Ocular axial length prediction program, device, method, ocular axial length prediction system, learning device, and trained model

Abstract

The present invention predicts progression of myopia. A program according to an embodiment of the present invention causes an ophthalmic medical device or a computer to execute: acquiring a measurement value of the ocular axial length of a subject and / or a measurement value of a high-order aberration of the subject; and inputting at least one measurement value of the ocular axial length of the subject and / or at least one measurement value of the high-order aberration of the subject to a trained model that predicts a value related to elongation of the ocular axial length from the ocular axial length and / or the high-order aberration, and causing the trained model to output the value related to the elongation of the ocular axial length of the subject.

Description

Axial length prediction program, device, method, axial length prediction system, learning device, and learned model

[0001] The present invention relates to an axial length prediction program, device, method, axial length prediction system, learning device, and learned model.

[0002] In recent years, the number of myopic people has been increasing rapidly worldwide. Since it is known that when myopia progresses to high myopia, the risk of developing blindness diseases such as glaucoma, retinal detachment, and myopic maculopathy increases, early preventive intervention is important, and the need for myopia progression prediction is increasing.

[0003] Japanese Patent No. 6247722

[0004] However, the factors that centrally contribute to myopia progression have not been identified, and it is considered that a large number of factors are involved complexly, and the prediction technology has not been established. For example, in Patent Document 1, only basic factors such as age are used in a polynomial regression model, so the differences in myopia progression between individuals cannot be reflected. Furthermore, in Patent Document 1, the refractive value is used as an index of myopia, but in children, the influence of accommodation is strong, and it is necessary to use an accommodative paralytic drug for accurate evaluation of the refractive value. Therefore, since the refractive value under non-accommodative paralysis cannot accurately evaluate the progression of myopia, and accommodative paralytic drugs are hardly used in daily medical practice, the myopia progression prediction method according to Patent Document 1 alone is insufficient.

[0005] An object of the present invention is to predict the progression of myopia.

[0006] A program according to an embodiment of the present invention causes an ophthalmic medical device or a computer to acquire a measured value of the axial length of a subject and / or a measured value of the higher-order aberration of the subject, and to input at least one measured value of the axial length of the subject and / or at least one measured value of the higher-order aberration of the subject into a learned model that predicts a value related to the elongation of the axial length from the axial length and / or the higher-order aberration, and outputs a value related to the elongation of the axial length of the subject.

[0007] According to the present invention, the progression of myopia can be predicted.

[0008] This figure shows the overall configuration (Configuration Example 1 and Configuration Example 2) according to one embodiment of the present invention. This figure shows the overall configuration (Configuration Example 3 and Configuration Example 4) according to one embodiment of the present invention. This is a functional configuration diagram of an ophthalmic medical device and computer according to one embodiment of the present invention. This is a flowchart of the axial length elongation prediction process according to one embodiment of the present invention. This is a sequence diagram of the axial length elongation prediction process according to one embodiment of the present invention. This figure is for explaining the structure of the dataset according to one embodiment of the present invention. This figure is for explaining the data used to generate the trained model according to one embodiment of the present invention. This figure shows the 95% confidence interval of the RMSE of each model according to one embodiment of the present invention. This figure shows the Brand-Altman plot of a two-year model using higher-order aberrations according to one embodiment of the present invention. This figure compares a one-year model using higher-order aberrations with a two-year model using higher-order aberrations according to one embodiment of the present invention. This figure compares a two-year model without higher-order aberrations with a two-year model using higher-order aberrations according to one embodiment of the present invention. This figure compares the prediction errors with and without higher-order aberrations according to one embodiment of the present invention. This figure is for explaining the PFI of the two-year model and the PFI of the one-year model according to one embodiment of the present invention. This figure is for explaining higher-order aberrations according to one embodiment of the present invention. This is a hardware configuration diagram of an ophthalmic medical device and computer according to one embodiment of the present invention.

[0009] Embodiments of the invention will be described below with reference to the drawings.

[0010] <Explanation of Terms> ・"Ophthalmic medical device" refers to any device used to measure the condition of the eye. For example, an ophthalmic medical device is an axial length measuring device (e.g., an optical axial length measuring device). Note that an axial length measuring device may also have a function to measure higher-order aberrations. ・"Axial length" is the length from the cornea to the retina. ・"Higher-order aberration" refers to the deviation from the ideal wavefront (leading or lagging the ideal wavefront) that occurs when multiple light rays from a single light source do not converge to a single point on the retina. In other words, higher-order aberration is the difference (wavefront aberration) between the ideal wavefront (the wavefront when multiple light rays from a single light source converge to a single point) and the wavefront actually measured. Higher-order aberrations will be explained in detail later.

[0011] <System Configuration> Figures 1 and 2 show the overall configuration according to one embodiment of the present invention.

[0012] In the <Configuration Example 1> in Figure 1, the ophthalmic medical device 10 predicts the elongation of the axial length of the subject (for example, a child) 11. In other words, the ophthalmic medical device 10 has a function to predict the elongation of the axial length. For example, the ophthalmic medical device 10 is an optical axial length measuring device.

[0013] In the configuration example 2 shown in Figure 1, a computer (server) 30 capable of sending and receiving data from multiple ophthalmic medical devices 10 predicts the elongation of the axial length of the subject (for example, a child) 11. In other words, the server 30 has a function to predict the elongation of the axial length.

[0014] In the configuration example 3 shown in Figure 2, the ophthalmic medical device 10 and a computer (terminal) 20 capable of sending and receiving data predict the elongation of the axial length of the subject 11. In other words, the terminal 20 has a function to predict the elongation of the axial length. For example, the terminal 20 can be a personal computer, tablet, smartphone, etc.

[0015] In the configuration example 4 shown in Figure 2, a computer (server) 30 capable of sending and receiving data with multiple computers (terminals) 20 predicts the elongation of the axial length of the subject 11. In other words, the server 30 has the function of predicting the elongation of the axial length. It is assumed that each terminal 20 is capable of sending and receiving data with each ophthalmic medical device 10.

[0016] <Functional Configuration> Figure 3 is a functional configuration diagram of an ophthalmic medical device 10 (when the ophthalmic medical device 10 predicts the elongation of the axial length of the eye), and computers 20 and 30 (when the computer (terminal) 20 predicts the elongation of the axial length, and when the computer (server) 30 predicts the elongation of the axial length) according to one embodiment of the present invention.

[0017] The ophthalmic medical device 10 and computers 20 and 30 may include a measurement value acquisition unit 101, a prediction unit 102, a predicted value presentation unit 103, and a trained model storage unit 104. The ophthalmic medical device 10 and computers 20 and 30 can function as the measurement value acquisition unit 101, the prediction unit 102, and the predicted value presentation unit 103 by executing a program. Each of these will be described below.

[0018] The measurement acquisition unit 101 acquires the measured axial length of the subject 11 and the measured higher-order aberrations of the subject 11. It is desirable that the axial length and higher-order aberrations be measured when the subject is not undergoing myopia treatment.

[0019] [Measurement of axial length and higher-order aberrations] Both the axial length and higher-order aberrations of the subject 11 may be measured by an ophthalmic medical device (for example, a device that can measure both axial length and higher-order aberrations) 10, or the axial length of the subject 11 may be measured by an ophthalmic medical device (for example, a device that can measure axial length but not higher-order aberrations) 10, and the higher-order aberrations of the subject 11 may be measured by a device other than the ophthalmic medical device 10 (for example, a device that can measure higher-order aberrations).

[0020] The prediction unit 102 predicts the elongation of the subject's axial length. Specifically, the prediction unit 102 inputs at least one measured value of the subject's axial length and at least one measured value of the subject's higher-order aberrations, acquired by the measurement value acquisition unit 101, into a trained model that predicts values ​​related to the elongation of the subject's axial length from the axial length and higher-order aberrations, and outputs values ​​related to the elongation of the subject's axial length.

[0021] Here, we will explain the inputs and outputs of the trained model. Note that the following inputs and outputs can be used in any combination.

[0022] [Input to the trained model (measured axial length and higher-order aberrations)] The input to the trained model includes the measured axial length and higher-order aberrations of subject 11. Factors related to myopia, such as grade level, gender, height, weight, BMI, close work time, outdoor activity time, and sleep time, may also be used as input to the trained model.

[0023] The input to the trained model may be a single set of measurements (e.g., measurements of axial length and higher-order aberrations for year n) or two sets of measurements (e.g., measurements of axial length and higher-order aberrations for year n, and measurements of axial length and higher-order aberrations for year n-1). While this specification uses an example based on years, it may also use semi-annual, monthly, weekly, etc.

[0024] [When using a single measurement] For example, the input to the trained model is "one measurement of axial length" and "one measurement of higher-order aberration" (i.e., one measurement is used). For example, the prediction unit 102 can predict the amount of elongation of axial length over one year (i.e., the difference between the axial length of subject 11 in year n and the axial length of subject 11 in year n+1) from a single year's measurement (specifically, the measurement of axial length and higher-order aberration of subject 11 in year n).

[0025] [When using two sets of measurements] For example, the input to the trained model is: "the measured axial length at the first time (e.g., year n-1)", "the measured axial length at the second time (e.g., year n) which is later than the first time", "the difference between the measured axial length at the first time and the measured axial length at the second time", "the ratio between the measured axial length at the first time and the measured axial length at the second time", "the measured higher-order aberration at the first time (e.g., year n-1)", "the measured higher-order aberration at the second time (e.g., year n) which is later than the first time", "the difference between the measured higher-order aberration at the first time and the measured higher-order aberration at the second time", and "the ratio between the measured higher-order aberration at the first time and the measured higher-order aberration at the second time" (i.e., two sets of measurements (the first and second times) are used).

[0026] For example, the input to a trained model is the change from a measured axial length at a first time (e.g., year n-1) to a measured axial length at a second time (e.g., year n) that is later than the first time, and the change (difference or ratio) from a measured higher-order aberration at the first time to a measured higher-order aberration at the second time.

[0027] For example, the input to a trained model is the difference between the axial length measured at the first time and the axial length measured at the second time, and the difference between the higher-order aberration measured at the first time and the higher-order aberration measured at the second time (i.e., two sets of measurements (from the first and second times) are used).

[0028] For example, the input to a trained model is the ratio of the axial length measured at the first time to the axial length measured at the second time, and the ratio of the higher-order aberration measured at the first time to the higher-order aberration measured at the second time (i.e., two sets of measurements (from the first and second times) are used).

[0029] For example, the input to a trained model is "a measurement of axial length taken at the first time (e.g., year n-1)", "a measurement of axial length taken at the second time after the first time (e.g., year n)", "a measurement of higher-order aberrations taken at the first time (e.g., year n-1)", and "a measurement of higher-order aberrations taken at the second time after the first time (e.g., year n)" (i.e., two sets of measurements (from the first and second times) are used).

[0030] [Output from the trained model (values ​​related to axial length elongation)] - For example, if the input to the trained model is one measurement (specifically, the measurement of axial length and higher-order aberrations in year n), the output of the trained model is the amount of axial length elongation of subject 11 over one year (i.e., the difference (or ratio) between the axial length in year n and the axial length in year n+1). Note that the value of the axial length in year n+1 may also be output. - For example, if the input to the trained model is two measurement values ​​(measurements from the first and second time; specifically, the measurement of axial length and higher-order aberrations in year n-1 and year n), the output of the trained model is the difference (amount of elongation) between the measurement of subject 11's axial length in year n and the future value of subject 11's axial length (e.g., year n+1; however, this may also be year n+2, year n+3, etc.). - For example, if the input to the trained model is two sets of measurements (measurements from the first and second time periods; specifically, measurements of axial length and higher-order aberrations from year n-1 and year n), the output of the trained model is the ratio of the measured axial length of subject 11 in year n to the future axial length of subject 11 (for example, year n+1; however, this could also be year n+2, year n+3, etc.). - For example, if the input to the trained model is two sets of measurements (measurements from the first and second time periods; specifically, measurements of axial length and higher-order aberrations from year n-1 and year n), the output of the trained model is the future axial length of subject 11 (for example, year n+1; however, this could also be year n+2, year n+3, etc.).

[0031] Furthermore, the interval between the first time when the axial length and higher-order aberrations are measured and the second time, which is after the first time, and the interval between the second time and the future that is the subject of prediction, may be the same period (for example, one year).

[0032] The prediction value presentation unit 103 presents the prediction results from the prediction unit 102 to the ophthalmic medical device 10 or computer (terminal) 20 (for example, by displaying text, images, or playing audio). The prediction value presentation unit 103 may also generate and present a risk score based on the predicted elongation of the axial length of the eye, and present treatment methods, etc., in order to indicate the risk level of myopia progression for the subject 11.

[0033] The trained model storage unit 104 stores a trained model that predicts values ​​related to axial length elongation from axial length and higher-order aberrations. The trained model is a machine learning model that outputs values ​​related to axial length elongation when axial length and higher-order aberrations are input. Note that the trained model may be updated (for example, new training data may be added to the training data used to generate the previous trained model and machine learning may be performed, or machine learning may be performed using only the new training data).

[0034] [Machine Learning] Here, we will explain the machine learning method. For example, an arbitrary learning device such as server 30 generates a trained model and stores the generated trained model in the trained model storage unit 104. The learning device uses training data (for example, values ​​related to the elongation of axial length of multiple children and higher-order aberrations of axial length are used as training data) to train the model so that when axial length and higher-order aberrations are input, a value related to the elongation of axial length is output. For example, in one embodiment of the present invention, a Gradient Boosting Decision Tree (GBDT) is used (for example, LightGBM). Therefore, it is a nonlinear regression and can handle multidimensional features (multidimensional data including higher-order aberrations and data on axial length over time).

[0035] In addition, the above describes an embodiment in which a value related to the elongation of the axial length is predicted from the axial length and higher-order aberrations, and an embodiment in which a model is trained that outputs a value related to the elongation of the axial length when the axial length and higher-order aberrations are input. However, the present invention can be applied when at least one of the axial length and higher-order aberrations is used (that is, only the axial length may be used, or only the higher-order aberrations may be used).

[0036] [When using only axial length] For example, the prediction unit 102 inputs at least one measurement value of the subject 11's axial length, acquired by the measurement value acquisition unit 101, into a trained model that predicts a value related to the elongation of the axial length from the axial length (which may be one measurement value or two measurement values), and outputs a value related to the elongation of the subject 11's axial length. The learning device learns a model that outputs a value related to the elongation of the axial length when the axial length (which may be one measurement value or two measurement values) is input, and generates a trained model.

[0037] [When using only higher-order aberrations] For example, the prediction unit 102 inputs at least one measurement value of the subject 11's higher-order aberrations acquired by the measurement value acquisition unit 101 to a trained model that predicts a value related to the elongation of the subject's axial length from higher-order aberrations (which may be one measurement value or two measurement values), and outputs a value related to the elongation of the subject 11's axial length. The learning device learns a model that outputs a value related to the elongation of the axial length when higher-order aberrations (which may be one measurement value or two measurement values) are input, and generates a trained model.

[0038] <Processing Method> Figure 4 is a flowchart of the process for predicting the elongation of the axial length of the eye according to one embodiment of the present invention. Figure 4 shows the processing method when the ophthalmic medical device 10 predicts the elongation of the axial length of the eye (<Configuration Example 1> in Figure 1).

[0039] In step 101 (S101), both the axial length and higher-order aberrations of the subject 11 are measured by the ophthalmic medical device 10, or the axial length of the subject 11 is measured by the ophthalmic medical device 10 and the higher-order aberrations of the subject 11 are measured by a device other than the ophthalmic medical device 10.

[0040] In step 102 (S102), the measurement value acquisition unit 101 of the ophthalmic medical device 10 acquires the measured values ​​of the axial length of the subject 11 and the higher-order aberrations of the subject 11, which were measured in S101.

[0041] In step 103 (S103), the prediction unit 102 of the ophthalmic medical device 10 inputs the measured values of the axial length and higher-order aberration obtained in S102 into the learned model stored in the learned model storage unit 104, and outputs a value related to the elongation of the axial length.

[0042] In step 104 (S104), the predicted value presentation unit 103 of the ophthalmic medical device 10 displays the result of the prediction in S103 on a display or the like of the ophthalmic medical device 10.

[0043] FIG. 5 is a sequence diagram of the prediction process of the elongation of the axial length according to an embodiment of the present invention. FIG. 5 shows a processing method when a computer (terminal 20 or server 30) predicts the elongation of the axial length (Configuration Example 2 in FIG. 1, Configuration Examples 3 and 4 in FIG. 2).

[0044] In step 201 (S201), both the axial length and higher-order aberration of the subject 11 are measured by the ophthalmic medical device 10, or the axial length of the subject 11 is measured by the ophthalmic medical device 10, and the higher-order aberration of the subject 11 is measured by a device other than the ophthalmic medical device 10.

[0045] In step 202 (S202), the ophthalmic medical device 10 transmits the measured value of the axial length of the subject 11 and the measured value of the higher-order aberration of the subject 11 measured in S201 to a computer (terminal 20 or server 30). Note that the ophthalmic medical device 10 may transmit the measured value of the axial length of the subject 11 and the measured value of the higher-order aberration of the subject 11 to the server 30 via the terminal 20 (Configuration Example 4 in FIG. 2).

[0046] In step 203 (S203), the measurement value acquisition unit 101 of the computer (terminal 20 or server 30) acquires the measured value of the axial length of the subject 11 and the measured value of the higher-order aberration of the subject 11 measured in S201.

[0047] In step 204 (S204), the prediction unit 102 of the computer (terminal 20 or server 30) inputs the measured values of the axial length and higher-order aberration obtained in S203 into the learned model stored in the learned model storage unit 104, and outputs a value related to the elongation of the axial length.

[0048] In step 205 (S205) and step 206 (S206), the prediction value presentation unit 103 of the computer (terminal 20 or server 30) transmits the prediction result in S204 to the ophthalmic medical device 10.

[0049] In step 207 (S207), the ophthalmic medical device 10 receives the prediction result in S204 and displays it on a display or the like.

[0050] Note that the prediction result in S204 may be displayed on the display or the like of the terminal 20.

[0051] <Effect> Thus, in one embodiment of the present invention, by using factors different from the conventional ones, particularly by using higher-order aberrations in learning, and by using the axial length of the eye measured using a high-precision optical axial length measuring device, the accuracy of myopia progression prediction can be improved.

[0052] In recent years, as myopia progression suppression treatment (axial length elongation suppression treatment), there have emerged many treatment methods with a high level of evidence, such as orthokeratology, low-concentration atropine eye drops, focus depth expansion type / multifocal contact lenses, outdoor activities, and supplements. When the axial length elongation can be predicted according to the present invention, it is possible to strengthen the treatment method or the like in terms of how much progression is expected in the future, and it is considered that there is an effect that it becomes easier to individualize the myopia progression suppression treatment.

[0053] <Example> Hereinafter, generation of a learned model which is an example of the present invention will be described. Note that data of elementary school students' health examinations, ophthalmic examinations, and answers to questionnaires regarding the eyes was used.

[0054] FIG. 6 is a diagram for explaining the structure of a data set according to one embodiment of the present invention.

[0055] In order to create feature amounts and objective variables for training the model, based on the records of each student for each year ([records for one year] in FIG. 6), records for three consecutive years ([records for three years] in FIG. 6) were created.

[0056] The 1-year model takes "features for the current year (i.e., axial length, height, etc.)" as input and outputs "the amount of elongation from the current year's axial length to the next year's axial length (i.e., the difference)". The 2-year model takes "features for the previous year and the current year, their respective values ​​(previous year's value and current year's value), the difference between them (previous year and current year), and the ratio between them (previous year and current year) (i.e., the previous year's value for axial length, the current year's value, the difference between them, and the ratio between them; the previous year's value for height, the current year's value, the difference between them, and the ratio between them, etc.)" as input and outputs "the amount of elongation from the current year's axial length to the next year's axial length (i.e., the difference)".

[0057] The following features were used as input to the model. Although data from the right eye was used, it is believed that similar predictions can be made using data from the left eye. • Grade level • Gender • Axial length (measured for the right eye) • Higher-order aberrations of the right eye (measured spherical aberration) • Higher-order aberrations of the right eye (measured S3) • Higher-order aberrations of the right eye (measured S4) • Higher-order aberrations of the right eye (measured S5) • Higher-order aberrations of the right eye (measured S6) • Higher-order aberrations of the right eye (Sqrt(S3^2 + S4^2 + S5^2 + S6^2)) • Height (measured) • Weight (measured) • BMI • Daily work time (total daily work time (questionnaire response)) • Daily work time (time spent watching TV (questionnaire response)) • Daily work time (time spent using a smartphone (questionnaire response)) • Daily work time (time spent using a computer (questionnaire response)) • Daily work time (time spent reading (questionnaire response)) • Time spent working at close range (distance between book and eyes (as per questionnaire response)) • Time spent outdoors (as per questionnaire response) • Time spent sleeping (as per questionnaire response)

[0058] Figure 7 is a diagram illustrating the data used to generate a trained model according to one embodiment of the present invention. (1) The original dataset had 1150 students and 3372 records (records for one year). 545 students from the original dataset were excluded because they did not have records for three consecutive years. (2) The three-year dataset (records for three consecutive years) had 605 students and 1075 records (records for three years). Of the three-year dataset, 13 records were excluded because they did not have axial length, 83 records were excluded because their spherical equivalent (SE) > 0 indicated myopia, and 447 records were excluded because they did not have higher-order aberration (HOA). (3) The curated three-year dataset had 372 students and 532 records (records for three years). (4) The three-year dataset after curation was divided into a training dataset (with 299 students and 427 records) and a test dataset (with 73 students and 105 records).

[0059] The breakdown of the three-year dataset after curation was as follows: • Number of students (372): 187 boys, 185 girls; 212 students in the third year of follow-up; 160 students in the fourth year of follow-up. • Number of records (532): 136 students in the second year; 110 students in the third year; 138 students in the fourth year; 148 students in the fifth year.

[0060] Using the above training data, we generated trained models using the LightGBM method, a type of Gradient Boosting Decision Tree (GBDT). The accuracy of each model is described below.

[0061] Figure 8 shows the 95% confidence intervals for the RMSE (root-mean-square error) of each model according to one embodiment of the present invention.

[0062] - "2Y" is a two-year model that does not use higher-order aberrations (HOA) as input (features). - "2Y (HOA)" is a two-year model that uses higher-order aberrations (HOA) as input (features). - "1Y" is a one-year model that does not use higher-order aberrations (HOA) as input (features). - "1Y (HOA)" is a one-year model that uses higher-order aberrations (HOA) as input (features).

[0063] As shown in Figure 8, the 2-year model using higher-order aberrations (HOA) as input (features) showed high prediction accuracy and an RMSE of 0.151.

[0064] Figure 9 shows a Brand-Altman plot of a two-year model using higher-order aberrations according to one embodiment of the present invention. The vertical axis shows the difference between the model's predicted value and the target value (actual measured value). The horizontal axis shows the average of the model's predicted value and the target value (actual measured value). The results of the Brand-Altman analysis show that only a small proportional error exists, and the systematic error is considered to be minute.

[0065] Figure 10 compares a one-year model and a two-year model using higher-order aberrations according to one embodiment of the present invention. The p-value of the Wilcoxon signed-rank test was 0.023. From Figure 10, it is considered that when comparing models using higher-order aberrations, the two-year model (i.e., using time-series data including the previous year's data) improves the squared error more than the one-year model.

[0066] Figure 11 compares a two-year model using higher-order aberrations and a two-year model without higher-order aberrations according to one embodiment of the present invention. When evaluated on samples with large prediction errors (squared error >= 0.2) rather than all samples, the p-value of the Wilcoxon signed-rank test was 0.033. From Figure 11, it can be seen that when comparing two-year models, the model using higher-order aberrations (HOA) improves the squared error more than the model without higher-order aberrations (HOA). Thus, since the model using higher-order aberrations (HOA) improves the squared error more than the model without higher-order aberrations (HOA) rather than all samples, even in samples with large prediction errors, higher-order aberrations can be said to be particularly useful in cases where prediction is difficult.

[0067] Figure 12 compares the prediction errors with and without higher-order aberrations according to one embodiment of the present invention. The vertical axis shows the prediction error, and the horizontal axis shows the number of data points. The model using higher-order aberrations (HOA) has a smaller squared error than the model without higher-order aberrations (HOA).

[0068] Figure 13 is a diagram illustrating the permutation feature importance (PFI) of a two-year model and a one-year model according to one embodiment of the present invention.

[0069] - "Demographic" refers to grade level and gender. - "Axial Length" refers to axial length of the eye. - "HOA" refers to higher-order aberrations (measured spherical aberration, measured S3, measured S4, measured S5, measured S6, and the square root of the sum of squares of S3-S6). - "Anthropometric" refers to height, weight, and BMI. - "Near-work" refers to near-work time (total near-work time, time spent watching TV, time spent using a smartphone, time spent using a computer, time spent reading, and distance between the eyes and the book). - "Activity" refers to time spent in outdoor activities and time spent sleeping.

[0070] As shown in Figure 13, the axial length of the eye, a feature belonging to the "Axial Length" group, is thought to be the main factor influencing the prediction. Higher-order aberrations belonging to the "HOA" group (measured values ​​of spherical aberration, S3, S4, S5, S6, and the square root of the sum of squares of S3-S6), height, weight, and BMI belonging to the "Anthropometric" group, and close-work time belonging to the "Near-work" group (total close-work time, time spent watching television, time spent using a smartphone, time spent using a computer, time spent reading, and distance between the eyes and the book) are also thought to influence the prediction.

[0071] Figure 14 is a diagram illustrating higher-order aberrations according to one embodiment of the present invention. Higher-order aberrations refer to the deviation from the ideal wavefront (leading or lagging the ideal wavefront) that occurs when multiple light rays from a single point light source do not converge to a single point on the retina. In other words, higher-order aberrations are the difference (wavefront aberration) between the ideal wavefront (the wavefront when multiple light rays from a single point light source converge to a single point) and the wavefront actually measured.

[0072] The wavefronts that are actually measured can be quantitatively expressed using the Zernike polynomial. Figure 14 shows the Zernike polynomial for the cases where n = 0, 1, and 2 (low-order aberrations) and for the cases where n = 3, 4, 5, and 6 (higher-order aberrations).

[0073] In one embodiment of the present invention, the following can be used as input to the model: • Higher-order aberration (measured value of spherical aberration) ... Spherical aberration in Figure 14 • Higher-order aberration (measured value of S3) ... n = 3 in Figure 14 • Higher-order aberration (measured value of S4) ... n = 4 in Figure 14 • Higher-order aberration (measured value of S5) ... n = 5 in Figure 14 • Higher-order aberration (measured value of S6) ... n = 6 in Figure 14 • Higher-order aberration (square root of the sum of squares of S3 - S6)

[0074] Furthermore, the present invention allows the use of higher-order aberrations measured by any measuring instrument using any measuring method (see, for example, "Measurement Method for Wavefront Aberration, Relationship between Cornea / Lens and Aberration, Simulation of Visual Perception," Yoko Hirohara, 2017 https: / / www.jstage.jst.go.jp / article / jorthoptic / 46 / 0 / 46_046K001 / _article / -char / ja / ). For example, higher-order aberrations may be measured using commercially available measuring instruments as described below.

[0075] [Measurement Methods and Equipment] Typical methods include the Hartmann-Shack wavefront sensor (representative models: KR-1W (Topcon), Wavescan (Abbott Medical Optics), WASCA Analyzer (Carl Zeiss Meditec)). Other methods include Laser Ray Tracing, which measures ocular aberrations by irradiating various points in the eye with a narrow beam of light and measuring the position of the reflected light (representative model: iTrace System (Tracey Technologies) - this is the measurement method and equipment used in the above experiment), a method that projects a grid pattern onto the fundus and observes the image on the retina (Tscherning aberration meter) (representative model: WaveLight Analyzer (WaveLight Technologie Inc.)), and a method that measures refractive distribution using retinoscopy at various points in the pupil (representative model: OPD-Scan (Nidek)). There are no reports comparing all of these simultaneously, but it is generally believed that similar results can be obtained using any of the measurement methods. Furthermore, when tracking changes over time in the same patient, it is desirable to continue measurements with the same equipment.

[0076] <Hardware Configuration> Figure 15 is a hardware configuration diagram of an ophthalmic medical device 10 and computers 20 and 30 according to one embodiment of the present invention. The ophthalmic medical device 10 and computers 20 and 30 may include a control unit 1001, a main memory unit 1002, an auxiliary memory unit 1003, an input unit 1004, an output unit 1005, and an interface unit 1006. Each of these will be described below.

[0077] Furthermore, the ophthalmic medical device 10 shall also be equipped with a function to measure the axial length of the eye using any method (for example, an optical method). The ophthalmic medical device 10 may also be equipped with a function to measure higher-order aberrations.

[0078] The control unit 1001 is a processor (for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc.) that executes various programs installed in the auxiliary storage unit 1003.

[0079] The main memory unit 1002 includes non-volatile memory (ROM (Read Only Memory)) and volatile memory (RAM (Random Access Memory)). The ROM stores various programs, data, etc., necessary for the control unit 1001 to execute various programs installed in the auxiliary memory unit 1003. The RAM provides a work area that is expanded when various programs installed in the auxiliary memory unit 1003 are executed by the control unit 1001.

[0080] The auxiliary storage unit 1003 is an auxiliary storage device that stores various programs and information used when various programs are executed.

[0081] The input unit 1004 is an input device that allows operators of the ophthalmic medical device 10 and computers 20 and 30 to input various instructions to the ophthalmic medical device 10 and computers 20 and 30.

[0082] The output unit 1005 is an output device that outputs the internal status of the ophthalmic medical device 10, computers 20 and 30, etc.

[0083] The interface unit 1006 is a communication device for connecting to a network and communicating with other devices.

[0084] Although embodiments of the present invention have been described in detail above, the present invention is not limited to the specific embodiments described above, and various modifications and changes are possible within the scope of the gist of the present invention as described in the claims.

[0085] This international application claims priority under Japanese Patent Application No. 2024-169545, filed on 27 September 2024, and the entire contents of Patent Application No. 2024-169545 are incorporated herein by reference.

[0086] 1. Axial length prediction system 10. Ophthalmic medical device 11. Target person 20. Computer (terminal) 30. Computer (server) 101. Measurement acquisition unit 102. Prediction unit 103. Prediction value presentation unit 104. Trained model storage unit 1001. Control unit 1002. Main memory unit 1003. Auxiliary memory unit 1004. Input unit 1005. Output unit 1006. Interface unit

Claims

1. A program that causes an ophthalmic medical device or computer to perform the following actions: acquire a measured value of the subject's axial length and / or a measured value of the subject's higher-order aberrations; and input at least one measured value of the subject's axial length and / or a measured value of the subject's higher-order aberrations into a trained model that predicts a value related to the elongation of the subject's axial length from the axial length and / or aberrations, causing the model to output a value related to the elongation of the subject's axial length.

2. The program according to claim 1, which causes the ophthalmic medical device or the computer to perform the following actions: acquire a measured value of the subject's axial length and a measured value of the subject's higher-order aberrations; and input at least one measured value of the subject's axial length and at least one measured value of the subject's higher-order aberrations into a trained model that predicts a value relating to the elongation of the subject's axial length from the axial length and higher-order aberrations, causing the model to output a value relating to the elongation of the subject's axial length.

3. The program according to claim 1, wherein the measured value of the subject's axial length is the measured value taken at the first time, the measured value taken at the second time which is later than the first time, the difference between the measured value taken at the first time and the measured value taken at the second time, and the ratio between the measured value taken at the first time and the measured value taken at the second time; and the measured value of the subject's higher-order aberration is the measured value taken at the first time, the measured value taken at the second time, the difference between the measured value taken at the first time and the measured value taken at the second time, and the ratio between the measured value taken at the first time and the measured value taken at the second time.

4. The program according to claim 1, wherein the measured value of the subject's axial length is one measurement value, and the measured value of the subject's higher-order aberrations is one measurement value.

5. The program according to claim 1, wherein the measured value of the subject's axial length shows a change from the measured value at the first time to the measured value at the second time, which is later than the first time, and the measured value of the subject's higher-order aberrations shows a change from the measured value at the first time to the measured value at the second time.

6. The program according to claim 5, wherein the change is the difference between the measurement value measured at the first time and the measurement value measured at the second time.

7. The program according to claim 5, wherein the change is the ratio of the measured value measured at the first time to the measured value measured at the second time.

8. The program according to claim 1, wherein the measured value of the subject's axial length is the measured value taken at the first time and the measured value taken at the second time, which is after the first time, and the measured value of the subject's higher-order aberration is the measured value taken at the first time and the measured value taken at the second time.

9. The program according to claim 3, wherein the value relating to the elongation of the subject's axial length is the difference between the measured value of the axial length measured at the second time and the future value of the subject's axial length.

10. The program according to claim 3, wherein the value relating to the elongation of the subject's axial length is the ratio of the measured value of the axial length measured at the second time to the value of the subject's future axial length.

11. The program according to any one of claims 3 to 8, wherein the value relating to the elongation of the subject's axial length is the value of the subject's future axial length.

12. A device comprising: a measurement value acquisition unit that acquires a measurement value of the subject's axial length and / or a measurement value of the subject's higher-order aberrations; and a prediction unit that inputs at least one measurement value of the subject's axial length and / or a measurement value of the subject's higher-order aberrations to a trained model that predicts a value related to the elongation of the subject's axial length from the axial length and / or higher-order aberrations, and outputs a value related to the elongation of the subject's axial length.

13. The apparatus according to claim 12, wherein the apparatus is an ophthalmic medical device or a computer.

14. A method performed by an ophthalmic medical device or computer, comprising: obtaining a measured value of the axial length of a subject and / or a measured value of the subject's higher-order aberrations; and inputting at least one measured value of the subject's axial length and / or a measured value of the subject's higher-order aberrations into a trained model that predicts a value relating to the elongation of the axial length of the subject from the axial length and / or higher-order aberrations, causing the model to output a value relating to the elongation of the subject's axial length.

15. An axial length prediction system including a server capable of sending and receiving data with an ophthalmic medical device or terminal, wherein the server comprises: a measurement value acquisition unit that acquires a measured value of the subject's axial length and / or a measured value of the subject's higher-order aberrations from the ophthalmic medical device or terminal; a prediction unit that inputs at least one measured value of the subject's axial length and / or a measured value of the subject's higher-order aberrations to a trained model that predicts a value related to the elongation of the axial length from the axial length and / or higher-order aberrations, and causes the model to output a value related to the elongation of the subject's axial length; and a prediction value presentation unit that presents the result of the prediction to the ophthalmic medical device or terminal.

16. A learning device that, when given axial length and / or higher-order aberrations as input, trains a model to output values ​​related to the elongation of the axial length, thereby generating a trained model.

17. A trained model that, when given axial length and / or higher-order aberrations as input, outputs a value related to axial length elongation.

Images