WO2025253124A1

WO2025253124A1 - Method for estimating eye growth

Info

Publication number: WO2025253124A1
Application number: PCT/GB2025/051234
Authority: WO
Inventors: Yuan SUN; David Hammond
Original assignee: CooperVision International Ltd
Current assignee: CooperVision International Ltd
Priority date: 2024-06-06
Filing date: 2025-06-05
Publication date: 2025-12-11
Anticipated expiration: 2026-12-06
Also published as: GB2700694A; GB202508901D0

Abstract

A computer-implemented method for estimating a future change in an axial length and/or a refractive error of an eye (104). The method comprises the following steps: receiving a dataset relating to a subject individual (102), the dataset comprising data defining a plurality of patient parameters (106) at a reference time point; operating a machine learning agent (134) to estimate a future change in the axial length and/or the refractive error of the eye (104) over a treatment period during which the eye (104) is treated with a myopia control treatment; and outputting the future change (148) estimated by the machine learning agent (134). The machine learning agent has been trained using a corpus of training data (150) comprising a plurality of training datasets, each relating to a different treated individual having an eye which has been treated using the myopia control treatment. Each training dataset comprises data defining each of the patient parameters (106) for the treated individual, and a change in an axial length and/or a refractive error of the eye of the treated individual. The patient parameters may include an age (110) of the individual.

Description

METHOD FOR ESTIMATING EYE GROWTH

Field

The present disclosure concerns a method for estimating eye growth. More particularly, but not exclusively, the disclosure concerns a computer-implemented method for estimating a future change in an axial length and/or refractive error of an eye which is subject to a myopia control treatment. The disclosure also concerns a method of treating myopia, a computer program product, and a computer system.

Background

Myopia (short-sightedness) affects a significant number of people including both children and adults. Myopia typically develops because the axial length of the eye grows to be longer than the focal length of the lens of the eye. Particularly in children and young adults, myopia may progress (i.e. worsen) with age. This is typically caused by the axial length of the eye increasing too quickly, relative to the speed at which the focal length of the eye increases as the optical components of the eye grow.

Figures la and lb provide schematic cross-sectional views through an eye with normal vision and an eye with myopia. Figure la shows the eye la with normal vision (i.e. an emmetropic eye). The optical components (principally the lens 3a) of the eye la cause the image 5a of a distant object 7a to be focussed on the retina 9a. This allows the image to be seen clearly. Figure lb shows the eye lb with myopia. The optical components (principally the lens 3b) of the eye lb cause the image 5b of a distant object 7b to be focussed in front of the retina 9b. The axial length of the eye lb is longer than the focal length. The light from the object 7b is therefore out of focus on arrival at the retina 9b. This causes the image to appear blurred. Figure 1c shows an eye 1c with hyperopia. The optical components (principally the lens 3 c) of the eye 1c cause the image 5c of a distant object 7c to be focussed behind the retina 9c. The axial length of the eye 1c is shorter than the focal length. The light from the object 7c is therefore out of focus on arrival at the retina 9c. This causes the image to again appear blurred.

Conventional lenses (e.g. spectacle lenses and contact lenses) for correcting the effect of myopia cause divergence of incoming light from distant objects before the light reaches the eye, so that the focus is shifted onto the retina. It has been found that conventional lenses which simply shift focus onto the retina do not generally treat the underlying cause of myopia, i.e. the imbalance between the axial length and the focal length of the eye, nor do they generally prevent myopia progression.

The term myopia control refers to methods which aim to slow or prevent the progression of myopia. A recent approach to myopia control involves providing contact lenses having regions that provide full correction of distance vision and regions that under-correct, or deliberately induce, myopic defocus. It has been suggested that this approach can prevent or slow down the development or progression of myopia in children and young people, whilst providing good distance vision. The regions that provide full-correction of distance vision are usually referred to as base-power regions or distance regions. The regions that provide under-correction or deliberately induce myopic defocus are usually referred to as add-power regions or myopic defocus regions, because the dioptric power is more positive, or less negative, than the power of the base-power regions.

The add-power regions are designed to focus incoming parallel light (i.e. light from a distance) within the eye in front of the retina (i.e. closer to the lens), whilst the base-power power regions are designed to focus light and form an image at the retina (i.e. further away from the lens). A surface (typically the anterior surface) of the add-power regions has a smaller radius of curvature than that of the base-power power regions in order to provide the more positive or less negative power to the eye.

Figures 2a to 2c show an example type of contact lens 10 for use in a myopia control treatment to slow the progression of myopia. The contact lens 10 comprises an optic zone 12 which covers the pupil. The optic zone 12 comprises a central region 14 which is surrounded by an annular region 16. The optic zone 12 provides the optical functionality of the contact lens 10. A peripheral zone 18 surrounds the optical zone 12. The peripheral zone 18 sits over the iris and provides mechanical functions, including increasing the size of the contact lens 10 to make it easier to handle, providing ballasting to prevent rotation of the contact lens 10 (in the case of a toric lens, for example), and providing a shaped region that improves comfort for the wearer.

The central region 14 has a base power which, in use, provides full correction of distance vision and focuses light onto the retina 20 of the eye of the user. The radius of curvature of the anterior (front) surface of the annular region 16 is smaller than the radius of curvature of the anterior surface of the central region 14. The annular region 16 thereby has a greater power than the power of the central region 14. The focal length of the annular region 16 is therefore shorter, and the focus lies on a focal surface 22 which is closer to the contact lens 10 and in front of the retina 20. The annular region 16 therefore under-corrects distance vision and deliberately induces myopic defocus.

The contact lens 10 shown in Figures 2a to 2c is a relatively straightforward example of a myopia control contact lens. Such lenses can have drawbacks. Light passing through the annular region 16 creates an unfocussed annular image 24 at the retina 20. It has been found that this can result in a wearer of the contact lens 10 seeing a ‘halo’ around focused distance images. Various other types of myopia control contact lenses exist, at least some of which may help to solve this problem. In some types of myopia control contact lenses, the central region and the annular region do not share the same optical axis. Instead, the annular region creates an annular region of focus around the optical axis of the central region. In other types of myopia control contact lenses, the annular region provides varying amounts of additional power over the base power of the central region, the power of the annular region varying with angular position. Commercially available contact lenses for myopia control include those available under the name MiSight® (CooperVision, Inc.).

For an eye care practitioner to determine appropriate treatment options for an individual displaying signs of myopia, it may be useful for the eye care practitioner to be able to predict how the eye might grow, and therefore how myopia might progress, in that individual. It may be useful for the eye care practitioner to predict myopia progression both with and without intervention using a myopia control treatment. There are several known methods for estimating future eye growth. At least some of these methods rely on a linear formula to calculate an estimated future growth in a predetermined time period.

EP3395233B1, for example, describes a method of calculating a future axial elongation ( AL) of an eye based on the eye’s refractive error change in the previous year (RECIPY) in Diopters, the age of the individual in years, and the current axial length of the eye in mm. The following linear formula is used:

AAL = a * RECIPY — b * Age + c * Axial Length — d

The values of a, b, c and d were calculated using data from a control group.

It has been found that methods of estimating the future growth of an eye which involve the use of such formula may have a limited accuracy and/or a relatively high margin of error. Furthermore, formula-based approaches can make it difficult to take into account categorical (non-numerical) variables, and it can be complicated to define non-linear relationships between the input parameters and the output.

“Axial length growth and the risk of developing myopia in European children” (Tideman et. al, Acta Ophthalmologica 2018, DOI: 10.1111/aos.13603) provides normative values for axial length which can be used for monitoring eye growth in children. This technique seems to rely on the assumption that the increase in axial length will typically continue along the same percentile curve as a function of time. Children with a rate of axial length growth higher than expected based on their percentile line can be identified for myopia control treatment.

These and other known models focus on predicting the growth of untreated eyes. To estimate the effect of a myopia control treatment, the predicted axial elongation of an untreated eye may be reduced by a predetermined percentage. However, estimating the effect of a myopia control treatment in this way may have limited accuracy. Accordingly, there is a limit to how useful such an estimate is to an eye care practitioner seeking to determine an appropriate treatment option for an individual displaying signs of myopia.

The present disclosure seeks to mitigate the above-mentioned problems. Alternatively or additionally, the present disclosure seeks to provide an improved method of estimating a future change in the axial length and/or the refractive error of an eye should the eye be treated with a myopia control treatment.

Summary

A first aspect of the disclosure provides a computer-implemented method for estimating a future change in an axial length and/or a refractive error of an eye; wherein one or more patient parameters are obtainable for an individual, the patient parameters including: an age of the individual; wherein the method comprises the following steps: receiving, at a processor, a dataset relating to a subject individual, the dataset comprising data defining each of the patient parameters for the subject individual at a reference time point; operating, using the processor, a machine learning agent to estimate a future change in the axial length and/or the refractive error of an eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment, wherein the future change in the axial length and/or the refractive error of the eye is estimated based on each of the patient parameters in the received dataset; outputting the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent; wherein the machine learning agent has been trained using a corpus of training data comprising a plurality of training datasets, each training dataset relating to a different treated individual having an eye which has been treated using the myopia control treatment over a treated period; wherein each training dataset comprises data defining: each of the patient parameters for the treated individual at a treated time point when the treated individual was treated, and a change in an axial length and/or a refractive error of the eye of the treated individual over the treated period.

A second aspect of the disclosure provides a method of treating myopia using the computer-implemented method according to the first aspect, the method comprising the following steps: obtaining the one or more patient parameters of the subject individual; performing the computer-implemented method to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment; and providing the myopia control treatment to the subject individual.

A third aspect of the disclosure provides computer program product being executable to perform the computer-implemented method according to the first aspect.

A fourth aspect of the disclosure provides a computer system comprising: memory storing instructions; and a processor coupled to the memory, said processor being configured to run the stored instructions to perform the computer-implemented method according to the first aspect.

Description of the Drawings

Embodiments of the present disclosure will now be described by way of example only with reference to the accompanying schematic drawings of which:

Figure la shows a cross sectional view through an eye with normal vision;

Figure lb shows a cross sectional view through an eye with myopia;

Figure 1c shows a cross sectional view through an eye with hyperopia;

Figure 2a shows a plan view of a contact lens for use in a myopia control treatment;

Figure 2b shows a perspective view of the contact lens for use in a myopia control treatment; Figure 2c shows a ray diagram for the contact lens for use in a myopia control treatment;

Figure 3 shows a subject individual and patient parameters obtainable from the subject individual;

Figure 4 shows a computer system in accordance with a first embodiment of the disclosure;

Figure 5 shows a machine learning agent in accordance with the first embodiment of the disclosure;

Figure 6 shows a method in accordance with a second embodiment of the disclosure;

Figure 7 shows a method in accordance with a third embodiment of the disclosure;

Figure 8 shows a computer system in accordance with a fourth embodiment of the disclosure;

Figure 9 shows a machine learning agent in accordance with the fourth embodiment of the disclosure; and

Figure 10 shows a method in accordance with a fifth embodiment of the disclosure;

Figure 11 is a graph indicating the importance of each variable to how a first example machine learning agent estimates a change in the axial length of an eye;

Figure 12 is a graph indicating the importance of each variable to how a second example machine learning agent estimates a change in the axial length of an eye;

Figure 13 shows a SHAP summary plot for samples used to test the second example machine learning agent; and

Figures 14 and 15 show SHAP force plots for two different samples in the data used to test the second example machine learning agent.

Detailed Description

The present disclosure provides, according to the first aspect, a computer-implemented method for estimating a future change in an axial length and/or a refractive error of an eye.

One or more patient parameters are obtainable for an individual. The individual may be any given individual referred to herein. The patient parameters may include an age of the individual. The method comprises a step of receiving, at a processor, a dataset relating to a subject individual, the dataset comprising data defining each of the patient parameters for the subject individual at a reference time point.

The method comprises a step of operating, using the processor, a machine learning agent to estimate a future change in the axial length and/or the refractive error of an eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment. The future change in the axial length and/or the refractive error of the eye is estimated based on each of the patient parameters in the received dataset.

The method comprises a step of outputting the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent.

The machine learning agent has been trained using a corpus of training data comprising a plurality of training datasets, each training dataset relating to a different treated individual having an eye which has been treated using the myopia control treatment over a treated period.

Each training dataset comprises data defining each of the patient parameters for the treated individual at a time when the treated individual was treated. Each training dataset also comprises data defining a change in an axial length and/or a refractive error of the eye of the treated individual over the treated period.

The method may allow the change in the axial length and/or the refractive error of the eye to be estimated more accurately than by prior art methods, particularly those which rely on the use of a linear equations to determine a future axial length of an untreated eye, the value of which is then reduced by a predetermined percentage to account for the effects of a myopia control treatment. The present method requires the machine learning agent to have been trained on training datasets comprising data for treated individuals. This may allow for more accurate modelling of the effect of myopia control treatment on the change in the axial length and/or the refractive error of the eye. Alternatively or additionally, this may allow more accurate modelling of how the change in the axial length and/or the refractive error of the eye correlates with the patient parameters.

The machine learning agent may comprise a tree-based model. It has been found that tree-based models may provide accurate estimates of the future change in the axial length and/or the refractive error. Tree-based models typically comprise at least one tree structure (decision tree) which defines the relationship between the input(s) and an output. The branches in the decision tree may have different weightings which have been calculated using the training data. A decision tree can be used for classification to predict a category (wherein it is referred to as a “classifier”), or regression to predict a continuous numeric value (wherein it is referred to as “regressor”). The machine learning agent may comprise a tree-based regressor.

Various tree-based models are known to the skilled person. The tree-based model may be a decision tree model, a random forest model, a gradient-boost decision tree (GBDT) model (such as Extreme Gradient Boosting (XGBoost)), or another advanced model based on decision tree(s).

As will be understood by a person skilled in the art, random forest uses a technique called bagging to build a plurality of full decision trees in parallel from random bootstrap samples of a data set. The final prediction is an average of all of the decision tree predictions. GBDT iteratively trains a plurality of shallow decision trees, with each iteration using the error residuals of the previous model to fit the next model. The final prediction is a weighted sum of all of the tree predictions. XGBoost is an advanced tree-based model which builds trees in parallel, rather than sequentially like GBDT, and evaluates the quality of splits at every possible split in a training set. In embodiments of the disclosure, XGBoost may be the preferred tree-based model.

Example tree-based models available in Python programming language include: extremely randomized tree regressor (ExtraTreeRegressor), extreme Gradient Boosting regressor (XGBRegressor), random forest regressor (RandomForestRegressor), and decision tree regressor (DecisionTreeRegressor).

The performance of a plurality of different models may be evaluated using one or more evaluation metrics, such as MSE (Mean squared error) and R-squared (Coefficient of determination), SMAPE (Symmetric mean absolute percentage error), and computation time. The models may be evaluated using a corpus of test data. A model may be selected for the machine learning agent based on the evaluation. It may be that the preferred model depends on the data defined by the training datasets.

In alternative embodiments, the machine learning agent may comprise another category of model, which is not tree-based.

The machine learning agent may be re-trained with more data, i.e. with a corpus of training data comprising additional training datasets, when more training data becomes available, for example following conclusion of a new clinical study. Re-training may improve the generalization ability and/or the accuracy of the estimates generated by the machine learning agent. Generalization ability refers to the model's ability to adapt properly to new, previously unseen data, drawn from the same distribution as the one used to create the model. After retraining, hyperparameters and parameters of the model may be different.

The myopia control treatment may comprise treatment using an ophthalmic lens. The ophthalmic lens may be a contact lens, a spectacle lens or an intraocular lens. The ophthalmic lens may comprise a base-power region for correcting distance vision. The ophthalmic lens may comprise an add-power region for inducing a myopic defocus. The base-power region may form a central region of the lens. The add-power region may form an annular region around the central region.

The ophthalmic lens may comprise features that reduce the contrast of images viewed through the lens. The features that reduce the contrast of images viewed through the lens may increase defocus in a region of the lens. The features that reduce the contrast of images viewed through the lens may be scattering elements and/or scattering centres.

The myopia control treatment may comprise pharmacological treatment, for example using atropine or pirenzepine. The myopia control treatment may comprise light therapy, for example light therapy using red light.

It has been found that the change in the axial length and/or refractive error of the eye may be affected by an optical treatment previously used by the individual. The optical treatment type previously used may be an optical treatment type used in a period immediately before the reference time point / treatment period. It has been found that the change in the axial length and/or refractive error of the eye may also be affected by the length of time that the individual previously used the optical treatment, e.g. in the period immediately before the reference time point / treatment period.

Accordingly, the patient parameters may include an optical treatment type previously used by the individual (prior to the reference time point / treatment period). The patient parameters may include a length of time (e.g. in days) the individual had previously used the optical treatment type (prior to the reference time point / treatment period).

The optical treatment type previously used may be selected from: a single vision lens, the myopia control treatment, and another myopia control treatment (which is different from the myopia control treatment). It has been found that the change in the axial length and/or refractive error of the eye may be affected by: a biological sex of the individual, an ethnicity of the individual, an ethnicity of the mother of the individual, an ethnicity of the father of the individual, an axial length of the eye of the individual and/or a refractive error of the eye of the individual.

Accordingly, the patient parameters may include one or more of: a biological sex of the individual, an ethnicity of the individual, an ethnicity of the mother of the individual, an ethnicity of the father of the individual, an axial length of the eye of the individual and/or a refractive error of the eye of the individual.

It is thought that environmental factors may also have an effect on myopia progression. Those factors may be tied to the geographic area in which the person has grown up and/or is presently located. The patient parameters may therefore also include one or more of: a geographic area of residence of the individual, a country of residence of the individual (e.g. the USA, the UK, Australia, etc), a continent of residence of the individual (e.g. North America, Europe, Oceana, etc.), a selection of whether the individual is resident in an urban or rural environment (e.g. as defined by population density or self-identification by the individual).

Thus, it may be the case that the future change in the axial length and/or the refractive error of the eye is estimated based on, for the subject individual at a reference time point, one or more of the following parameters: an age of the subject individual, an optical treatment type previously used by the subject individual (prior to the reference time point / treatment period), a length of time the subject individual had previously used the optical treatment type (prior to the reference time point / treatment period), a biological sex of the subject individual, an ethnicity of the subject individual, an ethnicity of the mother of the subject individual, an ethnicity of the father of the subject individual, an axial length and/or a refractive error of the eye of the subject individual, a geographic area of residence of the individual, a country of residence of the individual, a continent of residence of the individual, and/or a selection of whether the individual is resident in an urban or rural environment.

Similarly, it may be the case that each training dataset comprises data defining, for the respective treated individual at the respective treated time point, one or more of the following parameters: an age of the treated individual, an optical treatment type previously used by the treated individual (prior to the treated time point / treated period), a length of time the individual had previously used the optical treatment type (prior to the treated time point / treated period), a biological sex of the treated individual, an ethnicity of the treated individual, an ethnicity of the mother of the treated individual, an ethnicity of the father of the treated individual, an axial length and/or a refractive error of the eye of the treated individual, a geographic area of residence of the individual, a country of residence of the individual, a continent of residence of the individual, and/or a selection of whether the individual is resident in an urban or rural environment.

The categorical variables mentioned above (e.g. biological sex and ethnicity) may be difficult to take into account using prior art methods, particularly those which rely on the use of a linear equation to determine the future axial length of the eye.

As used herein, ethnicity refers to the ethnic group an individual is biologically (i.e. genetically) associated with (or closest associated with). It may be that ethnicity is determined by testing to detect a genetic characteristic, however, it may be that ethnicity is simply determined by self-identification. Food and Drug Administration (FDA) of the US Government Office guidance, based on the Office of Management and Budget (OMB) Statistical Policy Directive No. 15, recommends a self-identification approach. The skilled person will appreciate that different embodiments may include different selections (e.g. lists) of ethnic groups from which ethnicity may be chosen. In one example, the ethnic groups may include the following minimum choices as per FDA guidance: American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White. In another example, the ethnic groups may be based on those used in census data, for example, the 2021 Census of England and Wales. By way of example, the ethnic groups may include one or more of the following: Asian (which may include Indian, Pakistani, Bangladeshi, Chinese and/or any other Asian background); Black (which may include Caribbean, African and/or any other Black background); White (which may include Caucasian (White European) and/or any other White background); Arab (which may include Middle Eastern and/or any other Arab background); Mixed or multiple ethnic groups (which may include Asian and White, Asian and Black, Asian and Arab, (and so on... ), and/or any other Mixed or multiple ethnic background); Other (which may include any other ethnic background not previously listed).

It may be that, in the received dataset, the data defining each of the patient parameters is data from which the patient parameters for the subject individual at the reference time point can be calculated. For example, the data in the received dataset could comprise the date of birth of the subject individual and the date of the reference time point, the age of the subject individual being calculable using these dates. The same may be the case in relation to the training datasets.

The reference time point may be the time at which one or more patient parameters are obtained, for example by an eye care practitioner such as an optometrist. The reference time point may be the time at which a measurement of the axial length and/or the refractive error of the eye is taken, for example by the eye care practitioner. The treatment period is in the future relative to the reference time point. The reference time point may define the start of the treatment period. An end time point may define the end of the treatment period. The treatment period may, for example, be 12 months. In other examples, the treatment period may be 1 month, 3 months, 6 months, 18 months, or 24 months or more than 24 months. The future change may be the change in the axial length and/or the refractive error of the eye, relative to the axial length and/or the refractive error of the eye at the reference time point.

The future change in the axial length of the eye may be estimated based on the length of the treatment period. The length of the treatment period may be specified by the user. The method may comprise a step of receiving data defining the length of the treatment period.

The treatment period may comprise a time period which is in the future relative to when the present method is performed. The method may therefore be used to make a prediction/forecast of what the axial length and/or the refractive error of the eye might be at the end of the (proposed) treatment period.

Alternatively, the subject individual may have previously been treated with the myopia control treatment for the treatment period. The treatment period may thereby comprise a time period which is in the past relative to when the present method is performed. The reference time point may thus be the beginning of the past treatment period. The method may comprise a step of comparing (i) the estimated future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period with (ii) a measured change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period. The method may therefore be used to determine the extent to which the subject individual has previously responded to the myopia control treatment.

The method may comprise a step of outputting an indication, based on the comparison, of the extent to which the subject individual has responded to the myopia control treatment over the treatment period. The step of comparing may comprise calculating a difference between the estimated future change and the measured change. The method may comprise a step of comparing the difference to a threshold value. The individual may be determined to be a non-responder to the myopia control treatment if the difference exceeds the threshold value. There may be a step of outputting an indication of whether the individual is determined to be a non-responder.

The method provides an estimate of the magnitude of the future change in the axial length and/or the refractive error of the eye over the treatment period. With knowledge of the axial length and/or the refractive error of the eye at the start of the treatment period, it is possible to determine an estimate of the axial length and/or the refractive error of the eye at the end of the treatment period. It will therefore be appreciated that the comparison could be performed using estimated and measured values of the magnitude of the change in the axial length and/or the refractive error of the eye, or using estimated and measured values of the axial length and/or the refractive error of the eye at the end of the treatment period.

For each training dataset, the treated time point may be the beginning of the treated period, a time during the treated period, or the end of the treated period. Preferably, the treated time point is the same, relative to the treated period, for each of (e.g. all of) the training datasets used to train the machine learning agent. For example, it may be that each of the training datasets comprise patient parameters related to the beginning (or end) of the treated period. For example, each of the training datasets may comprise the age of the individual at the beginning (or end) of the treated period. Similarly, each of the training datasets may comprise the axial length and/or the refractive error of the eye at the beginning (or end) of the treated period. However, this does not necessarily need to be the case.

The training datasets may be parsed (e.g. prior to training the machine learning agent) to make the training datasets consistent with each other. For example, the training datasets may be parsed so that each of (e.g. all of) the training datasets comprise patient parameters related to the beginning (or end) of the treated period. For example, the training datasets may be parsed so that each of the training datasets contain the age of the individual at the beginning (or end) of the treated period. Similarly, the training datasets may be parsed so that each of the training datasets contain the axial length and/or the refractive error of the eye at the beginning (or end) of the treated period. The change in the axial length and/or the refractive error of the eye of the treated individual over the treated period may be defined using one or more of: a starting axial length at the beginning of the treated period, a finishing axial length at the end of the treated period, and/or a magnitude of the change over the treated period.

Each training dataset may define a length of the treated period. For each training dataset, the treated period may, for example, be 1 month, 3 months, 6 months, 12 months, 18 months, or 24 months or more than 24 months. The treated period may have a length corresponding to the length of the treatment period. Preferably, the length of the treated period is the same for each of the training datasets used to train the machine learning agent. However, this does not necessarily need to be the case. The training datasets may be parsed (e.g. through normalisation, extrapolation, etc.) so that each of the training datasets define a change in the axial length and/or the refractive error of the eye of the treated individual over the same length of treated period.

The above-mentioned myopia control treatment may be a first myopia control treatment. The above-mentioned machine learning agent may be a first machine learning agent. The above- mentioned corpus of training data may be a corpus of first training data. The above-mentioned training datasets may be first training datasets, each first training dataset relating to a different first treated individual having an eye which has been treated using the first myopia control treatment over a treated period.

The method may comprise a step of operating (e.g. using the processor) a machine learning agent (e.g. the first or a second machine learning agent) to estimate a future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period during which the eye is treated with a second myopia control treatment. The future change in the axial length and/or refractive error of the eye treated with the second myopia control treatment may be estimated based on each of the patient parameters in the received dataset.

The method may comprise a step of outputting the future change in the axial length and/or refractive error of the eye treated by the second myopia control treatment estimated by the (e.g. first or second) machine learning agent.

The (e.g. first or second) machine learning agent may have been trained using a corpus of second training data comprising a plurality of second training datasets. Each second training dataset may relate to a different second treated individual having an eye which has been treated using the second myopia control treatment over a treated period. Each second training dataset may comprise data defining each of the patient parameters for the second treated individual at a treated time when the second treated individual was treated. Each second training dataset may comprise data defining a value of a change in an axial length and/or a refractive error of the eye of the second treated individual over the treated period.

The method may comprise a step of comparing (i) the estimated future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period during which the eye is treated with the first myopia control treatment and (ii) the estimated future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period during which the eye is treated with the second myopia control treatment. The step of comparing may comprise calculating a difference between the two estimated future changes. The method may thereby be used to determine which of the first and second myopia control treatment the subject individual will respond to best. The method may comprise a step of a result of the comparison.

The method may comprise a step of selecting a myopia control treatment for the subject individual based on the estimated future change in the axial length and/or the refractive error of the eye. The method may comprise a step of outputting an indication of the selected myopia control treatment.

Selecting a myopia control treatment for the subject individual may comprise selecting between the first myopia control treatment and the second myopia control treatment based on the comparison between: the estimated future change should the eye be treated with the first myopia control treatment, and the estimated future change should the eye be treated with the second myopia control treatment. The selected myopia control treatment may be the myopia control treatment having the lowest estimated future change in the axial length and/or the refractive error of the eye.

It may be that the first and second myopia control treatments are different types of treatment. For example, the treatments could use different types of contact lens. It may be that the first and second myopia control treatments are the same type of treatment at different strengths. For example, the treatments could both use the same type of contact lens, but with each contact lens having a different amount of base power, add power and/or myopic defocus. Selecting a myopia control treatment may comprise selecting a type and/or a power of a myopia control contact lens for the subject individual.

The step of selecting may be performed using the processor or another processor. The step of selecting may be performed by a selector. The selector may perform the above-mentioned comparison between the estimated future changes. The selector may be a software module. The selector may be an automatic selector. The selector may be a user, such as the eye care practitioner. The above-mentioned steps of outputting may comprise outputting to the selector.

In embodiments, there may be more than two different myopia control treatments to select between, the machine learning agent being trained on training data comprising training datasets corresponding to each of the different myopia control treatments.

The method may comprise a step of operating (e.g. using the processor) a machine learning agent (e.g. the first, the second or a third machine learning agent) to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a time period (e.g. the treatment period) should the eye be untreated for the time period (e.g. the treatment period). For example, should the eye not be subject to a myopia control treatment, or not be subject to any optical treatment method. The future change in the axial length and/or the refractive error of the untreated eye may be estimated based on each of the patient parameters in the received dataset.

The method may comprise a step of outputting the future change in the axial length and/or the refractive error of the untreated eye estimated by the (e.g. first, second or third) machine learning agent.

The (e.g. first, second or third) machine learning agent may have been trained using a corpus of third training data comprising a plurality of third training datasets. Each third training dataset may relate to a different untreated individual having an eye which has been untreated over an untreated period. Each third training dataset may comprise data defining each of the patient parameters for the untreated individual at an untreated time point when the untreated individual was untreated. Each third training dataset may comprise data defining a change in an axial length and/or a refractive error of the eye of the untreated individual over the untreated period. It might be the case that some of the untreated individuals have never been treated with an optical treatment. It might be the case that some of the untreated individuals have, at some point outside the untreated period, been treated with an optical treatment.

The method may comprise a step of comparing (i) the estimated future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period during which the eye is treated with the (e.g. first or second) myopia control treatment and (ii) the estimated future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of the treatment period should the eye be untreated for the treatment period. The step of comparing may comprise calculating a difference between the two estimated future changes. The method may comprise a step of comparing the difference to a threshold value. The method may comprise a step of outputting a result of the comparison.

The method may comprise a step of determining (e.g. using the processor) whether the (e.g. first, second or selected) myopia control treatment is a suitable myopia control treatment for the subject individual based on the future change in the axial length and/or the refractive error of the eye estimated by the (e.g. first or second) machine learning agent. The method may comprise a step of outputting an indication of whether the (e.g. first, second or selected) myopia control treatment is determined to be a suitable myopia control treatment for the subject individual.

Determining whether the (e.g. first, second or selected) myopia control treatment is a suitable myopia control treatment may be based on the comparison between: the estimated future change should the eye be treated with (e.g. first, second or selected) myopia control treatment, and the estimated future change should the eye be untreated. The (e.g. first, second or selected) myopia control treatment may be determined to be suitable if the difference exceeds the threshold value. The (e.g. first, second or selected) myopia control treatment may be determined to be unsuitable if the difference does not exceed the threshold value.

The estimated future change in the axial length and/or the refractive error of the untreated eye may be output to the selector. The selector may perform the step of determining whether the myopia control treatment is a suitable myopia control treatment. The selector may perform the comparison between the treated and untreated estimated future changes.

The selected myopia control treatment may be the myopia control treatment having the lowest estimated future change in the axial length and/or the refractive error, relative to the estimated future change should the eye be untreated. The method may comprise a step of obtaining a measurement of the axial length and/or the refractive error of the eye of the subject individual. The measurement may be obtained after providing the (e.g. first, second or selected) myopia control treatment to the subject individual for a period of time. The period of time may be at least 1 month, or at least 3 months, or at least 6 months, or at least one year. The method may comprise a step of storing the measurement in a memory of a computer system. The method may comprise a step of inputting the measurement into a patient record. The patient record may be an electronic patient record stored in the memory of the computer system.

The method may comprise monitoring and/or determining (e.g. using the or a processor) the efficacy of the myopia control treatment using the measurement. For example, the measurement may be used to determine a measured change in the axial length and/or the refractive error of the eye over the period of time. The measured change may be compared with an estimated future change over the same period of time determined by the machine learning agent.

It may be that the measurement is used to augment the corpus of training data. There may be a step of generating a training dataset using the measurement. It may be that the machine learning agent is re-trained using the corpus of training data which has been augmented using the training dataset generated using the measurement. This may allow the machine learning agent to be updated/improved on an ongoing basis.

It may, however, be that the properties of the machine learning agent are fixed after approval by a regulatory body, and an updated/improved version is only deployed to users after further approval by the regulatory body. This may be appropriate where the machine learning agent, and/or the computer program product to which it belongs, is regarded as Software as a Medical Device (SaMD) for regulatory purposes.

The step of taking the measurement, and optionally also the step of inputting the measurement into the patient record, may be repeated one or more times after further period(s) of time, for example at regular intervals. Those further measurements may also be used to monitor and/or determine the efficacy of the myopia control treatment, and/or used to generate one or more further training datasets used to augment the corpus of training data and/or retrain the machine learning agent. The (or each or any other) measurement of the axial length and/or the refractive error of the eye of the subject individual may be obtained using an electronic measurement device communicatively coupled to the processor.

It has been found that certain patients experience a rebound effect after a myopia control treatment is ceased, whereby myopia progresses faster than it would have done via natural progression should the eye have been untreated. This may result in the individual losing (in whole or in part) the effects of the myopia control treatment. The phenomenon has been observed for individuals treated with 1% and 0.5% atropine, and may also occur with lower concentrations of atropine. Rebound accelerated eye growth may also be a feature of red light therapy.

The method may comprise a step of operating (e.g. using the processor) a machine learning agent (e.g. the first, the second, the third or a fourth machine learning agent) to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a post-treatment period during which the eye is untreated. The posttreatment period may begin after, optionally immediately after, the treatment period. The future change in the axial length and/or the refractive error of the eye may be estimated based on each of the patient parameters in the received dataset.

The method may comprise a step of outputting the future change in the axial length and/or the refractive error of the eye estimated by the (e.g. first, second, third or fourth) machine learning agent.

It may be that the (e.g. first, second or third) machine learning agent, having been trained using the corpus of third training data relating to untreated individuals, is used to estimate the future change over the duration of the post-treatment period (which is also an untreated period).

Alternatively, the (e.g. first, second, third or fourth) machine learning agent may have been trained using a corpus of fourth training data comprising a plurality of fourth training datasets. Each fourth training dataset may relate to a different treated individual having an eye which has been treated over a treated period and then untreated for a post-treatment period. Each fourth training dataset may comprise data defining each of the patient parameters for the treated individual at a post-treatment time point when the treated individual was untreated after finishing the myopia control treatment. Each fourth training dataset may comprise data defining a change in an axial length and/or a refractive error of the eye of the treated individual over the posttreatment period.

The method may comprise a step of comparing (i) the estimated future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of the post-treatment period during which the eye is untreated to (ii) a measured change in the axial length and/or refractive error of the eye of the subject individual over the duration of the posttreatment period.

The step of comparing may comprise calculating a difference between the estimated future change and the measured change. The method may comprise a step of comparing the difference to a threshold value. The individual may be determined to be experiencing a rebound effect if the measured change is greater than the estimated future change, or greater than the estimated future change by an amount greater than the threshold value. The method may comprise a step of outputting a result of the comparison. The method may comprise a step of outputting an indication of whether the individual is determined to be experiencing a rebound effect.

To determine whether a rebound effect has been experienced, it is not necessary to estimate the future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of the treatment period itself. Accordingly, there may be aspects and embodiments of the disclosure which do not comprise the step of operating the machine learning agent to estimate the future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of the treatment period. There may be aspects and embodiments where the only estimate is of the future change in the axial length and/or the refractive error of an untreated eye.

The second myopia control treatment may have any of the features set out above in relation to the first myopia control treatment. The second, third and/or fourth machine learning agents may have any of the features set out above in relation to the first machine learning agent. The first, second, third and/or fourth machine learning agents may each be substantially distinct software elements wherein each machine learning agent can be operated independently, or they may each be sub-agents which form a part of an over-arching machine learning agent.

The corpus of second training data, the corpus of third training data and/or the corpus of fourth training data may have any of the features set out above in relation to the corpus of first training data. For example, the second, third and/or fourth training datasets may have any of the features set out above in relation to the first training datasets.

For example, in relation to the third training datasets, the untreated time point may be the beginning of the untreated period, a time during the untreated period, or the end of the untreated period. Preferably, untreated time point is the same, relative to the untreated period, for each of the third training datasets. For example, it may be that each of the third training datasets comprise patient parameters related to the beginning (or end) of the untreated period. The third training datasets could be parsed as set out above.

The change in the axial length and/or the refractive error of the eye of the untreated individual over the untreated period may be defined using one or more of: a starting axial length at the beginning of the untreated period, a finishing axial length at the end of the untreated period, and a magnitude of the change over the untreated period.

Similarly, it may be the case that each third training dataset comprises data defining, for the respective untreated individual at the respective untreated time point, one or more of the following parameters: an age of the untreated individual, an optical treatment type previously used by the untreated individual (prior to the untreated time point / untreated period), a length of time the individual had previously used the optical treatment type (prior to the untreated time point / untreated period), a biological sex of the untreated individual, an ethnicity of the untreated individual, an ethnicity of the mother of the untreated individual, an ethnicity of the father of the untreated individual, an axial length and/or a refractive error of the eye of the untreated individual, a geographic area of residence of the individual, a country of residence of the individual, a continent of residence of the individual, and/or a selection of whether the individual is resident in an urban or rural environment.

The above-mentioned steps of receiving may comprise receiving via an interface. The interface may comprise a user interface. The above-mentioned steps of outputting may comprise outputting by the processor. The above-mentioned steps of outputting may comprise communicating the output information to the subject individual and/or the eye care practitioner (and/or any other user). The above-mentioned steps of outputting may comprise outputting via the (user) interface. The above-mentioned steps of outputting may comprise outputting to a display. The display may be a display of the (user) interface. The display may be a computer screen. The above-mentioned steps of outputting may comprise causing the output information to be displayed on the display. The method may comprise a step of displaying the output information on the display. It will be understood that the output information may be any information which is output in the or each outputting step. For example, the output information may be the estimated future change, the indication of whether the myopia control treatment is determined to be a suitable, and/or the indication of the selected myopia control treatment, etc.

The step of receiving the dataset at the processor may comprise the processor accessing the dataset. The dataset may be stored in a memory, from which the dataset is accessed.

The present disclosure provides, according to the second aspect, a method of treating myopia. The method may be regarded as computer-implemented. The method may use the computer-implemented method of the first aspect. The method may comprise a step of obtaining the one or more patient parameters of the subject individual. The time of obtaining the patient parameters may define the reference time point. For example, the method may comprise a step of obtaining a measurement of the axial length and/or the refractive error of the eye of the subject individual. The time of the measurement may define the reference time point. The method may comprise a step of performing the computer-implemented method according to the first aspect of the invention to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment. The method may comprise a step of providing the myopia control treatment to the subject individual.

The step of providing the myopia control treatment may comprise providing a contact lens to the subject individual and/or providing a contact lens prescription to the subject individual.

The method may comprise a step of determining whether the myopia control treatment is a suitable myopia control treatment for the subject individual based on the estimated future change in the axial length and/or the refractive error of the eye. The myopia control treatment may be provided only if the myopia control treatment is determined to be suitable. The step of providing the myopia control treatment may comprise providing the myopia control treatment which is determined to be suitable for the subject individual.

The method may comprise a step of performing the computer-implemented method according to the first aspect of the invention to estimate, for each of a plurality of different myopia control treatments (e.g. a first and a second myopia control treatments), a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a treatment period during which the eye is treated with the myopia control treatment.

The method may comprise a step of selecting (e.g. by the selector) a myopia control treatment for the subject individual from the plurality of myopia control treatments. The myopia control treatment may be selected based on the plurality of estimated future changes in the axial length and/or the refractive error of the eye. The step of providing the myopia control treatment may comprise providing the selected myopia control treatment.

The method may comprise monitoring the progress of the myopia control treatment provided to the subject individual. The monitoring may comprise a step of obtaining a measurement of the axial length and/or the refractive error of the eye of the subject individual after providing the myopia control treatment for a period of time. The monitoring may comprise a step of storing the obtained measurement in a memory. The monitoring may comprise a step of inputting the obtained measurement into a patient record. The monitoring may comprise comparing the measured change in the axial length and/or the refractive error of the eye with a future change estimated over the same period of time using the machine learning agent.

The present disclosure provides, according to the third aspect, a computer program product being executable to perform the computer-implemented method according to the first aspect.

The computer program product may comprise a non-transitory computer readable storage medium comprising a program executable by a processor to perform said computer-implemented method. The program may be in the form of non-transitory source code, object code, or in any other non-transitory form suitable for use in the implementation of the methods described herein. The carrier may be any entity or device capable of carrying the program, such as a RAM, a ROM, or an optical memory device, etc.

The present disclosure provides, according to the fourth aspect, a computer system comprising: a memory storing instructions; and a processor coupled to the memory. The processor is configured to run the stored instructions to perform the computer-implemented method according to the first aspect.

The computer system may further comprise an interface. The interface may comprise a user interface. The user interface may comprise a webpage, for example a webpage displayed on a display. The interface may comprise a software interface. The dataset relating to the subject individual may be partly received via the user interface. For example, the axial length and/or refractive error may be received via the user interface. The dataset relating to the subject individual may be partly received via the software interface from a database. For example, previously obtained data on the subject individual, such as the age of the subject individual, may be received via the software interface. The memory and/or processer may be provided by a remote server (e.g. remote from the user interface).

The processor may be a single device or a collection of processing devices distributed remotely from each other. The processor may comprise a central processing unit (CPU). The processor may comprise a graphics processing unit (GPU). The processor may comprise one or more of a field programmable gate array (FPGA), a programmable logic device (PLD), or a complex programmable logic device (CPLD). The processor may comprise an application specific integrated circuit (ASIC). It will be appreciated by the skilled person that many other types of devices, in addition to the examples provided, may be used to provide the processor. The processor may comprise multiple co-located processors or multiple disparately located processors. Operations performed by the processor may be carried out by one or more of hardware, firmware, and software.

The processor may comprise data storage. The data storage may comprise one or both of volatile and non-volatile memory. The data storage may comprise one or more of random access memory (RAM), read-only memory (ROM), a magnetic or optical disk and disk drive, or a solid- state drive (SSD). It will be appreciated by the skilled person that many other types of memory, in addition to the examples provided, may also be used. It will be appreciated by a person skilled in the art that the processor may comprise more, fewer and/or different components from those described.

It will of course be appreciated that features described in relation to one aspect of the present disclosure may be incorporated into other aspects of the present disclosure. For example, the method of the disclosure may incorporate any of the features described with reference to the apparatus of the disclosure and vice versa.

Figure 3 shows schematically a subject individual 102 having an eye 104. A plurality of patient parameters 106 can be obtained from the subject individual 102. In accordance with the below embodiments of the disclosure, the patient parameters 106 include: the axial length 108 of the eye 104, the age 110 of the subject individual 102, an optical treatment type 112 previously used by the subject individual 102, a length of time 114 the subject individual 102 had previously used the optical treatment type, a biological sex 116 of the subject individual 102, an ethnicity of the mother 118 of the subject individual 102, and an ethnicity of the father 120 of the subject individual 102.

Figure 4 shows a computer system 126 in accordance with a first embodiment of the disclosure. The computer system 126 comprises a memory 128 which stores instructions in the form of a computer program product 132 comprising a machine learning agent 134. A processor 136 is coupled to the memory 128 and is configured to run the stored instructions. Both the memory 128 and processor 136 are provided by a remote server 138.

The computer system 126 is configured such that the processor 136 receives (i.e. can access) a dataset relating to the subject individual 102, that dataset comprising data defining each of the patient parameters 106 for the subject individual 102.

The processor 136 is coupled to an interface 140 via which data defining at least some of (possibly all of) the patient parameters 106 is received. The interface 140 comprises a user interface 140a via which a user 142, such as an eye care practitioner, can input at least some of (possibly all of) the patient parameters 106. The user 142 may input the patient parameters 106 after obtaining them from the subject individual 102, for example during an examination of the subject individual 102. In the present embodiment, the user interface 140a comprises a webpage containing fields for entry of the patient parameters 106. Upon entry of the patient parameters 106, they are uploaded to the remote server 138 and accessible to the processor 136.

In some embodiments, the interface 140 also comprises a software interface 140b. The software interface 140b is configured such that at least some of the patient parameters 106 can be obtained automatically from a database 144 via the software interface. The database 144 stores data which has previously been obtained from the subject individual 102. The database 144 could, for example, be maintained by the user 142. Alternatively or additionally, the software interface 140b is configured such that at least some of the patient parameters 106 can be obtained automatically from a measurement apparatus 146 via the software interface 140b. The measurement apparatus could, for example, be an apparatus configured to measure the axial length of the eye 104 of the subject individual 102. In some embodiments, at least some of the patient parameters 106 are already stored in the memory 128. This may negate the need for the processor 136 to receive all of the patient parameters via the user interface 140a and/or software interface 140b.

The machine learning agent 134 is configured to estimate a future change 148 in the axial length of the eye 104 of the subject individual 102 should the eye 104 be treated with a myopia control treatment. The user interface 140a is arranged to output the estimated future change 148 to the user 142.

As shown schematically in Figure 5, the estimated future change 148 is based on each of the patient parameters 106 in the received dataset. The future change 148 over the duration of a treatment period is estimated. The duration of the treatment period is specified by the user, for example via the user interface 140a. In the present embodiment, the treatment period is 12 months.

The machine learning agent 134 comprises a tree-based model. The model comprises a tree structure which defines the relationship between the patient parameters 106 as inputs and the future change 148 in the axial length as an output. The branches in the tree structure have different weightings which were calculated when the machine learning agent 134 was trained. As set out above, example tree-based models available in Python programming language include: extremely randomized tree regressor (ExtraTreeRegressor), extreme Gradient Boosting regressor (XGBRegressor), random forest regressor (RandomForestRegressor), and decision tree regressor (DecisionTreeRegressor).

The machine learning agent 134 was trained using a corpus of training data 150 comprising a plurality of training datasets 150a... 150n. Each of the training datasets 150a... 150n relates to a different treated individual. Each treated individual has an eye which has been treated using the myopia control treatment over the duration of a respective treated period. In the present embodiment, each treated individual has an eye which has been treated using the myopia control treatment over a 12 month treated period. In alternative embodiments, at least some of the treated individuals have been treated for different lengths of time, the training dataset defining that length of time, such that the data can be normalised prior to the machine learning agent 134 being trained.

Each of the training datasets 150a... 150n comprises data defining the patient parameters 106 for the respective treated individual at a treated time point when the eye of the treated individual was treated with the myopia control treatment. Each of the training datasets 150a... 150n therefore defines the following parameters: the axial length 108a, n of the eye at the treated time point, the age 110a, n of the treated individual at the treated time point, an optical treatment type 112a,n previously used by the treated individual, a length of time 114a,n the treated individual had previously used the optical treatment type, a biological sex 116a,n of the treated individual, an ethnicity of the mother 118a, n of the treated individual, and an ethnicity of the father 120a,n of the treated individual.

The treated time point is, for each of the training datasets 150a... 150n, the beginning of the respective treated period. In alternative embodiments, the treated time point is the end of the respective treated period, or a known time during the respective treated period.

Each of the training datasets 150a... 150n also comprises data defining a change 148a, n in the axial length of the eye of the respective treated individual over the duration of the treated period. The change has previously been determined by measurement, for example carried out by an eye care practitioner. In the present embodiment, the training datasets 150a... 150n each comprise a value of the magnitude of the change 148a,n. In alternative embodiments, the training datasets 150a... 150n comprise start and end values of the axial length of the eye, from which the magnitude of the change 148a,n can be calculated or inferred.

As will be understood by the skilled person, training the machine learning agent involves the machine learning agent determining a relationship, in the form of a tree-based model, which relates the patient parameters 106 as input variables to the future change in the axial length 148 as an output.

Figure 6 shows a method 1000 in accordance with a second embodiment of the disclosure. The method utilises the computer system 126 described above.

At step 1002, a user 142 in the form of an eye care practitioner obtains from the subject individual 102 a plurality of patient parameters 106. The axial length 108 of the eye 104 of the subject individual is obtained by measurement using a measurement apparatus. The time of the measurement defines a reference time point. The following further patient parameters 106 are obtained: the age 110 of the subject individual 102 at the reference time point, an optical treatment type 112 previously used by the subject individual 102, a length of time 114 the subject individual 102 had previously used the optical treatment type, a biological sex 116 of the subject individual 102, an ethnicity of the mother 118 of the subject individual 102, and an ethnicity of the father 120 of the subject individual 102.

At step 1004, the user 142 inputs the obtained patient parameters via the user interface 140a, the patient parameters 106 then being uploaded to the remote server 138 so they are accessible to the processor 136.

At step 1006, the processor 136 receives a dataset comprises data defining the plurality of patient parameters 106 obtained from the subject individual 102.

At step 1008, the processor 136 operates the machine leaning agent 134. The machine learning agent 134 estimates, based on each of the patient parameters 106, the future change 148 in the axial length of the eye 104 of the subject individual 102. The future change 148 is estimated over the duration of the user-specified 12 month treatment period beginning at the reference time point.

At step 1010, the interface 140 outputs the estimated future change 148 in the axial length of the eye 104 estimated by the machine learning agent. The future change 148 is displayed on a screen of the user interface 140a. The eye care practitioner 142 may use the estimated future change 148 to help them determine whether it would be beneficial for the subject individual 102 to commence or continue treatment with the myopia control treatment.

Figure 7 shows a method 2000 in accordance with a third embodiment of the disclosure. The method utilises the computer system 126 described above. In this example, the subject individual 102 has previously been treated with the myopia control treatment over a treatment period spanning the previous 12 months.

At step 2002, a user 142 in the form of an eye care practitioner retrieves a plurality of patient parameters 106 previously obtained from the subject individual 102 at the beginning of the previous 12 month treatment period. The beginning of that treatment period defines a reference time point. The following patient parameters 106 were previously obtained: the axial length 108 of the eye 104 of the subject individual 102 at the reference time point, the age 110 of the subject individual 102 at the reference time point, an optical treatment type 112 previously used by the subject individual 102, a length of time 114 the subject individual 102 had previously used the optical treatment type, a biological sex 116 of the subject individual 102, an ethnicity of the mother 118 of the subject individual 102, and an ethnicity of the father 120 of the subject individual 102 At step 2004, the user 142 inputs the previously obtained patient parameters 106 via the user interface 140a, the patient parameters 106 then being uploaded to the remote server 138 so they are accessible to the processor 136.

At step 2006, the processor 136 receives a dataset comprises data defining the plurality of previously obtained patient parameters 106.

At step 2008, the processor 136 operates the machine leaning agent 134. The machine learning agent 134 estimates, based on each of the patient parameters 106, the future change 148 in the axial length of the eye 104 of the subject individual 102. The future change 148 is estimated over the duration of the user-specified 12 month treatment period beginning at the reference time point.

At step 2010, the interface 140 outputs the estimated future change 148 in the axial length of the eye 104 estimated by the machine learning agent. The future change 148 is displayed on a screen of the user interface 140a.

At step 2012, the eye care practitioner 142 obtains from the subject individual 102 the current axial length of the eye 104 by measurement using a measurement apparatus. The eye care practitioner then inputs the current axial length of the eye 104 via the user interface 140, the value then being uploaded to the remote server 138 so it is accessible to the processor 136.

At step 2014, the processor 136 calculates the actual (measured) change in the axial length 108 of the eye 104 over the duration of the treatment period.

At step 2016, the processor 136 then calculates a difference between: the estimated future change 148 in the axial length of the eye 104 with the actual change in the axial length over the duration of the treatment period.

At step 2018, the processor 136 determines the extent to which the subject individual 102 has responded to the myopia control treatment over the previous treatment period by comparing the calculated difference to a threshold value. The individual is determined to be a non-responder to the myopia control treatment if the difference exceeds the threshold value.

At step 2020, the interface 140 outputs an indication of the extent to which the subject individual 102 has responded to the myopia control treatment, the indication comprising the determination as to whether the subject individual 102 is a non-responder. The eye care practitioner 142 may use the indication to help them determine whether it would be beneficial for the subject individual 102 to continue or stop treatment with the myopia control treatment. Figure 8 shows a computer system 226 in accordance with a fourth embodiment of the disclosure. The computer system 226 is configured in a similar way to the computer system 126 described above in relation to the first embodiment. The computer system 226 comprises a memory 228 and a processor 236 provided in a remote server 238. The memory 228 stores instructions in the form of a computer program product 232 comprising a machine learning agent 234. The processor 236 is coupled to an interface 240 comprising a user interface 240a and optionally a software interface 240b. Patient parameters 106 obtained from the subject individual 102 may be inputted by a user 142 via the user interface 240a, and/or automatically obtained from a database 144 or a measurement apparatus 146 via the software interface 240b.

The machine learning agent 234 comprises a tree-based model and is configured to estimate: a future change 248a in the axial length of the eye 104 of the subject individual 102 should the eye 104 be treated with a first myopia control treatment, a future change 248b in the axial length of the eye 104 of the subject individual 102 should the eye 104 be treated with a second myopia control treatment, and a future change 248c in the axial length of the eye 104 of the subject individual 102 should the eye 104 be untreated. The future changes 248a-c are estimated over the duration of a user-specified 12 month treatment period.

The user interface 240a is arranged to output each of the estimated future changes 248a-c to the user 142.

Figure 9 shows schematically the machine learning agent 234. The estimated future changes 248a-c are based on each of the patient parameters 106 in a received dataset. The machine learning agent 234 is trained using a corpus of first training data 250 comprising a plurality of first training datasets 250a...250n, a corpus of second training data 252 comprising a plurality of second training datasets 252a...252n, and a corpus of third training data comprising a plurality of third training datasets 254a...254n.

Each of the first training datasets 250a...250n relates to a different first treated individual having an eye which has been treated using the first myopia control treatment over the duration of a respective treated period. Each of the second training datasets 252a...252n relates to a different second treated individual having an eye which has been treated using the second myopia control treatment over the duration of a respective treated period. Each of the third training datasets 254a...254n relates to a different untreated individual having an eye which has not been treated with an optical treatment over the duration of a respective untreated period. Each of the first and second training datasets 250a...250n, 252a...252n comprises data defining the patient parameters 106 for the respective treated individual at a treated time point when the eye of the treated individual was treated with the myopia control treatment. Each of the third training datasets 254a...254n comprises data defining the patient parameters 106 for the respective untreated individual at an untreated time point when the eye of the treated individual was not treated.

Each of the training datasets 250a...250n, 252a...252n, 254a...254n also comprises data defining a change in the axial length of the eye of the respective treated/untreated individual over the duration of the treated/untreated period.

Figure 10 shows a method 3000 of treating myopia in accordance with a fifth embodiment of the disclosure. The method utilises the computer system 226 described above.

At step 3002, a user 142 in the form of an eye care practitioner obtains from the subject individual 102 a plurality of patient parameters 106. The axial length 108 of the eye 104 of the subject individual is obtained by measurement using a measurement apparatus. The time of the measurement defines a reference time point. The following further patient parameters 106 are obtained: the age 110 of the subject individual 102 at the reference time point, an optical treatment type 112 previously used by the subject individual 102, a length of time 114 the subject individual 102 had previously used the optical treatment type, a biological sex 116 of the subject individual 102, an ethnicity of the mother 118 of the subject individual 102, and an ethnicity of the father 120 of the subject individual 102.

At step 3004, the user 142 inputs the obtained patient parameters via the user interface 240a, the patient parameters 106 then being uploaded to the remote server 238 so they are accessible to the processor 236.

At step 3006, the processor 236 receives a dataset comprises data defining the plurality of patient parameters 106 obtained from the subject individual 102.

At step 3008, the processor 236 operates the machine learning agent 234. The machine learning agent 234 estimates, based on each of the patient parameters 106, the respective future changes 248a-c in the axial length of the eye 104 of the subject individual 102 should the eye 104 be treated with the first myopia control treatment, should the eye 104 be treated with the second myopia control treatment, and should the eye 104 remain untreated. The future changes 248a-c are estimated over the duration of the user-specified 12 month treatment period beginning at the reference time point.

At step 3010, the processor 236 selects a myopia control treatment for the subject individual 102 by comparing: the estimated future change 248a should the eye 104 be treated with the first myopia control treatment, and the estimated future change 248b should the eye be treated with the second myopia control treatment. The selected myopia control treatment is the myopia control treatment having the lowest estimated future change 248a, b in the axial length of the eye 104.

At step 3012, the processor 236 determines whether the selected myopia control treatment is a suitable myopia control treatment by comparing: the estimated future change 248a, b should the eye be treated with the selected myopia control treatment, and the estimated future change 248c should the eye be untreated. The selected myopia control treatment is determined to be suitable if the difference exceeds a predetermined threshold value, otherwise the selected myopia control treatment is determined to be unsuitable.

At step 3012, the interface 240 outputs an indication of the selected myopia control treatment and an indication of whether the selected myopia control treatment is suitable for the subject individual 102.

At step 3014, if the selected myopia control treatment is suitable, the user 242 provides the selected myopia control treatment to the subject individual 102.

In some embodiments, the user 242 provides the myopia control treatment by dispensing the myopia control treatment directly to the subject individual 102. For example, if the selected myopia control treatment is a specific type of contact lens, the user 242 dispenses the contact lens to the subject individual. In some embodiments, the user 242 provides the myopia control treatment by providing a prescription for the myopia control treatment.

A first example machine learning agent was developed by the applicant using six years of clinical data comprising 1241 samples. Each sample related to a single individual and comprised a dataset comprising: the axial length of the eye of the individual when the clinical study began, the refractive error of an eye of the individual when the clinical study began, the age of the individual when the clinical study began, the biological sex of the individual, the ethnicity of the mother of the individual (selected from Caucasian, Asian and Other), the ethnicity of the father of the individual (selected from Caucasian, Asian and Other), and the annual change in the axial length of the eye of the individual.

Each sample also defined the type of optical treatment used by the individual, which was either a single vision lens or a single specific type of commercially available myopia control lens. The sample therefore defined whether the individual was an untreated individual or a treated individual.

Each sample also defined the number of treatment days the individual was treated by an optical treatment. The sample therefore defined the length of the untreated period or treated period.

The clinical data was randomly divided by an 80:20 split into 992 samples used for training and 249 samples reserved for testing.

Using the data in the 992 samples, a corpus of first training data comprising a plurality of first training datasets relating to the treated individuals and a corpus of second training data comprising a plurality of second training datasets relating to the untreated individuals was produced. Each training datasets comprised data defining the following patient parameters: the age of the individual, the biological sex of the individual, the ethnicity of the mother of the individual, and the ethnicity of the father of the individual; along with data defining: the change in the axial length of the eye over the treated/untreated period, and the length of the treated/untreated period.

The machine learning agent was trained using the first and second training data. The machine learning agent was trained using Python programming language using an extreme Gradient Boosting regressor (XGBRegressor) model.

Figure 11 is a graph which shows the importance of each variable to how the first example machine learning agent estimates a change in the axial length of the eye. Along the x- axis, “xgbr2 Feature Importance” is a measure of how useful or valuable each variable was in the construction of the decision trees within the model. The more a variable is used to make key decisions within the decision trees, the higher its relative importance.

As can be seen, the optical treatment type (“Lens”) was by far the most important variable. This indicates that whether an individual used single vision lenses or the myopia control lenses - i.e. whether the individual was an untreated individual or a treated individual - had the largest impact on the change in the axial length of the eye. The patient parameters were important in the following descending order: age of the individual (“Age BL”), ethnicity of the father of the individual (“F Ethnicity”), biological sex of the individual (“Gender”), and ethnicity of the mother of the individual (“M Ethnicity”). The length of the treated/untreated period (“TreatmentDays”) was determined to be more important than the biological sex of the individual, but less than the ethnicity of the father of the individual.

A second example machine learning agent was developed by the applicant using the above-mentioned clinical data plus an additional set of clinical data. This increased the number of samples used for training to 1088 samples.

The second example machine learning agent was trained using a more limited number of patient parameters in the training datasets. Only the age of the individual and the biological sex of the individual were used as patient parameters. With the extra samples added, a Phi k correlation determined that the ethnicity of the father of the individual and the ethnicity of the mother of the individual had a zero correlation with the change in the axial length of the eye. The machine learning agent was still based on the XGBRegressor model.

Figure 12 shows the importance of each variable to how the second example machine learning agent estimates a change in the axial length of the eye. Again, the optical treatment type was most important. This was followed in descending order by: the age of the individual, the length of the treated/untreated period, and the biological sex of the individual.

Figure 13 shows a SHAP summary plot for 273 samples used to test the second example machine learning agent. The SHAP summary plot indicates how much each variable contributed, either positively or negatively, to the prediction of the change in the axial length (the model output). Feature importance is ranked in descending order, optical treatment type being the most important variable. The SHAP value is the average marginal contribution of a feature value across all coalitions of features. The horizontal location of each dot shows whether the effect of that value is associated with a higher (right) or lower (left) prediction. The colour (converted to greyscale in Figure 13) shows whether that feature is high (darker) or low (lighter) in value.

Correlations between the variables and the change in axial length can be seen in the plot. For example, optical treatment type can be seen as having has a clear correlation with the model output, using a myopia control lens led to lower change in the axial length as all the darker dots were on the left side. Furthermore, increasing age of the individual (indicated by a high level of “Age BL”) can be seen as having a negative impact on the model output as darker dots are primarily to the left side. Therefore, increasing age is negatively correlated with the change in the axial length.

Figures 14 and 15 show SHAP force plots for two different samples in the test data. SHAP force plots show how much each variable contributed to the prediction of the change in the axial length of each individual. In the Figure 14 plot, the variables “TreatmentDays”, “Age BL” and “Lens” (0 indicating single vision lens) drove the prediction to a higher value, “Gender” (0 indicating female) and drove it to a lower value. The final prediction was 0.42. In the Figure 15 plot, the variables “Age BL”, “Lens” (1 indicating myopia control lens), and “TreatmentDays” drove the prediction to lower value, and “Gender” drove it to higher value, the final prediction was at 0.06.

Whilst the present invention has been described and illustrated with reference to particular embodiments, it will be appreciated by those of ordinary skill in the art that the invention lends itself to many different variations not specifically illustrated herein. By way of example only, certain possible variations will now be described.

In alternative embodiments, the memory 128, 228 and the processor 136, 236 are provided locally to the user 142, for example in a personal computer. In such embodiments, the computer program product 132, 232 may be in the form of a software package installable on the personal computer.

In alternative embodiments, the computer program product 232 comprises first, second and third machine learning agents. The first machine learning agent is trained on the corpus of first training data 250 for estimating the future change 248a should the eye 104 be treated with the first myopia control treatment. The second machine learning agent is trained on the corpus of second training data 252 for estimating the future change 248b should the eye 104 be treated with the second myopia control treatment. The third machine learning agent is trained on the corpus of third training data 254 for estimating the future change 248c should the eye 104 be untreated.

In alternative embodiments, the method may make use of more, fewer or different patient parameters. For example, the method could make use of only the age 110 and the biological sex 116 of the subject individual 102, whereby only those patient parameters are contained in the received dataset, and only those patient parameters are used (in combination with a change in the axial length) to train the machine learning agent. In alternative embodiments, the age 110 of the subject individual 102 is not used as a patient parameter.

In alternative embodiments, equivalent computer systems and methods are used to estimate a future change in the refractive error of the eye 104. Similarly, equivalent computer systems and methods could be used to estimate a future change in both the axial length of the eye 104 and the refractive error of the eye 104 at the same time. In the above-described embodiments, references to the axial length of the eye could be changed for references to the refractive error of the eye, or changed for references to the axial length and the refractive error of the eye, as appropriate.

The steps of the methods described above could be carried out in any technically feasible order, they do not necessarily need to be carried out in the order described above.

In embodiments, the computer implemented method disclosed herein may further comprise measuring or determining the axial length and/or the refractive error of the eye of the subject individual and inputting into the dataset.

In embodiments, the computer implemented method disclosed herein may further comprise measuring or determining the axial length and/or the refractive error of the eye of the subject individual after providing a myopia control treatment for a period of time and inputting into either the dataset and/or the corpus of training data. The period of time may be at least 1 month, or at least 3 months, or at least 6 months, or at least one year.

In embodiments, the outputting the future change in the axial length and/or the refractive error of the eye of the subject individual is communicated to the subject individual and/or an eye care practitioner.

In embodiments, a measurement device for measuring the axial length and/or the refractive error of the eye of the subject individual is utilized on said subject individual to obtain eye data, and said measurement device communicates said eye data to the processor.

The measurement device, for example, can be an axial measurement device and/or a refractive error measurement device. More specific examples can include, but are not limited to, an applanation ultrasound device, an optical biometry device, an autorefractor, an interferometry measurement instrument, an OCT imaging device, a refractive error phoropter, and the like.

It will be appreciated that the techniques described herein may be implemented in software or hardware, or may be implemented using a combination of software and hardware. They may include configuring an apparatus to carry out and/or support any or all of techniques described herein. Although at least some aspects of the examples described herein with reference to the drawings comprise computer processes performed in processing systems or processors, examples described herein also extend to computer programs, for example computer programs on or in a carrier, adapted for putting the examples into practice. The carrier may be any entity or device capable of carrying the program. The carrier may comprise a computer readable storage media. Examples of tangible computer-readable storage media include, but are not limited to, an optical medium (e.g., CD-ROM, DVD-ROM or Blu-ray), flash memory card, floppy or hard disk or any other medium capable of storing computer-readable instructions such as firmware or microcode in at least one ROM or RAM or Programmable ROM (PROM) chips.

Where in the foregoing description, integers or elements are mentioned which have known, obvious or foreseeable equivalents, then such equivalents are herein incorporated as if individually set forth. Reference should be made to the claims for determining the true scope of the present invention, which should be construed so as to encompass any such equivalents. It will also be appreciated by the reader that integers or features of the invention that are described as preferable, advantageous, convenient or the like are optional and do not limit the scope of the independent claims. Moreover, it is to be understood that such optional integers or features, whilst of possible benefit in some embodiments of the invention, may not be desirable, and may therefore be absent, in other embodiments.

Claims

1. A computer-implemented method for estimating a future change in an axial length and/or a refractive error of an eye; wherein one or more patient parameters are obtainable for an individual, the patient parameters including:

- an age of the individual; wherein the method comprises the following steps: receiving, at a processor, a dataset relating to a subject individual, the dataset comprising data defining each of the patient parameters for the subject individual at a reference time point; operating, using the processor, a machine learning agent to estimate a future change in the axial length and/or the refractive error of an eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment, wherein the future change in the axial length and/or the refractive error of the eye is estimated based on each of the patient parameters in the received dataset; outputting the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent; wherein the machine learning agent has been trained using a corpus of training data comprising a plurality of training datasets, each training dataset relating to a different treated individual having an eye which has been treated using the myopia control treatment over a treated period; wherein each training dataset comprises data defining:

- each of the patient parameters for the treated individual at a treated time point when the treated individual was treated, and

- a change in an axial length and/or a refractive error of the eye of the treated individual over the treated period.

2. A computer-implemented method according to any preceding claim, wherein the machine learning agent comprises a decision tree model or a regression tree model.

3. A computer-implemented method according to any preceding claim, wherein the myopia control treatment comprises treatment using an ophthalmic lens comprising a base-power region for correcting distance vision, and an add-power region for inducing a myopic defocus.

4. A computer-implemented method according to claim 3, wherein base-power region forms a central region of the lens, and the add-power region forms an annular region around the central region.

5. A computer-implemented method according to any preceding claim, wherein the patient parameters include:

- an optical treatment type previously used by the individual, and

- a length of time the individual had previously used the optical treatment type.

6. A computer-implemented method according to claim 5, wherein the optical treatment type previously used may be selected from: a single vision lens, the myopia control treatment, and another myopia control treatment.

7. A computer-implemented method according to any preceding claim, wherein the patient parameters include:

- an ethnicity of the individual.

8. A computer-implemented method according to any preceding claim, wherein the patient parameters include:

- an ethnicity of the mother of the individual.

9. A computer-implemented method according to any preceding claim, wherein the patient parameters include:

- an ethnicity of the father of the individual.

10. A computer-implemented method according to any preceding claim, wherein the patient parameters include: - a biological sex of the individual.

11. A computer-implemented method according to any preceding claim, wherein the patient parameters include:

- an axial length and/or a refractive error of an eye of the individual.

12. A computer-implemented method according to any preceding claim, wherein the subject individual has previously been treated with the myopia control treatment for the treatment period, the treatment period thereby comprising a time period which is in the past relative to a point in time when the method is performed, the reference time point being the beginning of the treatment period; the method comprising a step of: comparing (i) the estimated future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period with (ii) a measured change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period.

13. A computer-implemented method according to claim 12, wherein the step of comparing comprises calculating a difference between the estimated change in the axial length and/or refractive error of the eye and the measured axial length and/or the refractive error of the eye; wherein the method comprises a step of comparing the difference to a threshold value; and wherein the individual is determined to be a non-responder to the myopia control treatment if the difference exceeds the threshold value.

14. A computer-implemented method according to any preceding claim, wherein the myopia control treatment is a first myopia control treatment, the method further comprising a step of: operating a machine learning agent to estimate a future change in the axial length and/or refractive error of the eye of the subject individual over the duration of the treatment period during which the eye is treated with a second myopia control treatment, wherein the future change in the axial length and/or refractive error of the eye is estimated based on each of the patient parameters in the received dataset; outputting the future change in the axial length and/or refractive error of the eye treated by the second myopia control treatment estimated by the machine learning agent; wherein the machine learning agent has been trained using a corpus of second training data comprising a plurality of second training datasets, each second training dataset relating to a different second treated individual having an eye which has been treated using the second myopia control treatment over a treated period; wherein each second training dataset comprises data defining:

- each of the patient parameters for the second treated individual at a treated time when the second treated individual was treated, and

- a change in an axial length and/or a refractive error of the eye of the second treated individual over the treated period.

15. A computer-implemented method according to any preceding claim, comprising steps of: operating a machine learning agent to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of the treatment period should the eye be untreated for the treatment period, wherein the future change in the axial length and/or the refractive error of the untreated eye is estimated based on each of the patient parameters in the received dataset; outputting the future change in the axial length and/or the refractive error of the untreated eye estimated by the machine learning agent; wherein the machine learning agent has been trained using a corpus of third training data comprising a plurality of third training datasets, each third training dataset relating to a different untreated individual having an eye which has been untreated over an untreated period; wherein each third training dataset comprises data defining:

- each of the patient parameters for the untreated individual at an untreated time point when the untreated individual was untreated,

- a change in an axial length and/or a refractive error of the eye of the untreated individual over the untreated period.

16. A computer-implemented method according to any preceding claim, comprising the steps of: determining whether the myopia control treatment is a suitable myopia control treatment for the subject individual based on the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent; and outputting an indication of whether the myopia control treatment is determined to be a suitable myopia control treatment for the individual.

17. A computer-implemented method according to any preceding claim, wherein the step of outputting the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent comprises outputting the future change via a user interface and/or communicating the future change to the subject individual and/or an eye care practitioner.

18. A computer-implemented method according to any preceding claim, wherein the method comprises: obtaining a measurement of the axial length and/or the refractive error of the eye of the subject individual, and storing the measurement in a memory of a computer system.

19. A computer-implemented method according to claim 18, wherein the measurement is obtained after providing the myopia control treatment for a period of time.

20. A computer-implemented method according to claim 19, wherein the period of time is at least 1 month, or at least 3 months, or at least 6 months, or at least one year.

21. A computer-implemented method according to any preceding claim, comprising the steps of: selecting a myopia control treatment for the subject individual based on the future change in the axial length and/or the refractive error of the eye estimated by the machine learning agent; and outputting an indication of the selected myopia control treatment.

22. A method of treating myopia using the computer-implemented method according any of claims 1 to 21, the method comprising the following steps: obtaining the one or more patient parameters of the subject individual; performing the computer-implemented method to estimate a future change in the axial length and/or the refractive error of the eye of the subject individual over the duration of a treatment period during which the eye is treated with a myopia control treatment; and providing the myopia control treatment to the subject individual.

23. A method of treating myopia according to claim 22, wherein providing the myopia control treatment comprises providing a contact lens to the subject individual and/or providing a contact lens prescription to the subject individual.

24. A method of treating myopia according to claim 22 or 23, comprising monitoring the progress of the myopia control treatment provided to the subject individual by obtaining a measurement of the axial length and/or the refractive error of the eye of the subject individual after providing the myopia control treatment for a period of time and inputting the obtained measurement into a patient record.

25. A computer program product being executable to perform the computer-implemented method according to any of claims 1 to 21.

26. A computer system comprising: a memory storing instructions; and a processor coupled to the memory, said processor being configured to run the stored instructions to perform the computer-implemented method according to any of claims 1 to 21.