WO2019193362A2

WO2019193362A2 - Determining a clinical outcome for a subject suffering from a macular degenerative disease

Info

Publication number: WO2019193362A2
Application number: PCT/GB2019/051003
Authority: WO
Inventors: Nilkunj DODHIA; Nigel Davies
Original assignee: Macusoft Ltd.
Priority date: 2018-04-05
Filing date: 2019-04-05
Publication date: 2019-10-10
Also published as: GB202017496D0; GB201805642D0; GB2587551A; WO2019193362A3

Abstract

A computer-implemented method of determining a clinical outcome for a subject suffering from a macular degenerative disease comprises: obtaining patient data relating to the subject and a plurality of macular images of the subject, the plurality of macular images comprising a plurality of images of at least a macular region in the subject's eye captured over a time period; inputting the obtained patient data and plurality of macular images to a machine learning algorithm, wherein the machine learning algorithm is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period; and outputting the determined clinical outcome for the subject. Apparatus for performing the method is also disclosed.

Description

Determining a Clinical Outcome for a Subject Suffering from a Macular Degenerative Disease

Technical Field

The present invention relates to a method, apparatus and computer program for determining a clinical outcome for a subject suffering from a macular degenerative disease.

Background

Common diseases that can affect the centre of the back of the eye and cause devastating loss of vision and blindness include wet Age-related Macular Degeneration (AMD), Diabetic Macula Oedema (DMO), and Retinal Vein Occlusion (RVO). Collectively these types of diseases can be referred to as macular degenerative diseases, as they cause degeneration of the macula. It has been estimated that 237 million people globally suffer from long-term macular eye diseases that lead to devastating sight loss, which could otherwise be preventable if monitored and treated regularly. Recent

breakthroughs in imaging of the eye and treatments have transformed the outcome for patients with these diseases, most notably through intravitreal injections. However, in many countries demand for these treatments is now overwhelming clinical capacity and leading to a rise in sight loss primarily due to delayed treatment. One recent study found that 78% of patients experienced permanent vision deterioration directly attributable to delayed treatment (Foot & MacEwen, 2017). There is therefore a need for an improved method of monitoring and treating macular degenerative diseases.

Summary of the Invention

According to a first aspect of the present invention, there is provided a computer- implemented method of determining a clinical outcome for a subject suffering from a macular degenerative disease, the method comprising: obtaining patient data relating to the subject and a plurality of macular images of the subject, the plurality of macular images comprising a plurality of images of at least a macular region in the subject’s eye captured over a time period; inputting the obtained patient data and plurality of macular images to a machine learning algorithm, wherein the machine learning algorithm is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period; and outputting the determined clinical outcome for the subject.

In some embodiments according to the first aspect, the time period over which the plurality of images are captured is at least one month.

In some embodiments according to the first aspect, the machine learning algorithm is trained to determine one or more of the following as the clinical outcome: a recommended time at which to schedule a follow-up appointment; a recommended frequency of a plurality of follow-up appointments; an anti vascular endothelial growth factor treatment; a referral to a medical practitioner; and a recommended course of treatment.

In some embodiments according to the first aspect, the plurality of macular images include one or more retinal images and/ or one or more optical coherence tomography OCT images of the subject’s eye.

In some embodiments according to the first aspect, the patient data comprises one or more of: information indicative of the subject’s age, gender and/or ethnicity; clinical history information related to a history of the current subject; and/or one or more retinal image parameters obtained from a retinal image of the subject.

In some embodiments according to the first aspect, the method further comprises: subsequently obtaining an updated macular image of the subject, after determining said clinical outcome; and inputting the patient data, the updated macular image and one or more of the plurality of macular images to the machine learning algorithm to determine a subsequent clinical outcome in dependence on a progression of the macular degenerative disease since the treatment was applied. In some embodiments according to the first aspect, the macular degenerative disease comprises: wet age-related macular degeneration; retinal vein occlusion; diabetic macular oedema; cystoid macular oedema; and/or choroidal neovascularization.

In some embodiments according to the first aspect, the method further comprises treating the subject according to the determined clinical outcome. In some embodiments according to the first aspect, the patient data comprises genomic information indicative of whether the subject has one or more genetic variations associated with the macular degenerative disease. In some embodiments according to the first aspect, the macular degenerative disease comprises wet age-related macular degeneration, and the one or more genetic variations include variations associated with wet age-related macular degeneration in one or more of the following genes: CFH; C3; CFI; C2; CFB; CETP; LIPC; ABCAl;

APOE; TIMP3; COL8A1; and ARMS2.

In some embodiments according to the first aspect, the one or more genetic variations comprise a plurality of possible variations within the same gene, each of the plurality of possible variations being associated with different levels of risk of the macular degenerative disease, and the genomic information is indicative of whether the subject has one of the plurality of possible variations within said gene.

In some embodiments according to the first aspect, the genomic information is obtained by obtaining genetic data indicative of the subject’s genome, and searching the genetic data to determine whether the subject has one or more of the genetic variations associated with the macular degenerative disease.

According to a second aspect of the present invention, there is provided a computer program comprising computer program instructions which, when executed, perform a method according to the first aspect.

According to a third aspect of the present invention, there is provided a non-transitory computer readable storage medium arranged to store a computer program according to the second aspect. According to a fourth aspect of the present invention, there is provided apparatus for determining a clinical outcome for a subject suffering from a macular degenerative disease, the apparatus comprising: an input for receiving patient data relating to the subject and a plurality of macular images of the subject, the plurality of macular images comprising a plurality of images of at least a macular region in the subject’s eye captured over a time period; a machine learning algorithm configured to receive the obtained patient data and plurality of macular images as data inputs, wherein the machine learning algorithm is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period; and an output configured to output the determined clinical outcome for the subject.

In some embodiments according to the fourth aspect, the time period over which the plurality of images are captured is at least one month.

In some embodiments according to the fourth aspect, the machine learning algorithm is trained to determine one or more of the following as the clinical outcome: a

recommended time at which to schedule a follow-up appointment; a recommended frequency of a plurality of follow-up appointments; an anti vascular endothelial growth factor treatment; a referral to a medical practitioner; and a recommended course of treatment.

In some embodiments according to the fourth aspect, the plurality of macular images include one or more retinal images and/or one or more optical coherence tomography OCT images of the subject’s eye. In some embodiments according to the fourth aspect, the patient data comprises one or more of: information indicative of the subject’s age, gender and/or ethnicity; clinical history information related to a history of the current subject; and/or a retinal image parameter obtained from a retinal image of the subject. In some embodiments the clinical history information may comprise one or more visual acuity test scores of the subject.

In some embodiments according to the fourth aspect, the macular degenerative disease comprises: wet age-related macular degeneration; retinal vein occlusion; diabetic macular oedema; cystoid macular oedema; and/or choroidal neovascularization.

Brief Description of the Drawings

Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure 1 illustrates apparatus for determining a clinical outcome for a subject suffering from a macular degenerative disease, according to an embodiment of the present invention; Figure 2 is a flowchart showing a computer-implemented method of determining a clinical outcome for a subject suffering from a macular degenerative disease, according to an embodiment of the present invention; and

Figure 3 is a diagram illustrating the macular region of the eye.

Detailed Description

In the following detailed description, only certain exemplary embodiments of the present invention have been shown and described, simply by way of illustration. As those skilled in the art would realise, the described embodiments may be modified in various different ways, all without departing from the scope of the present invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not restrictive. Like reference numerals designate like elements throughout the specification. Figure 1 schematically illustrates apparatus for determining a clinical outcome for a subject suffering from a macular degenerative disease, according to an embodiment of the present invention. Figure 2 is a flowchart showing a computer-implemented method performed by the apparatus of Fig. 1, according to the present embodiment.

The functional blocks illustrated in Fig. 1 may, for example, be implemented as software instructions executed by one or more processors, or may be implemented in dedicated hardware such as an application-specific integrated circuit (ASIC) or field programmable gate array (FPGA). In embodiments in which a software

implementation is used, the apparatus may comprise processing means in the form of one or more processors, and memory arranged to store computer program instructions which, when executed by the processing means, perform any of the methods disclosed herein.

The apparatus of the present embodiment comprises an input 100 configured to receive patient data 101 and a plurality of macular images 102 in step S201. The apparatus also comprises a machine learning algorithm 110 and an output 120. The machine learning algorithm 110 is configured to receive the patient data 101 and plurality of macular images 102 as input data in step S202. In the present embodiment the machine learning algorithm 110 comprises an artificial neural network, but in other

embodiments a different type of machine learning algorithm maybe used. The plurality of macular images 102 may, for example, include one or more retinal images of the subject’s eye, such as a fundus image. In other embodiments the plurality of macular images 102 may include other types of image instead of or in addition to a retinal image. For example, in some embodiments the plurality of retinal images may include one or more optical coherence tomography (OCT) images of the subject’s eye.

The machine learning algorithm 110 is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period, as indicated by the input patient data 101 and the macular images 102. The determined clinical outcome may also provide a personalised outcome that is suitable for that particular individual, meaning an outcome that takes into account individual characteristics of the current subject as indicated by the patient data 101 and the macular images 102. The determined clinical outcome is outputted in step S203 via a suitable output 120. For example, the clinical outcome maybe a recommendation to schedule a follow-up appointment 121 without treatment at this stage, or may be a recommendation to begin a particular course of treatment 122. In some embodiments the method may further comprise a step of treating the subject according to the determined clinical outcome, after step S203 in Fig. 3. The machine learning algorithm 110 can be trained to recognise various types of macular degenerative diseases and recommend appropriate courses of action, taking into account the patient data 101 and the progression of the disease in the current subject as indicated by the patient data 101 and the plurality of macular images 102 captured over time. Examples of macular diseases that the machine learning algorithm 110 may be trained to recognise in embodiments of the present invention include, but are not limited to:

• Wet Age-related macular degeneration

• Retinal vein occlusion

• Diabetic macular oedema

• Cystoid macular oedema from central retinal vein occlusion

• Cystoid macular oedema from branch retinal vein occlusion; and

• Choroidal neovascularization. In some embodiments, the machine learning algorithm may be provided with information about a type of macular disease from which the patient is already known to be suffering, and accordingly may not need to automatically determine the type of disease.

Here, the term‘outcome’ is used to denote a particular recommended course of action to be followed for the current subject, and may include a recommended treatment and/or monitoring regime. For example, the machine learning algorithm no maybe configured to determine one or more of the following as the clinical outcome, in dependence on the particular patient data 101 and macular images 102 that were inputted in step S202: a recommended time at which to schedule a follow-up appointment; a recommended frequency of a plurality of follow-up appointments; a recommended course of treatment, such as an anti vascular endothelial growth factor (anti-VEGF) treatment or stem cell therapy. In this way, the machine learning algorithm 110 can provide an automated triage process for determining a suitable course of action without requiring the intervention of a skilled medical professional, helping to free up valuable health service resources.

Studies by the inventor have shown that a machine learning algorithm 110 trained in this way can determine a suitable course of action which meets or exceeds the so-called ‘Gold standard’ based on the recommendation of an expert medical professional, for example a Consultant Ophthalmologist. In one study carried out by the inventor, the machine learning algorithm 110 was used to determine an outcome comprising a recommended course of treatment and follow-up for 53 macular degeneration patients. The outcomes determined by the algorithm 110 for the 53 patients were then compared against the recommendation of a Consultant Ophthalmologist. In this study, the outcomes determined by the algorithm 110 for 47 of the 53 patients was found to agree with the recommendation of the Consultant Ophthalmologist, indicating an accuracy of 88.7%. The outcomes determined by the algorithm 110 for the remaining 6 patients were deemed to be borderline in terms of determining a change to a pre-existing treatment plan for each patient, with the‘Gold standard’ re-review of these cases concluding that the difference in opinion was not going to cause harm to a patient due to the clinician’s caution towards risk. In other words, the clinician’s recommendation in these cases was deemed to lead to over-treatment. In comparison, previous studies comparing recommendations from a number of Ophthalmologists to the‘Gold-standard’ have shown that medical professionals typically achieve an accuracy of around 85% when compared to the Gold-standard. Therefore the machine learning algorithm 110 is able to determine the most suitable outcome for patients with an accuracy that matches or exceeds that of a human.

In some embodiments the machine learning algorithm 110 may be configured to recommend a referral to a medical practitioner, such as an ophthalmologist, as the determined clinical outcome. For example, the machine learning algorithm 110 may be arranged to attempt to classify the subject into one of a plurality of predefined clinical outcome categories in dependence on the input patient data 101 and plurality of macular images 102. In the event that the machine learning algorithm 110 is unable to assign the current subject to a particular one of the predefined categories with at least a certain level of confidence, the machine learning algorithm 110 may be configured to recommend a referral to a medical practitioner as the determined clinical outcome.

This approach can ensure that the machine learning algorithm 110 does not

inadvertently recommend an inappropriate course of action in a situation where the subject displays unusual symptoms and/or has a type of disease for which the algorithm 110 has not been trained, in which case the algorithm 110 may not be able to recommend the most appropriate course of action.

The plurality of macular images 102 include images that show at least the macular region in the subject’s eye. The macular region is an area of the retina 300 which contains the macula 310. The macula is a pigmented area near the centre of the retina 300, as shown in Fig. 3, and is responsible for high-resolution colour vision at the centre of the visual field. Degeneration of the macula 310 can therefore have a significant impact on visual acuity. Each one of the macular images 102 may show only the macular region, or may show both the macular region and other parts of the eye, for example part of the retina outside the macular region.

The plurality of macular images 102 include images that have been captured over a time period, and can therefore be used to track a progression of the macular degenerative disease in the subject. The images may be captured over relatively long time periods to enable the progression of the macular degenerative disease to be tracked over weeks, months or years. For example, the time period over which the macular images 102 are captured may be greater than or equal to one month. By tracking the progression of the macular degenerative disease over time in a specific subject, the machine learning algorithm no can more accurately determine the current stage of the disease and/or the likely response of that subject to a particular course of treatment.

The method shown in Fig. 2 may be repeated over time, to monitor the progression of the macular degenerative disease in the subject and recommend a new course of action as and when appropriate. For example, when an updated macular image of the subject is subsequently obtained, steps S202 and S203 can be repeated by inputting the updated macular image and one or more of the plurality of previous macular images to the machine learning algorithm to determine a subsequent clinical outcome.

The input 100 may, for example, comprise a user interface through which the patient data 101 can be entered, and/or through which a user can load copies of the macular images 102 to the apparatus. In some embodiments the input 100 may be configured to automatically retrieve the patient data 101 and/or one or more of the macular images 102 from storage, for example a database arranged to store patient data and/ or macular images for a plurality of subjects, such as an Electronic Health Record. The output 120 is configured to output the determined clinical outcome for the subject. For example, the output 120 may comprise a display unit for displaying the determined clinical outcome in a suitable graphical user interface. The output 120 may be configured to output the determined clinical outcome in a non-visual manner, for example by outputting the determined clinical outcome in the form of an audio message to communicate the outcome to a user. As a further example, in some embodiments the output 120 may include an interface configured to output the determined clinical outcome by transmitting a message indicative of the determined clinical outcome to another device, for instance in the form of a text message or email. For example, the output 120 may comprise a network interface for outputting the determined clinical outcome to another device over a network.

In some embodiments the apparatus may be implemented as a cloud-based computing platform in which a server is programmed to execute the machine learning algorithm based on input data received from a remote client. In such embodiments the input 100 may comprise a graphical user interface rendered in a web browser at the remote client, for example at a desktop computer, laptop computer, tablet computer, smartphone, or wearable electronic device. In another embodiment the machine learning algorithm no maybe executed locally, for example, the input 100, machine learning algorithm no and output 120 may all be included in the same physical device.

The patient data 101 may take various forms, depending on the particular embodiment. The patient data 101 can allow the machine learning algorithm 110 to take into account information specific to the current patient, such as their medical history and individual characteristics, when determining the clinical outcome. By inputting patient data 101 in addition to the plurality of macular images 102 to the machine learning algorithm 110, the accuracy of the determination can be improved. Depending on the

embodiment the patient data may be inputted manually and/ or may be acquired automatically via the input 100. For example, in some embodiments the input 100 may be arranged to receive the patient data from one or more sensors included in one or more wearable electronic devices such as a fitness band or smartwatch. The machine learning algorithm 110 may be trained using equivalent patient data and macular images from a large population of individuals, so as to allow the machine learning algorithm 110 to more accurately predict the response of a particular individual to a specific course of treatment, and determine an appropriate clinical outcome accordingly. For example, the machine learning algorithm 110 may be configured to take into account the individual’s personal history and the previous progression of the macular degenerative disease in their particular case, as indicated by the plurality of macular images 102, when determining the clinical outcome.

For example, the patient data may include information indicative of one or more characteristics of the current subject, such as their age, gender, and/or ethnicity. The patient data may include clinical history information related to a history of the current subject. For example, the clinical history information may indicate whether the subject has any other underlying medical conditions such as diabetes or cardiac conditions, and/ or may include information about the subject’s lifestyle, such as whether the subject is a smoker. In some embodiments the clinical history information may comprise information about dates and outcomes of previous appointments, for example diagnostic data and historical test results such as visual acuity test scores of the subject. Taking into account one or more visual acuity test scores can improve the accuracy of the output of the algorithm, since the visual acuity test score provides information about the function of the eye whereas the plurality of images provide information about the form of the eye. Depending on the embodiment the clinical history information that is inputted at step S202 may include the full clinical history information for the current subject since they were initially diagnosed with the macular degenerative disease, or may only include selected clinical history information from a certain time period, for example data from their three most recent appointments.

In some embodiments the patient data may comprise one or more retinal image parameters measured or otherwise obtained from an image of the subject’s retina, for example one of the plurality of macular images 102 or a separate image. The retinal image parameters may be obtained from any suitable image of the retina, for example a fundus image, OCT scan, or other type of image. Examples of types of parameters that maybe obtained from a retinal image and included in the inputted patient data 101 include, but are not limited to:

• Foveal volume;

• Foveal thickness;

• Macular volume;

• Macular thickness;

• A number, size and/ or distribution of a type of feature, such as vessels, drusen, haemorrages, micro-aneurisms, bright lesions, and/or exudates;

• Pigmentation;

• A relative change in morphology, location and/ or pigmentation relative to a retinal image captured at a previous point in time.

In some embodiments the patient data can comprise genomic information indicative of whether the subject has one or more genetic variations associated with the macular degenerative disease. The genomic information for the current subject may, for example, be retrieved from a database storing genetic data for a number of patients. In some embodiments, the genomic information can be obtained by obtaining genetic data indicative of the subject’s genome, and searching the genetic data to determine whether the subject has one or more of the genetic variations associated with the macular degenerative disease. The genomic information inputted into the machine learning algorithm 110 is then dependent on the result of the search, and indicates whether any of the genetic variations were found in the genetic data. For example, in the case of wet age-related macular degeneration (AMD), certain variations in specific genes are indicative of different levels of risk of developing AMD, including 15 variations in 12 genes present across the following four biological pathways:

• the complement system (CFH, C3, CFI, C2 and CFB);

• cholesterol metabolism (CETP, LIPC, ABCAl and APOE);

• extracellular matrix remodeling (TIMP3 and COL8A1); and

• oxidative stress (ARMS2).

As such, in some embodiments the patient data inputted to the machine learning algorithm 110 can comprise genomic information indicative of whether the subject has any of the variations in one or more of these 12 genes that are known to be indicative of a level of risk of developing AMD. For example, a genotype of‘AA’ in the CFH gene may be indicative of a lower risk of developing AMD, a genotype of‘GA’ may be indicative of an intermediate risk of developing AMD, and a genotype of‘GG’ may be indicative of a higher risk of developing AMD. Similarly, a genotype of‘CC n the CFI gene may be indicative of a lower risk of developing AMD, a genotype of‘CT may be indicative of an intermediate risk of developing AMD, and a genotype of TG may be indicative of a higher risk of developing AMD.

It will be appreciated that the above-described variations and risk levels are provided merely by way of example, and should not be construed as limiting. The genetic variations and associated risk levels will differ for different types of diseases.

Embodiments of the present invention can provide automated clinical decision support tools with algorithms trained using deep learning. A machine learning algorithm may be trained on a small confined area of the eye, specifically the macular region, to monitor specific bio-markers that are indicative of the progression of a particular type of macular degenerative disease. This provides a high level of accuracy while managing the complexity and cost of data preparation, algorithm training, computer processing effort, scaling-up and commercialisation. Furthermore, the machine learning algorithm may continually learn and re-train itself based on newly-inputted

information in the form of patient data and macular images. A 2016 ECHoES study of ophthalmologists found that human beings can typically achieve an accuracy rate of around 85% when manually making treatment decisions on patient diagnostics upon presentation. Tests conducted by the inventors have shown that an automated apparatus and method such as the one described above with reference to Figs. 1 and 2 can provide a >85% accuracy rate, and can therefore provide a more reliable decision regarding treatment while simultaneously freeing up healthcare resources. Embodiments of the invention will therefore allow ophthalmologists to devote more time to delivering treatment to patients, allowing more patients to be seen and treated.

Whilst certain embodiments of the invention have been described herein with reference to the drawings, it will be understood that many variations and modifications will be possible without departing from the scope of the invention as defined in the

accompanying claims.

Claims

1. A computer-implemented method of determining a clinical outcome for a subject suffering from a macular degenerative disease, the method comprising:

obtaining patient data relating to the subject and a plurality of macular images of the subject, the plurality of macular images comprising a plurality of images of at least a macular region in the subject’s eye captured over a time period;

inputting the obtained patient data and plurality of macular images to a machine learning algorithm, wherein the machine learning algorithm is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period; and

outputting the determined clinical outcome for the subject.

2. The method of claim 2, wherein the time period over which the plurality of images are captured is at least one month.

3. The method of claim l or 2, wherein the machine learning algorithm is trained to determine one or more of the following as the clinical outcome:

a recommended time at which to schedule a follow-up appointment;

a recommended frequency of a plurality of follow-up appointments;

an anti vascular endothelial growth factor treatment;

a referral to a medical practitioner; and

a recommended course of treatment.

4. The method of claim 1, 2 or 3, wherein the plurality of macular images include one or more retinal images and/ or one or more optical coherence tomography OCT images of the subject’s eye.

5. The method of any one of the preceding claims, wherein the patient data comprises one or more of:

information indicative of the subject’s age, gender and/or ethnicity;

clinical history information related to a history of the current subject; and/or one or more retinal image parameters obtained from a retinal image of the subject.

6. The method of claim 5, wherein the clinical history information comprises one or more visual acuity test scores of the subject.

7. The method of any one of the preceding claims, further comprising:

subsequently obtaining an updated macular image of the subject, after determining said clinical outcome; and

inputting the patient data, the updated macular image and one or more of the plurality of macular images to the machine learning algorithm to determine a subsequent clinical outcome in dependence on a progression of the macular degenerative disease since the treatment was applied.

8. The method of any one of the preceding claims, wherein the macular degenerative disease comprises:

wet age-related macular degeneration;

retinal vein occlusion;

diabetic macular oedema;

cystoid macular oedema; and/ or

choroidal neovascularization.

9. The method of any one of the preceding claims, further comprising:

treating the subject according to the determined clinical outcome.

10. The method of any one of the preceding claims, wherein the patient data comprises genomic information indicative of whether the subject has one or more genetic variations associated with the macular degenerative disease.

11. The method of claim 10, wherein the one or more genetic variations comprise a plurality of possible variations within the same gene, each of the plurality of possible variations being associated with different levels of risk of the macular degenerative disease, and

wherein the genomic information is indicative of whether the subject has one of the plurality of possible variations within said gene.

12. The method of claim 10 or 11, wherein the genomic information is obtained by: obtaining genetic data indicative of the subject’s genome; and searching the genetic data to determine whether the subject has one or more of the genetic variations associated with the macular degenerative disease.

13. A computer program comprising computer program instructions which, when executed, perform a method according to any one of the preceding claims.

14. A non-transitory computer readable storage medium arranged to store a computer program according to claim 13.

15. Apparatus for determining a clinical outcome for a subject suffering from a macular degenerative disease, the apparatus comprising:

an input for receiving patient data relating to the subject and a plurality of macular images of the subject, the plurality of macular images comprising a plurality of images of at least a macular region in the subject’s eye captured over a time period; a machine learning algorithm configured to receive the obtained patient data and plurality of macular images as data inputs, wherein the machine learning algorithm is trained to determine a clinical outcome for the subject in dependence on a progression of the macular degenerative disease over said time period; and

an output configured to output the determined clinical outcome for the subject.

16. The apparatus of claim 15, wherein the time period over which the plurality of images are captured is at least one month.

17. The apparatus of claim 15 or 16, wherein the machine learning algorithm is trained to determine one or more of the following as the clinical outcome:

a recommended time at which to schedule a follow-up appointment;

a recommended frequency of a plurality of follow-up appointments;

an anti vascular endothelial growth factor treatment;

a referral to a medical practitioner; and

a recommended course of treatment.

18. The apparatus of claim 15, 16 or 17, wherein the plurality of macular images include one or more retinal images and/or one or more optical coherence tomography OCT images of the subject’s eye.

19. The apparatus of any one of claims 15 to 18, wherein the patient data comprises one or more of:

information indicative of the subject’s age, gender and/or ethnicity;

clinical history information related to a history of the current subject; and/or a retinal image parameter obtained from a retinal image of the subject.

20. The apparatus of any one of claims 15 to 19, wherein the macular degenerative disease comprises:

wet age-related macular degeneration;

retinal vein occlusion;

diabetic macular oedema;

cystoid macular oedema; and/ or

choroidal neovascularization.