of the perioptic nerve fiber layer defect (TSNIT plot); (6) the thickness value of the layer with the lower peak in the double hump pattern of the thickness distribution of the perioptic nerve layer defect (TSNIT plot); and (7) the value of the axial length estimated from the angle (direction) of the double hump pattern of the perioptic nerve layer defect (RNFL).

[0139] In this embodiment, the following three patterns were considered for the macular area: (A) a first pattern using data for the entire macular area (Chart 42A in Figure 4B, first calculation model); (B) a second pattern using data for a region excluding the peripheral part of the macular area to avoid the influence of relatively large blood vessels in the fundus (Chart 42B, second calculation model); and (C) a third pattern using data for a donut-shaped region excluding the peripheral part of the macular area, similar to the second pattern, as well as the central part where abnormalities do not (or rarely) occur in early glaucoma (Chart 42C, third calculation model).

[0140] In all three computational models of this embodiment, the odds ratios of variables related to vertical and horizontal differences in each layer are the most influential indicators of glaucoma. Furthermore, of these three computational models, the third model exhibits the highest sensitivity and specificity in risk assessment using only SD-OCT data.

[0141] In the verification steps of this embodiment, two steps are applied to consider the following two objectives. The first objective is to develop a score that enables the automatic detection of glaucoma findings using OCT, and the second objective is to screen for glaucoma. Therefore, in the first verification, OCT data was interpreted by a glaucoma specialist, and in the second verification, glaucoma diagnosis was evaluated using a Humphrey visual field analyzer. Cases with evenly distributed risk scores were selected, and cases that were difficult to diagnose as having glaucomatous eyes were added, considering that it is important to accurately judge even slight changes in early glaucoma in mass screening. The verification work of this embodiment, designed in this way, is intended to simulate the situation in actual mass screening (population-based mass screening) where an ophthalmologist determines the need for further detailed glaucoma examinations at a hospital.

[0142] In addition, for the purpose of glaucoma screening, a blinded panel of multiple experts was formed regarding the risk score. In practice, as mentioned above, the necessary decisions were made by the agreement of all four ophthalmologists.

[0143] To validate the accuracy of the risk score using real-world data, the dataset used for validation of this embodiment intentionally included cases with low-resolution data. The causes of low resolution included poor fixation, blinking, eye movement, and small pupils. In some of these cases, it was difficult to identify glaucoma findings due to artifacts, segmentation errors, and optic nerve tracking errors. However, the data was reviewed by four ophthalmologists until they reached a consensus before making a final decision. The risk score obtained in this embodiment proved to be highly relevant to the assessments necessary for detailed glaucoma examinations.

[0144] Furthermore, in this embodiment, the validity (effectiveness) of the risk score was examined using ophthalmic diagnosis based on Humphrey visual field testing and OCT reports, and it was confirmed that it achieved high performance and high accuracy, with a sensitivity of 85.4 percent and a specificity of 91.3 percent.

[0145] In this embodiment, four cases showing high risk scores but no visual field abnormalities were considered false positives. Several of these false positive cases were deemed highly likely to be pre-perimetric glaucoma accompanied by retinal nerve fiber layer defects detected by OCT. While this embodiment primarily aims to detect early and mid-stage glaucoma, it is also considered applicable to the identification of pre-perimetric glaucoma.

[0146] Nine glaucoma cases showed low risk scores and were false negatives. Examination of these false negative cases revealed that risk scores tend to be lower in cases of mild RNFL thinning, narrow retinal nerve fiber layer defects, or retinal layers showing diffuse thinning due to myopia.

[0147] In recent years, research into the visualization of algorithm-based artificial intelligence and deep learning models has been actively pursued. However, glaucoma identification technologies suitable for glaucoma mass screening have not yet been put into practical use due to problems such as the black-box nature of algorithms and the complexity of implementation in mass screening. The algorithm according to this embodiment, as mentioned above, is capable of detecting glaucoma with high reliability and can also detect normal-tension glaucoma. Furthermore, it can be implemented to provide a glaucoma risk score by installing simple calculation software on existing OCT devices and computers, making it suitable for the practical application of glaucoma mass screening.

[0148] Furthermore, in this embodiment, the cutoff value (a threshold for dividing the range of risk scores according to the degree of risk) can be adjusted or set according to conditions such as the prevalence of the target population for mass screening. In the example above, the lower limit of the risk score (cutoff value) for determining that further detailed examination is necessary is set to 90 points. The risk score in this example has been shown to provide a screening with a high positive predictive value, and in fact, 6 percent of the approximately 9,000 datasets were included in the glaucoma prevalence rate of 5 percent.

[0149] Although two glaucoma cases in the validation dataset used to verify sensitivity and specificity are judged to be false negatives, considering their risk scores of 88 and 89, sensitivity can be improved by adjusting (slightly lowering) the cutoff value. Since the most important characteristic in mass screening is considered to be a high positive predictive value, the computational model according to this embodiment can be said to have both high feasibility and high reliability.

[0150] As described above, this embodiment provides an OCT screening algorithm suitable for glaucoma screening. Given the characteristics of glaucoma as an irreversible progressive disease, observing changes over time is extremely important. Therefore, it is considered more appropriate to express the new findings of this embodiment from the perspective of a risk score rather than from the perspective of setting a clear cutoff value.

[0151] Furthermore, in highly myopic eyes, the axial length of the eye is elongated, which can lead to deformation of the eyeball. OCT images of such eyes may reduce the reliability of data for glaucoma screening. Therefore, differentiation from glaucoma becomes difficult as the degree of myopia increases. On the other hand, regardless of glaucoma, the distribution of RNFL thickness changes significantly with axial length. As mentioned above, using the estimated axial length obtained from the angle (direction) of the double hump in the TSNIT plot did not improve the accuracy of the risk score. Similarly, taking raw axial length data into account did not improve the quality of the risk score.

[0152] This embodiment has several advantages, some of which are described below.

[0153] The first strength lies in the fact that ophthalmologists actually interpret a large number of images used for analysis, and computational models are constructed using a larger dataset than conventional methods, obtained from these interpretations. In other words, the construction of the glaucoma risk score algorithm according to this embodiment utilizes a dataset that is suitable in terms of both quality and quantity (a high-quality and large-scale dataset).

[0154] The second strength is that all the data used to construct the algorithm in this embodiment is population-based data and is free from hospital bias (for example, differences in the average severity of diseases depending on the size, history, and location of the hospital).

[0155] A third strength is that the algorithm in this embodiment does not provide glaucoma diagnostic results (risk scores) as a black box output, but rather can explain the reasons for obtaining those results from an ophthalmological perspective. This makes it possible, for example, to improve the quality of informed consent.

[0156] The fourth strength is that the algorithm according to this embodiment was constructed using data from participants who received appropriate follow-up at the medical institutions to which the inventors belong. For example, the inventors conducted the final verification of the algorithm according to this embodiment based on the results of examinations performed at an appropriate time after the initial consultation (in fact, four years after the initial consultation), taking into account the progression of glaucoma.

[0157] The primary purpose of the glaucoma risk score according to this embodiment is not to provide a definitive diagnosis of glaucoma, but rather to provide information (decision-making support information) that helps physicians determine whether a detailed examination at a hospital is necessary for the subject. As explained above, the glaucoma risk score according to this embodiment makes it possible to provide a high-quality glaucoma identification technology suitable for mass screening.

[0158] Furthermore, the use of the glaucoma risk score according to this embodiment is not limited to the provision of the decision-making support information described above (typically, glaucoma mass screening). For example, the glaucoma risk score according to this embodiment can be used for the final diagnosis and treatment of glaucoma.

[0159] <Ophthalmological equipment> Next, we will describe an ophthalmic device that is a non-limiting example of the glaucoma screening method according to the embodiment described above. Any aspect of the glaucoma screening method according to the embodiment can be combined, at least partially, with the ophthalmic device according to any of the embodiments. Furthermore, any known technology can be combined with the ophthalmic device according to any of the embodiments.

[0160] The ophthalmic device according to this embodiment is used, for example, in mass screening for ophthalmic diseases including glaucoma, and generates information indicating the glaucoma risk of the examined eye (glaucoma risk information) by processing fundus OCT data using a computational model.

[0161] The calculation model in this embodiment may be any of the first calculation model Y (Model 1), the second calculation model Y (Model 2), and the third calculation model Y (Model 3) described above, but is not limited to these.

[0162] In some cases, two or more calculation models can be selectively used. The method of selecting the calculation models is arbitrary. For example, consistency in screening can be ensured by constructing a system that has a function to record the type of calculation model applied to each subject (each eye examined, each subject examined, each group, etc.) for each screening, and a function to notify the facility where the new screening is conducted (or the ophthalmic equipment used for the new screening) of the type of calculation model used in past screenings when a new screening is conducted. Furthermore, the quality of screening can be improved by constructing a system that has a function to notify when a new calculation model is introduced and / or when an existing calculation model is updated.

[0163] The method by which the ophthalmic device in this embodiment acquires fundus OCT data may be arbitrary. Some examples of ophthalmic devices may include an OCT scanner that applies an OCT scan to the fundus of the eye being examined. Other examples of ophthalmic devices may include an element that acquires OCT data generated by applying an OCT scan using another ophthalmic device (ophthalmic OCT device) that includes an OCT scanner to the fundus of the eye being examined, from another device. This other device may be, for example, the ophthalmic OCT device that generated the OCT data, a medical image archiving system (e.g., Picture Archiving and Communication System (PACS)) that holds the OCT data, or a recording medium on which the OCT data is recorded. The element that acquires OCT data from another device may be, for example, a communication device for data communication with the ophthalmic OCT device or the medical image archiving system, or a drive device for reading OCT data from the recording medium.

[0164] The glaucoma risk information in this embodiment may be information that expresses any type of glaucoma risk in any form or manner. Some examples of glaucoma risk information may include the glaucoma risk score described above, or they may include information indicating the degree of glaucoma risk (e.g., high risk, medium risk, low risk, etc.).

[0165] Figure 12 shows one example of the configuration of an ophthalmic device according to this embodiment. The ophthalmic device 120 includes a processor 121, a memory 125, an OCT data acquisition unit 127, and a user interface (UI) 128.

[0166] The processor 121 is configured to perform various processes based on programs installed in the ophthalmic device 120 and / or programs stored in devices accessible by the ophthalmic device 120. For example, the processor 121 can perform control of each element of the ophthalmic device 120, control of peripheral devices of the ophthalmic device 120, and various data processing.

[0167] Memory 125 is the storage unit of the ophthalmic device 120. Memory 125 may include any form of storage device, for example, primary storage devices such as main memory, cache memory, and working memory, secondary storage devices such as hard disk drives, flash memory, and RAM disks, tertiary storage devices, offline storage, and registers.

[0168] Memory 125 holds various data processed by the ophthalmic device 120. Memory 125 stores a calculation model 126. The calculation model 126 may be any calculation model according to the above embodiment. The calculation model 126 may be a linear equation including a linear combination of one or more optic disc parameters representing the retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing the retinal tissue thickness in the macular area.

[0169] The calculation model 126 is created in advance and stored in memory 125. For example, the calculation model 126 may be any form of calculation model created using at least part of the glaucoma screening method according to the above embodiment, and may include at least one of the first calculation model Y (Model 1), the second calculation model Y (Model 2), and the third calculation model Y (Model 3).

[0170] The OCT data acquisition unit 127 acquires OCT data of the fundus of a living eye (tested eye) that is the subject of glaucoma screening by the ophthalmic device 120. The OCT data acquisition unit 127 may include, for example, the aforementioned OCT scanner and / or an element for acquiring the aforementioned OCT data from another device.

[0171] An OCT scanner, for example, includes, as in conventional scanners, a light source, a demultiplexer that splits the light emitted from the light source into measurement light and reference light, a sample arm that projects the measurement light onto the fundus of the eye under examination and guides the reflected light (reflected light, backscattered light, etc.) from the eye under examination, a reference arm that guides the reference light, a multiplexer that superimposes the reflected light of the measurement light and the reference light to generate interference light, and a photodetector that detects the interference light and generates an electrical signal (detection signal). The sample arm is equipped with an optical deflector (optical scanner), such as a galvanometer scanner, to project the measurement light to multiple different positions on the fundus of the eye under examination.

[0172] Furthermore, the OCT scanner includes an image reconstruction unit (image reconstruction circuit, image reconstruction circuit configuration) for constructing OCT data (typically OCT image data) by applying signal processing to the detection signal generated by the photodetector. The image reconstruction unit, for example, generates a reflectance intensity profile (A-line profile) for each A-line from the spectral distribution based on data for each projection position of the measurement light (scan point, A-line), as in the conventional method, images each A-line profile to generate multiple A-scan image data, and arranges these A-scan image data according to a scan pattern (arrangement of multiple scan points). The signal processing for generating the A-line profiles includes noise reduction (denoising), filtering, and fast Fourier transform (FFT).

[0173] The OCT modality used in an OCT scanner can be arbitrary, for example, spectral domain OCT or swept-source OCT. Spectral domain OCT is an OCT modality that splits light emitted from a low-coherence light source (broadband light source) into measurement light and reference light, projects the measurement light onto the object under test, superimposes the reflected light from the object under test with the reference light to generate interference light, detects the spectral distribution of the generated interference light with a spectrometer, and constructs image data by applying data processing such as Fourier transform to the detected spectral distribution.

[0174] On the other hand, swept-source OCT is a method that splits light from a tunable light source into measurement light and reference light, projects the measurement light onto the object under test, and generates interference light by superimposing the reflected measurement light from the object with the reference light. The generated interference light is detected by a photodetector, and image data is constructed by applying data processing such as Fourier transform to the detection data collected according to the wavelength sweep and measurement light scan.

[0175] In short, spectral domain OCT is an OCT modality that acquires the spectral distribution of interference light using spatial resolution, while swept-source OCT is an OCT method that acquires the spectral distribution of interference light using time resolution. For non-specific examples of spectral domain OCT, see, for example, U.S. Patent No. 9,295,386. For non-specific examples of swept-source OCT, see, for example, U.S. Patent No. 1,101,3400.

[0176] The user interface 128 is an interface for exchanging information between the ophthalmic device 120 and the user, and includes, for example, an input interface and an output interface. The input interface is an interface for the user to input information to the ophthalmic device 120, and examples of such interfaces include a pointing device, keyboard, switch, voice input device, and image (video) input device. The output interface is an interface for outputting information from the ophthalmic device 120, and examples of such interfaces include a display, voice output device, and printer.

[0177] The ophthalmic device 120 in this example may include various elements not shown. For example, it may include any elements described in the above-mentioned literature (U.S. Patent No. 9,295,386, U.S. Patent No. 1,101,3400) or any elements described in other literature relating to OCT.

[0178] The processor 121 includes a control unit (control circuit, control circuit configuration) 122, a measurement value calculation unit (measurement value calculation circuit, measurement value calculation circuit configuration) 123, and a risk information generation unit (risk information generation circuit, risk information generation circuit configuration) 124.

[0179] The control unit 122 is configured to perform control related to the measurement value calculation unit 123, control related to the risk information generation unit 124, control related to the memory 125, control related to the OCT data acquisition unit 127, and control related to the user interface 128.

[0180] Control over the measurement value calculation unit 123 may include, for example, turning the measurement value calculation unit 123 on / off, controlling the parameters of the measurement value calculation unit 123, inputting data to the measurement value calculation unit 123, and receiving output data from the measurement value calculation unit 123. Control over the risk information generation unit 124 may be similar.

[0181] Control over memory 125 may include, for example, processes for storing (saving) data in memory 125, processes for retrieving data from memory 125, and processes for processing or editing data stored in memory 125.

[0182] Control of the OCT data acquisition unit 127 may include, for example, control of the light source, control of the optical scanner, control of the optical path length of the sample arm and / or reference arm, control of the focus of the measurement light, control of the polarization state of the measurement light and / or reference light, control of the light quantity (intensity) of the measurement light and / or reference light, control of the photodetector, and control of a data acquisition system (DAS, DAQ) not shown.

[0183] Control over the user interface 128 may include, for example, display control and graphical user interface (GUI) control.

[0184] The control unit 122 can perform control of other elements of the ophthalmic device 120. This control may be performed in the same manner as in the conventional method (e.g., alignment control, focus control, tracking control, etc.).

[0185] The control unit 122 inputs the OCT data of the fundus of the eye being examined, generated by the OCT data acquisition unit 127, to the measurement value calculation unit 123. The measurement value calculation unit 123 is configured to calculate a number of measurement values ​​based on this OCT data, which are then input into the calculation model 126 used for glaucoma screening of the eye being examined.

[0186] If the calculation model 126 includes two or more calculation models, for example, the control unit 122 sends information (measurement type information) indicating the type of measurement value corresponding to the calculation model used for glaucoma screening of the eye under examination to the measurement value calculation unit 123. Based on this measurement type information, the measurement value calculation unit 123 recognizes the types of multiple measurement values ​​to be calculated and calculates each of these measurement values ​​from the OCT data of the fundus of the eye under examination.

[0187] As described above, some example calculation models 126 may be linear equations that include a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area. When such a calculation model 126 is used for glaucoma screening of the eye under examination, the multiple measurements calculated by the measurement calculation unit 123 include one or more optic disc measurements corresponding to each of the one or more optic disc parameters and one or more macular measurements corresponding to each of the one or more macular parameters.

[0188] If the calculation model 126 used for glaucoma screening of the eye under examination is the first calculation model Y (Model 1), the measurement value calculation unit 123 calculates the measurement value corresponding to TSNITlower, the measurement value corresponding to cpRNFLqS, the measurement value corresponding to cpRNFLqIt, the measurement value corresponding to cpRNFLqIn, the measurement value corresponding to mGCIPL_IT, the measurement value corresponding to log_mGCIPL_STvsIT, the measurement value corresponding to log_mGCIPL_ITvsIN, and the measurement value corresponding to mGCC_IN based on the OCT data of the fundus of the eye under examination.

[0189] If the calculation model 126 used for glaucoma screening of the eye under examination is the second calculation model Y (Model 2), the measurement value calculation unit 123 calculates the measurement value corresponding to TSNITlower, the measurement value corresponding to cpRNFLqS, the measurement value corresponding to cpRNFLqIt, the measurement value corresponding to cpRNFLqIn, the measurement value corresponding to log_mGCIPL_STxvsITx, and the measurement value corresponding to log_mGCC_ITxvsINx based on the OCT data of the fundus of the eye under examination.

[0190] If the calculation model 126 used for glaucoma screening of the eye under examination is the third calculation model Y (Model 3), the measurement value calculation unit 123 calculates the measurement value corresponding to TSNITlower, the measurement value corresponding to cpRNFLqIn, the measurement value corresponding to mGCC, the measurement value corresponding to log_mGCIPL_SNyvsINy, the measurement value corresponding to log_mGCIPL_STyvsITy, and the measurement value corresponding to log_mGCC_ITyvsINy based on the OCT data of the fundus of the eye under examination.

[0191] Similarly, when a different type of calculation model 126 is used, the measurement value calculation unit 123 obtains the measurement values ​​corresponding to each of the multiple parameters (multiple variables) included in that calculation model 126 from the OCT data of the fundus of the eye being examined.

[0192] The risk information generation unit 124 generates glaucoma risk information for the eye under examination based on multiple measurement values ​​calculated from the OCT data of the fundus of the eye under examination by the measurement value calculation unit 123 and a calculation model 126 used for glaucoma screening of the eye under examination.

[0193] In some examples, the risk information generation unit 124 substitutes each measurement value calculated by the measurement value calculation unit 123 into the corresponding parameter (variable) of the calculation model 126 and calculates a value (e.g., glaucoma risk score) based on the calculation model 126. The glaucoma risk information generated and output by the risk information generation unit 124 in this example includes, for example, the result of a calculation using the calculation model 126 (e.g., glaucoma risk score), and information obtained by applying a predetermined algorithm (procedure, calculation method) to the result of the calculation (e.g., information indicating the degree of glaucoma risk), or both.

[0194] If the calculation model 126 used for glaucoma screening of the eye under examination is the first calculation model Y (Model 1), the risk information generation unit 124 will use multiple measured values ​​as shown below in the first calculation model Y (Model 1). The glaucoma risk score is calculated by substituting the values ​​into multiple parameters (multiple variables) in 1): (1) Substitute the measured value corresponding to TSNITlower into the variable TSNITlower; (2) Substitute the measured value corresponding to cpRNFLqS into the variable cpRNFLqS; (3) Substitute the measured value corresponding to cpRNFLqIt into the variable cpRNFLqIt; (4) Substitute the measured value corresponding to cpRNFLqIn into the variable cpRNFLqIn; (5) Substitute the measured value corresponding to mGCIPL_IT into the variable mGCIPL_IT; (6) Substitute the measured value corresponding to log_mGCIPL_STvsIT into the variable log_mGCIPL_STvsIT; (7) Substitute the measured value corresponding to log_mGCIPL_ITvsIN into the variable log_mGCIPL_ITvsIN; (8) Substitute the measured value corresponding to mGCC_IN into the variable mGCC_IN. Furthermore, the risk information generation unit 124 in this example can generate glaucoma risk information that includes the glaucoma risk score calculated in this manner, and / or information indicating the degree of glaucoma risk (e.g., high risk, medium risk, or low risk) derived from the glaucoma risk score.

[0195] If the calculation model 126 used for glaucoma screening of the eye under examination is the second calculation model Y (Model 2), the risk information generation unit 124 calculates the glaucoma risk score by substituting multiple measured values ​​into multiple parameters (multiple variables) of the second calculation model Y (Model 2) as follows: (1) Substitute the measured value corresponding to TSNITlower into the variable TSNITlower; (2) Substitute the measured value corresponding to cpRNFLqS into the variable cpRNFLqS; (3) Substitute the measured value corresponding to cpRNFLqIt into the variable cpRNFLqIt; (4) Substitute the measured value corresponding to cpRNFLqIn into the variable cpRNFLqIn; (5) Substitute the measured value corresponding to log_mGCIPL_STxvsITx into the variable log_mGCIPL_STxvsITx; (6) Substitute the measured value corresponding to log_mGCC_ITxvsINx into the variable log_mGCC_ITxvsINx. Furthermore, the risk information generation unit 124 in this example can generate glaucoma risk information that includes the glaucoma risk score calculated in this manner, and / or information indicating the degree of glaucoma risk (e.g., high risk, medium risk, or low risk) derived from the glaucoma risk score.

[0196] If the calculation model 126 used for glaucoma screening of the eye under examination is the third calculation model Y (Model 3), the risk information generation unit 124 calculates the glaucoma risk score by substituting multiple measured values ​​into multiple parameters (multiple variables) of the third calculation model Y (Model 3) as follows: (1) Substitute the measured value corresponding to TSNITlower into the variable TSNITlower; (2) Substitute the measured value corresponding to cpRNFLqIn into the variable cpRNFLqIn; (3) Substitute the measured value corresponding to mGCC into the variable mGCC; (4) Substitute the measured value corresponding to log_mGCIPL_SNyvsINy into the variable log_mGCIPL_SNyvsINy; (5) Substitute the measured value corresponding to log_mGCIPL_STyvsITy into the variable log_mGCIPL_STyvsITy; (6) Substitute the measured value corresponding to log_mGCC_ITyvsINy into the variable log_mGCC_ITyvsINy. Furthermore, the risk information generation unit 124 in this example can generate glaucoma risk information that includes the glaucoma risk score calculated in this manner, and / or information indicating the degree of glaucoma risk (e.g., high risk, medium risk, or low risk) derived from the glaucoma risk score.

[0197] Similarly, when a different type of calculation model 126 is used, the risk information generation unit 124 can substitute multiple measurements calculated by the measurement value calculation unit 123 into multiple variables of the calculation model 126 used for glaucoma screening of the eye under examination, in order to generate glaucoma risk information.

[0198] The control unit 122 performs control to provide glaucoma risk information generated by the risk information generation unit 124. In some examples, the control unit 122 can display the glaucoma risk information on the display of the user interface 128. In some examples, the control unit 122 can control a communication device (not shown) to transmit the glaucoma risk information to another device (e.g., a device, system, recording medium, etc.).

[0199] If the calculation model 126 includes multiple calculation models, and glaucoma screening of the eye under examination is performed using two or more of these calculation models, the measurement value calculation unit 123 calculates two or more groups of measurement values ​​corresponding to each of the two or more calculation models from the OCT data of the fundus of the eye under examination, and the risk information generation unit 124 generates glaucoma risk information based on each of the two or more groups of measurement values. As a result, two or more glaucoma risk information corresponding to each of the two or more calculation models used for glaucoma screening of the eye under examination is obtained.

[0200] In some cases, the risk information generation unit 124 can provide two or more pieces of glaucoma risk information.

[0201] In some cases, the risk information generation unit 124 can select one or more glaucoma risk information from the two or more glaucoma risk information and provide the selected one or more glaucoma risk information.

[0202] In some cases, the risk information generation unit 124 can generate final glaucoma risk information based on the two or more glaucoma risk information and provide the final glaucoma risk information (and the two or more glaucoma risk information).

[0203] In some examples, the risk information generation unit 124 can generate a final glaucoma risk score and / or final information indicating the degree of glaucoma risk based on two or more glaucoma risk scores. The process of generating final glaucoma risk information from two or more glaucoma risk information may include, for example, calculating statistical values ​​(e.g., mean, weighted mean, median, standard deviation, variance, etc.) of the two or more glaucoma risk scores, and removing outliers from the two or more glaucoma risk scores. The final information indicating the degree of glaucoma risk may be generated, for example, from the final glaucoma risk score.

[0204] In some cases, the risk information generation unit 124 can generate final information indicating the degree of glaucoma risk based on multiple pieces of information indicating the degree of glaucoma risk. This final information generation process may include, for example, a process of selecting one piece of information from the multiple pieces of information by majority vote, a process of comparing the information obtained from the eye (person) being examined in the current glaucoma screening with the multiple pieces of information and selecting one piece of information from the multiple pieces of information, or a process of comparing the information obtained from past examinations of the eye (person) being examined with the multiple pieces of information and selecting one piece of information from the multiple pieces of information.

[0205] As described above, the ophthalmic device 120 according to this embodiment performs glaucoma risk assessment of the eye under examination using the glaucoma screening method according to the above embodiment, and therefore can provide a high-quality glaucoma identification technology suitable for mass screening.

[0206] <Program, recording medium> A program and recording medium, which are non-limiting examples of the glaucoma screening method according to the above embodiment, will be described.

[0207] The program according to this embodiment is a novel program to be executed by a computer. This computer includes a processor and memory. This processor may have a configuration similar to that of the processor 121 of the exemplary ophthalmic device 120 shown in Figure 12. That is, the processor according to this embodiment may include a processor that functions as a control unit 122, a processor that functions as a measurement value calculation unit 123, and a processor that functions as a risk information generation unit 124. Furthermore, the memory according to this embodiment may hold a calculation model (126) similar to the memory 125 of the ophthalmic device 120.

[0208] The program according to this embodiment is configured to cause the processor to execute a procedure that includes the following series of steps.

[0209] In the first step, the processor performs control to store the computational model in memory according to the program of this embodiment. This computational model may be any of the computational models according to the above embodiments, for example, a linear equation including a linear combination of one or more optic disc parameters representing the retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing the retinal tissue thickness in the macular area.

[0210] In the second step, the processor performs control (OCT data reception control) to receive OCT data of the fundus of the eye being examined, in accordance with the program according to this embodiment.

[0211] In some examples, OCT data reception control may include the process of acquiring OCT data obtained from the fundus of the eye being examined by another ophthalmic device (ophthalmic OCT device) including an OCT scanner, from said ophthalmic OCT device, a medical image archiving system, or a recording medium.

[0212] In some examples, the computer is an element of an ophthalmic device (e.g., ophthalmic device 120). The ophthalmic device also includes an OCT data acquisition unit (e.g., OCT data acquisition unit 127). In this case, the OCT data reception control may include control to cause the OCT data acquisition unit to perform an OCT scan of the fundus of the eye under examination, and control to cause the OCT data acquisition unit to generate OCT data based on the data acquired by the OCT scan.

[0213] In other words, the OCT data reception control performed in the second step may include either control for the processor to receive OCT data generated by another ophthalmic device from outside the computer, or control for the processor to receive OCT data generated by the ophthalmic device itself, including the computer.

[0214] In the third step, the processor performs a process (measurement value calculation process) to calculate multiple measurement values ​​from the OCT data acquired in the second step, in accordance with the program according to this embodiment.

[0215] In some cases, the measurement calculation process calculates multiple measurements corresponding to the calculation model stored in memory in the first step, based on the OCT data acquired in the second step. If the calculation model is a linear equation including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area, the measurement calculation process includes calculating multiple measurements, based on the OCT data acquired in the second step, including one or more optic disc measurements corresponding to each of the one or more optic disc parameters and one or more macular measurements corresponding to each of the one or more macular parameters.

[0216] In the fourth step, the processor performs a process to input the multiple measurement values ​​calculated in the third step into the calculation model stored in memory in the first step, in accordance with the program according to this embodiment.

[0217] If the calculation model stored in memory in the first step is a linear equation including a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area, and if in the third step multiple measurements are calculated including one or more optic disc measurements corresponding to each of the one or more optic disc parameters and one or more macular measurements corresponding to each of the one or more macular parameters, then in the fourth step, the corresponding optic disc measurements and corresponding macular measurements are input to each optic disc parameter (variable) and each macular parameter (variable) of the calculation model. This derives a glaucoma risk score for the eye under examination. The processor can generate glaucoma risk information from this glaucoma risk score.

[0218] In the fifth step, the processor performs a process to provide glaucoma risk information generated in the fourth step, in accordance with the program according to this embodiment. In other words, in the fifth step, the processor performs a process to provide calculation results output from the calculation model in response to the input of multiple measurement values ​​calculated in the third step.

[0219] Thus, since the program according to this embodiment performs glaucoma risk assessment of the eye being examined using the glaucoma screening method according to the above embodiment, it is possible to provide a high-quality glaucoma identification technology suitable for mass screening.

[0220] The program according to this embodiment can be recorded on a recording medium. In other words, the recording medium according to this embodiment is a recording medium on which the program according to this embodiment is recorded. The recording medium according to this embodiment is a computer-readable non-temporary recording medium. The computer-readable non-temporary recording medium according to this embodiment may be any form of recording medium, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.

[0221] It is possible to combine, at least partially, any aspect of the glaucoma screening method according to the embodiment, and / or any aspect of the ophthalmic apparatus according to the embodiment, with a program according to any embodiment. Furthermore, it is possible to combine any known technology with a program according to any embodiment.

[0222] Similarly, any aspect of the glaucoma screening method according to the embodiment, and / or any aspect of the ophthalmic apparatus according to the embodiment, can be combined, at least partially, with a recording medium according to any embodiment. Furthermore, any known technology can be combined with a recording medium according to any embodiment.

[0223] <Several non-limiting features> The following describes some non-limiting features of the various embodiments detailed in this disclosure. However, the features of the embodiments are not limited to those described above.

[0224] A first example of the glaucoma screening method according to the embodiment includes a calculation model preparation step, an OCT data generation step, a measurement value calculation step, a measurement value input step, and a calculation result provision step.

[0225] In the computational model preparation step, a computational model is prepared that includes a linear combination of one or more optic disc parameters representing retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing retinal tissue thickness in the macular area.

[0226] In the OCT data generation step, optical coherence tomography (OCT) scans are applied to the fundus of the eye being examined to generate OCT data.

[0227] In the measurement calculation step, based on the OCT data generated in the OCT data generation step, multiple measurements are calculated, including one or more nipple measurements corresponding to one or more nipple parameters in the calculation model prepared in the calculation model preparation step, and one or more macular measurements corresponding to one or more macular parameters.

[0228] In the measurement input step, multiple measurement values ​​calculated in the measurement calculation step are input into the calculation model prepared in the calculation model preparation step.

[0229] In the calculation result provision step, the calculation results output from the calculation model are provided in response to the input of multiple measurement values ​​to the calculation model performed in the measurement value input step.

[0230] A second embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the first embodiment: One or more optic disc parameters in the computational model prepared in the computational model preparation step include optic disc parameters representing the thickness of the retinal nerve fiber layer in the optic disc area.

[0231] A third embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the second embodiment: The computational model prepared in the computational model preparation step includes a linear combination of two or more optic disc parameters. The two or more optic disc parameters include a first optic disc parameter (TSNITlower) that indicates the thickness value corresponding to the lower of the two peaks of a double hump exhibited by a TSNIT plot representing the distribution of retinal nerve fiber layer thickness in a predetermined circular region around the optic disc. Furthermore, the two or more optic disc parameters include a second optic disc parameter (cpRNFLqIn) that indicates the average value of the retinal nerve fiber layer thickness in an arc-shaped region in the circular region around the optic disc where the angle defined as zero in the temporal direction and positive in the direction of rotation upward from the temporal direction is in the range of 225 to 285 degrees.

[0232] A fourth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the third embodiment. The computational model prepared in the computational model preparation step includes a linear combination of four optic disc parameters. These four optic disc parameters include a first optic disc parameter (TSNITlower) and a second optic disc parameter (cpRNFLqIn), as well as a third optic disc parameter (cpRNFLqS) and a fourth optic disc parameter (cpRNFLqIt). The third optic disc parameter (cpRNFLqS) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 45 to 135 degrees within the circular region around the optic nerve head. The fourth optic disc parameter (cpRNFLqIt) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 285 to 315 degrees within the circular region around the optic nerve head.

[0233] A fifth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the first to fourth embodiments: The computational model prepared in the computational model preparation step includes a linear combination of two or more macular parameters. These two or more macular parameters include a first macular parameter based on the thickness of the complex tissue of ganglion cells and the inner plexiform layer (mGCIPL) in the macular area, and a second macular parameter based on the thickness of the retinal ganglion cell complex (mGCC) in the macular area.

[0234] A sixth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameter relating to mGCIPL) includes a layer thickness parameter (mGCIPL_IT) that represents the thickness of the composite tissue (mGCIPL).

[0235] A seventh embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameters relating to mGCIPL) includes comparison value parameters (log_mGCIPL_STvsIT, log_mGCIPL_ITvsIN;log_mGCIPL_STxvsITx;log_mGCIPL_SNyvsINy, log_mGCIPL_STyvsITy) that represent a comparison value between the first composite tissue thickness in a first sub-area of ​​the macular area and the second composite tissue thickness in a second sub-area.

[0236] An eighth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the seventh embodiment: The comparison value, represented by the comparison value parameter included in the first macular parameter (parameter related to mGCIPL), is calculated by applying a logarithmic operation to a number that includes the difference between the first composite tissue thickness in the first sub-area of ​​the macular area and the second composite tissue thickness in the second sub-area.

[0237] A ninth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the fifth to eighth embodiments: The second macular parameter (parameter related to mGCC) includes a layer thickness parameter (mGCC_IN;mGCC) representing the thickness of the retinal ganglion cell complex (mGCC).

[0238] A tenth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the fifth to eighth embodiments: The second macular parameter (parameter relating to mGCC) includes a comparison parameter (log_mGCC_ITxvsINx;log_mGCC_ITyvsINy) that represents a comparison value between the thickness of the first retinal ganglion cell complex in the third sub-area of ​​the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0239] An eleventh embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the tenth embodiment: The comparison value, represented by the comparison value parameter included in the second macular parameter (parameter related to mGCC), is calculated by applying a logarithmic operation to a number that includes the difference between the thickness of the first retinal ganglion cell complex in the third sub-area of ​​the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0240] A twelfth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the first to eleventh embodiments: In the computational model preparation step, a computational model is created based on a development dataset collected from a first population, a first validation dataset collected from a second population different from the first population, and a second validation dataset collected from a subpopulation in the first population that was determined to have suspected glaucoma.

[0241] A thirteenth embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the twelfth embodiment: The developed dataset includes fundus photographs acquired by digital fundus photography, fundus OCT data acquired by optical coherence tomography, and visual field data acquired by visual field testing.

[0242] A 14th embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of the 12th or 13th embodiment: In the computational model preparation step, the first validation dataset includes fundus OCT data acquired by optical coherence tomography and is used to validate the positive predictive value.

[0243] A 15th embodiment of the glaucoma screening method according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the 12th to 14th embodiments: In the computational model preparation step, the second validation dataset includes visual field data obtained by visual field testing and is used for sensitivity and specificity validation.

[0244] Any of the first to fifteen embodiments of the glaucoma screening method according to this embodiment can be combined, at least partially, with any of the matters described or suggested in this disclosure.

[0245] A first embodiment of the ophthalmic device according to this embodiment includes a storage unit, an OCT data acquisition unit, a measurement value calculation unit, and a risk information generation unit.

[0246] The memory unit holds a pre-created calculation model. The calculation model includes a linear combination of one or more optic disc parameters representing the retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing the retinal tissue thickness in the macular area. The memory 125 of the ophthalmic device 120 shown in Figure 12 is one example of the memory unit according to this embodiment.

[0247] The OCT data acquisition unit is configured to acquire optical coherence tomography data (OCT data) of the fundus of the eye being examined. The OCT data acquisition unit 127 of the ophthalmic device 120 shown in Figure 12 is one example of the OCT data acquisition unit according to this embodiment.

[0248] The measurement value calculation unit is configured to calculate multiple measurement values ​​based on OCT data acquired by the OCT data acquisition unit, including one or more optic disc measurement values ​​corresponding to one or more optic disc parameters in the calculation model stored in the memory unit, and one or more macular measurement values ​​corresponding to one or more macular parameters. The measurement value calculation unit 123 of the ophthalmic device 120 shown in Figure 12 is one example of the measurement value calculation unit according to this embodiment.

[0249] The risk information generation unit is configured to generate glaucoma risk information for the eye being examined based on a calculation model stored in the memory unit and a plurality of measurement values ​​calculated by the measurement value calculation unit. The risk information generation unit 124 of the ophthalmic device 120 shown in Figure 12 is one example of a risk information generation unit according to this embodiment.

[0250] A second embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of the first embodiment: One or more optic disc parameters in the computational model stored in the memory unit include optic disc parameters representing the thickness of the retinal nerve fiber layer in the optic disc area.

[0251] A third embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of the second embodiment: The computational model stored in the memory unit includes a linear combination of two or more optic disc parameters. The two or more optic disc parameters include a first optic disc parameter (TSNITlower) that indicates the thickness value corresponding to the lower of the two peaks of a double hump exhibited by a TSNIT plot representing the distribution of retinal nerve fiber layer thickness in a predetermined circular region around the optic nerve disc. Furthermore, the two or more optic disc parameters include a second optic disc parameter (cpRNFLqIn) that indicates the average value of the retinal nerve fiber layer thickness in an arc-shaped region in the circular region around the optic nerve disc where the angle defined as zero in the temporal direction and positive in the direction of rotation upward from the temporal direction is in the range of 225 to 285 degrees.

[0252] A fourth embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of the third embodiment: The calculation model stored in the memory unit includes a linear combination of four optic disc parameters. These four optic disc parameters include a first optic disc parameter (TSNITlower) and a second optic disc parameter (cpRNFLqIn), as well as a third optic disc parameter (cpRNFLqS) and a fourth optic disc parameter (cpRNFLqIt). The third optic disc parameter (cpRNFLqS) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 45 to 135 degrees within the circular region around the optic nerve head. The fourth optic disc parameter (cpRNFLqIt) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 285 to 315 degrees within the circular region around the optic nerve head.

[0253] A fifth embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the first to fourth embodiments: The computational model stored in the memory unit includes a linear combination of two or more macular parameters. These two or more macular parameters include a first macular parameter based on the thickness of the complex tissue of ganglion cells and the inner plexiform layer (mGCIPL) in the macular area, and a second macular parameter based on the thickness of the retinal ganglion cell complex (mGCC) in the macular area.

[0254] A sixth embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameter relating to mGCIPL) includes a layer thickness value parameter (mGCIPL_IT) that represents the thickness of the composite tissue (mGCIPL).

[0255] A seventh embodiment of the ophthalmic device according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameter relating to mGCIPL) includes comparison value parameters (log_mGCIPL_STvsIT, log_mGCIPL_ITvsIN;log_mGCIPL_STxvsITx;log_mGCIPL_SNyvsINy, log_mGCIPL_STyvsITy) that represent a comparison value between the first composite tissue thickness in a first sub-area of ​​the macular area and the second composite tissue thickness in a second sub-area.

[0256] An eighth embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of the seventh embodiment: The comparison value, represented by the comparison value parameter included in the first macular parameter (parameter relating to mGCIPL), is calculated by applying a logarithmic operation to a number that includes the difference between the first composite tissue thickness in the first sub-area of ​​the macular area and the second composite tissue thickness in the second sub-area.

[0257] A ninth embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the fifth to eighth embodiments: The second macular parameter (parameter relating to mGCC) includes a layer thickness parameter (mGCC_IN;mGCC) representing the thickness of the retinal ganglion cell complex (mGCC).

[0258] A tenth embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the fifth to eighth embodiments: The second macular parameter (parameter relating to mGCC) includes a comparison parameter (log_mGCC_ITxvsINx;log_mGCC_ITyvsINy) that represents a comparison value between the thickness of the first retinal ganglion cell complex in the third sub-area of ​​the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0259] An eleventh embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of the tenth embodiment: The comparison value, represented by the comparison value parameter included in the second macular parameter (parameter relating to mGCC), is calculated by applying a logarithmic operation to a number that includes the difference between the thickness of the first retinal ganglion cell complex in the third sub-area of ​​the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0260] A twelfth embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the first to eleventh embodiments: The computational model according to this embodiment is created by a computational model creation device. The computational model creation device may be an element of the ophthalmic apparatus according to this embodiment, or it may be a device separate from the ophthalmic apparatus according to this embodiment.

[0261] To create the computational model according to this embodiment, a development dataset, a first verification dataset, and a second verification dataset are prepared.

[0262] The development dataset is a dataset collected from the first population. That is, the development dataset is a dataset collected by applying one or more predetermined medical examinations (including ophthalmological examinations) to each subject included in the first population.

[0263] The first validation dataset is a dataset collected from a second population, distinct from the first population corresponding to the development dataset. Specifically, the first validation dataset is a dataset collected by applying one or more predetermined medical examinations (including ophthalmological examinations) to each subject in the second population. The one or more medical examinations performed for the first validation dataset and the one or more medical examinations performed for the development dataset may share at least one common examination, or they may not share any common examinations.

[0264] The second validation dataset is a dataset collected from a subgroup of individuals in the first population corresponding to the development dataset who were judged to have suspected glaucoma. This subgroup consists of individuals who belonged to the first population, which was the subject of the development dataset, and who were judged (diagnosed) to have suspected glaucoma. It is clear from its definition that this group is a part (subset) of the first population.

[0265] These development datasets, the first validation dataset, and the second validation dataset are input to a computational model creation device. The computational model creation device creates a computational model based on these datasets and a predetermined algorithm.

[0266] The algorithms used to construct the computational model may include machine learning algorithms. These machine learning techniques are typically supervised learning, but are not limited to it. In some examples, in machine learning for constructing computational models, any known technique, such as unsupervised learning, reinforcement learning, semi-supervised learning, transduction, or multitask learning, may be used alone or in combination with supervised learning, in addition to or instead of supervised learning.

[0267] The methods used to construct the computational model are not limited to the examples shown herein and may include any known methods, such as support vector machines, Bayesian classifiers, boosting, k-means algorithms, kernel density estimation, principal component analysis, independent component analysis, self-organizing maps, random forests, and generative adversarial networks (GANs).

[0268] A thirteenth embodiment of the ophthalmic apparatus according to the embodiment has the following non-limiting features in addition to the non-limiting features of the twelfth embodiment: The development dataset input to the computational model creation device includes fundus photographs acquired by digital fundus photography, fundus OCT data acquired by optical coherence tomography, and visual field data acquired by visual field testing.

[0269] The 14th exemplary aspect of the ophthalmic device according to the embodiment includes the following non-limiting features in addition to the non-limiting features of the 12th or 13th exemplary aspect. The first verification data set input to the calculation model creation device includes fundus OCT data acquired by optical coherence tomography. The calculation model creation device performs verification of the positive hit rate based on the input first verification data set.

[0270] The 15th exemplary aspect of the ophthalmic device according to the embodiment includes the following non-limiting features in addition to the non-limiting features in any of the 12th to 14th exemplary aspects. The second verification data set input to the calculation model creation device includes visual field data acquired by a visual field test. The calculation model creation device performs verification of sensitivity and specificity based on the input second verification data set.

[0271] Any of the 1st to 15th exemplary aspects of the ophthalmic device according to the embodiment can be at least partially combined with any matter described or suggested in the present disclosure.

[0272] The first exemplary aspect of the program according to the embodiment is a program for causing a computer including a processor and a memory to execute, and is configured to cause the processor to execute calculation model storage control, OCT data reception control, measurement value calculation processing, measurement value input processing, and calculation result providing processing.

[0273] In the calculation model storage control, the program causes the processor to execute a process of storing in the memory a calculation model including a linear combination of one or more papillary parameters representing the retinal tissue thickness in the optic nerve head area and a linear combination of one or more macular parameters representing the retinal tissue thickness in the macular area.

[0274] In the OCT data reception control, the program causes the processor to execute a process of receiving optical coherence tomography data (OCT data) of the fundus of the subject eye.

[0275] In the measurement value calculation process, the program causes the processor to execute a process of calculating a plurality of measurement values including one or more papillary measurement values respectively corresponding to one or more papillary parameters in the calculation model stored in the memory under the calculation model storage control, and one or more macular measurement values respectively corresponding to one or more macular parameters, based on the OCT data received in the OCT data reception process.

[0276] In the measurement value input process, the program causes the processor to execute a process of inputting the plurality of measurement values calculated in the measurement value calculation process into the calculation model stored in the memory under the calculation model storage control.

[0277] In the calculation result providing process, the program causes the processor to execute a process of providing the calculation result output from the calculation model corresponding to the input of the plurality of measurement values to the calculation model executed in the measurement value input process.

[0278] The first aspect example of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the first aspect example is recorded.

[0279] The second aspect example of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the first aspect example. One or more papillary parameters in the calculation model stored in the memory under the calculation model storage control include a papillary parameter representing the thickness of the retinal nerve fiber layer in the optic nerve papilla area.

[0280] The second aspect example of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the second aspect example is recorded.

[0281] A third embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the second embodiment: The computational model stored in memory by the computational model storage control includes a linear combination of two or more optic disc parameters. The two or more optic disc parameters include a first optic disc parameter (TSNITlower) that indicates the thickness value corresponding to the lower of the two peaks of a double hump exhibited by a TSNIT plot representing the distribution of retinal nerve fiber layer thickness in a predetermined circular region around the optic disc. Furthermore, the two or more optic disc parameters include a second optic disc parameter (cpRNFLqIn) that indicates the average value of the retinal nerve fiber layer thickness in an arc-shaped region in the circular region around the optic disc where the angle defined as zero in the temporal direction and positive in the direction of rotation upward from the temporal direction is in the range of 225 to 285 degrees.

[0282] A third embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the third embodiment is recorded.

[0283] A fourth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the third embodiment: The computation model stored in memory by the computation model storage control includes a linear combination of four optic disc parameters. These four optic disc parameters include a first optic disc parameter (TSNITlower) and a second optic disc parameter (cpRNFLqIn), as well as a third optic disc parameter (cpRNFLqS) and a fourth optic disc parameter (cpRNFLqIt). The third optic disc parameter (cpRNFLqS) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 45 to 135 degrees within the circular region around the optic nerve head. The fourth optic disc parameter (cpRNFLqIt) is an optic disc parameter that represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region corresponding to an angle in the range of 285 to 315 degrees within the circular region around the optic nerve head.

[0284] A fourth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the fourth embodiment is recorded.

[0285] A fifth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the first to fourth embodiments: The computational model stored in memory by the computational model storage control includes a linear combination of two or more macular parameters. These two or more macular parameters include a first macular parameter based on the thickness of the complex tissue of ganglion cells and the inner plexiform layer (mGCIPL) in the macular area, and a second macular parameter based on the thickness of the retinal ganglion cell complex (mGCC) in the macular area.

[0286] A fifth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the fifth embodiment is recorded.

[0287] A sixth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameter relating to mGCIPL) includes a layer thickness value parameter (mGCIPL_IT) that represents the thickness of the composite tissue (mGCIPL).

[0288] A sixth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the sixth embodiment is recorded.

[0289] A seventh embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the fifth embodiment: The first macular parameter (parameters relating to mGCIPL) includes comparison value parameters (log_mGCIPL_STvsIT, log_mGCIPL_ITvsIN;log_mGCIPL_STxvsITx;log_mGCIPL_SNyvsINy, log_mGCIPL_STyvsITy) that represent a comparison value between the first composite tissue thickness in a first sub-area of ​​the macular area and the second composite tissue thickness in a second sub-area.

[0290] A seventh embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the seventh embodiment is recorded.

[0291] An eighth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the seventh embodiment: The comparison value, represented by the comparison value parameter included in the first macular parameter (parameter relating to mGCIPL), is calculated by applying a logarithmic operation to a number that includes the difference between the first composite tissue thickness in the first sub-area of ​​the macular area and the second composite tissue thickness in the second sub-area.

[0292] An eighth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the eighth embodiment is recorded.

[0293] A ninth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the fifth to eighth embodiments: The second macular parameter (parameter relating to mGCC) includes a thickness value parameter (mGCC_IN;mGCC) representing the thickness of the retinal ganglion cell complex (mGCC).

[0294] A ninth embodiment of the recording medium is a computer-readable non-temporary recording medium on which a program according to the ninth embodiment is recorded.

[0295] The tenth aspect example of the program according to the embodiment includes the following non-limiting features in addition to the non-limiting features in any of the fifth to eighth aspect examples. The second macular parameter (parameter related to mGCC) includes a comparison value parameter (log_mGCC_ITxvsINx; log_mGCC_ITyvsINy) representing a comparison value between the thickness of the first retinal ganglion cell complex in the third sub-area of the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0296] The tenth aspect example of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the tenth aspect example is recorded.

[0297] The eleventh aspect example of the program according to the embodiment includes the following non-limiting features in addition to the non-limiting features of the tenth aspect example. The comparison value represented by the comparison value parameter included in the second macular parameter (parameter related to mGCC) is calculated by applying a logarithmic operation to the mantissa including the difference value between the thickness of the first retinal ganglion cell complex in the third sub-area of the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area.

[0298] The eleventh aspect example of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which the program according to the eleventh aspect example is recorded.

[0299] The twelfth aspect example of the program according to the embodiment includes the following non-limiting features in addition to the non-limiting features in any of the first to eleventh aspect examples. The program according to this aspect example is configured to further cause the processor to execute a calculation model creation process for creating a calculation model stored in a memory under calculation model storage control.

[0300] In the computational model creation process, the program instructs the processor to create a computational model based on a development dataset collected from a first population, a first validation dataset collected from a second population different from the first population, and a second validation dataset collected from a subpopulation of the first population that was determined to have suspected glaucoma.

[0301] A twelfth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the twelfth embodiment is recorded.

[0302] A thirteenth embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the twelfth embodiment: The development dataset input to the processor includes fundus photographs acquired by digital fundus photography, fundus OCT data acquired by optical coherence tomography, and visual field data acquired by visual field testing.

[0303] A thirteenth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the thirteenth embodiment is recorded.

[0304] A 14th embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of the 12th or 13th embodiment: A first validation dataset input to the processor includes fundus OCT data acquired by optical coherence tomography. The processor performs positive predictive value validation based on the input first validation dataset according to the program.

[0305] A 14th embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the 14th embodiment is recorded.

[0306] A 15th embodiment of the program according to the embodiment has the following non-limiting features in addition to the non-limiting features of any of the 12th to 14th embodiments: The second validation dataset input to the processor includes visual field data acquired by a visual field test. The processor performs sensitivity and specificity validation based on the input second validation dataset according to the program.

[0307] A fifteenth embodiment of the recording medium according to the embodiment is a computer-readable non-temporary recording medium on which a program according to the fifteenth embodiment is recorded.

[0308] Any of the first to fifteen embodiments of the program according to this embodiment can be combined, at least partially, with any of the matters described or suggested in this disclosure.

[0309] Any of the first to fifteen embodiments of the recording medium according to the embodiment can be combined, at least partially, with any of the matters described or suggested in this disclosure.

[0310] This disclosure presents several embodiments and some aspects thereof. These embodiments and aspects are merely non-limiting examples of the present invention. Therefore, any modifications (omissions, substitutions, additions, etc.) within the scope of the gist of the present invention can be made to the embodiments and aspects presented in this disclosure. [Explanation of Symbols]

[0311] 120 Ophthalmological equipment 121 processors 122 Control Unit 123 Measurement Value Calculation Unit 124 Risk Information Generation Unit 125 memory 126 Computational Models 127 OCT Data Acquisition Unit 128 User Interfaces

Claims

1. A storage unit for storing a pre-created calculation model including a linear combination of one or more optic disc parameters representing the thickness of retinal tissue in the optic disc area and a linear combination of one or more macular parameters representing the thickness of retinal tissue in the macular area, An OCT data acquisition unit that acquires optical coherence tomography (OCT) data from the fundus of the eye being examined, A measurement value calculation unit calculates a plurality of measurement values, including one or more nipple measurement values ​​corresponding to one or more nipple parameters and one or more macular measurement values ​​corresponding to one or more macular parameters, based on the OCT data. A risk information generation unit generates glaucoma risk information for the eye under examination based on the calculation model and the plurality of measured values. including, Ophthalmology equipment.

2. The one or more optic disc parameters in the calculation model stored in the memory unit include an optic disc parameter representing the thickness of the retinal nerve fiber layer in the optic disc area, An ophthalmic apparatus according to claim 1.

3. The aforementioned calculation model includes a linear combination of two or more nipple parameters, The two or more nipple parameters mentioned above are: A first optic disc parameter indicates the thickness value corresponding to the lower of the two peaks of the double hump shown in the TSNIT plot, which represents the distribution of retinal nerve fiber layer thickness in a predetermined circular region around the optic disc, and In the aforementioned circular region, a second optic disc parameter represents the average value of the retinal nerve fiber layer thickness in an arc-shaped region where the angle defined as zero in the temporal direction and the positive direction as the direction of rotation upward from the temporal direction corresponds to a range of 225 to 285 degrees. including, The ophthalmic apparatus according to claim 2.

4. The aforementioned calculation model includes a linear combination of four nipple parameters, The four nipple parameters mentioned above are: The first nipple parameter and, The second nipple parameter mentioned above, A third optic disc parameter representing the average value of the retinal nerve fiber layer thickness in the arc-shaped region where the angle corresponds to a range of 45 to 135 degrees within the circular region, A fourth optic disc parameter representing the average value of the retinal nerve fiber layer thickness in the arc-shaped region corresponding to the angle range of 285 to 315 degrees within the circular region, and including, The ophthalmic apparatus according to claim 3.

5. The aforementioned calculation model includes a linear combination of two or more macular parameters, The two or more macular parameters are, A first macular parameter based on the combined tissue thickness of ganglion cells and the inner plexiform layer in the macular area, A second macular parameter based on the thickness of the retinal ganglion cell complex in the aforementioned macular area, including, An ophthalmic device according to any one of claims 1 to 4.

6. The first macular parameter includes a layer thickness parameter representing the composite tissue thickness, The ophthalmic apparatus according to claim 5.

7. The first macular parameter includes a comparison parameter representing a comparison value between the first composite tissue thickness in a first sub-area of ​​the macular area and the second composite tissue thickness in a second sub-area. The ophthalmic apparatus according to claim 5.

8. The comparison value is calculated by applying a logarithmic operation to a number that includes the difference between the first composite structure thickness and the second composite structure thickness. The ophthalmic apparatus according to claim 7.

9. The second macular parameter includes a layer thickness parameter representing the thickness of the retinal ganglion cell complex, The ophthalmic apparatus according to claim 5.

10. The second macular parameter includes a comparison parameter representing a comparison between the thickness of the first retinal ganglion cell complex in the third sub-area of ​​the macular area and the thickness of the second retinal ganglion cell complex in the fourth sub-area. The ophthalmic apparatus according to claim 5.

11. The comparison value is calculated by applying a logarithmic operation to a number that includes the difference between the thickness of the first retinal ganglion cell complex and the thickness of the second retinal ganglion cell complex. An ophthalmic apparatus according to claim 10.

12. The calculation model is created based on a development dataset collected from a first population, a first validation dataset collected from a second population different from the first population, and a second validation dataset collected from a subpopulation of the first population that is suspected of having glaucoma. An ophthalmic device according to any one of claims 1 to 4.

13. A computational model creation device further comprising the development dataset, the first verification dataset, and the second verification dataset, which creates the computational model based on the development dataset, the first verification dataset, and the second verification dataset. The ophthalmic apparatus according to claim 12.

14. The aforementioned developed dataset includes fundus photographs acquired by digital fundus photography, fundus OCT data acquired by optical coherence tomography, and visual field data acquired by visual field testing. The ophthalmic apparatus according to claim 12.

15. The first validation dataset includes fundus OCT data acquired by optical coherence tomography and is used to validate the positive predictive value. The ophthalmic apparatus according to claim 12.

16. The second validation dataset includes visual field data obtained from visual field testing and is used to validate sensitivity and specificity. The ophthalmic apparatus according to claim 12.

17. A program to be executed on a computer, including a processor and memory, The aforementioned processor, Control for storing a computational model in the memory that includes a linear combination of one or more optic disc parameters representing the retinal tissue thickness in the optic disc area and a linear combination of one or more macular parameters representing the retinal tissue thickness in the macular area. Control for receiving optical coherence tomography (OCT) data from the fundus of the eye being examined, A process to calculate multiple measurements based on the OCT data, including one or more nipple measurements corresponding to one or more nipple parameters and one or more macular measurements corresponding to one or more macular parameters. The process of inputting the aforementioned multiple measurement values ​​into the calculation model, A process that provides calculation results output from the calculation model in response to the input of the multiple measurement values. To execute program.

18. A computer-readable non-temporary recording medium on which the program of claim 17 is recorded.

Images