Computer-implemented method and apparatus for determining a personalised dermatological treatment

Abstract

Examples disclosed herein provide computer-implemented methods and apparatus of determining a treatment for treating a subject's skin or hair. Such a method comprises: receiving, as input, dermatological data indicative dermatological data indicative of the subject's skin or hair following treatment by an initial treatment, the subject's skin or hair having an associated health condition; processing the dermatological data using an condition determination machine learning model to determine an updated health condition of the subject's skin or hair; determining a modified treatment for treating the updated health condition of the subject's skin or hair in dependence on the determined updated health condition of the subject's skin or hair; and providing, as output, the determined modified treatment.

Description

[0001] Computer-implemented method and apparatus for determining a personalised dermatological treatment FIELD OF INVENTION [1] The present disclosure relates to determining a treatment for treating a subject’s skin or hair. Examples relate to an apparatus, systems, computer-implemented methods, and machine-readable media having program code stored thereon configured to perform determination of a treatment for treating a subject’s skin or hair, BACKGROUND [2] There are many skin and hair conditions which may be improved with appropriate treatment. It can be challenging to determine the most appropriate treatment for a particular person. Often a skin or hair specialist will hold a consultation with the person to determine factors about the condition, that person’s lifestyle, demographics, medications including any allergies or sensitivities, other medicalconditions, family history, previous treatments, and preferences, as well as inspecting the skin or hairproblem to be treated, and propose a treatment for the condition. The treatment may provide some benefit to the person and treat their condition, but often, recurring appointments are needed with the specialist to check the efficacy of the treatment, note any side effects, update the person’s personal factors (for example, if they have changed their diet or use a different skin or hair product to before) and possibly change the provided treatment to improve efficacy, reduce any side effects, and accommodate any changes in the person’s personal factors. It will be appreciated that this process disadvantageously can, be time consuming, requires in-person consultations with an expert, and relies on an element of trial and error until a suitable treatment is identified. [3] Examples disclosed herein provide methods and apparatus which can be used to determine a treatment for treating a subject’s skin or hair which overcome one or more problems in the art. STATEMENTS OF INVENTION [4] According to an aspect, there is provided a computer-implemented method of determining a treatment for treating a subject’s skin or hair, the method comprising: receiving, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; processing the dermatological data using a condition determination machine learning model to determine an updated health condition of the subject’s skin or hair; determining a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; andproviding, as output, the determined modified treatment.[5] Determining the modified treatment may comprise: providing an indication of the initial treatment, and providing the determined updated skin or hair condition of the subject, to a treatment determination machine learning model; and obtaining, as output from the treatment determination machine learning model, the modified treatment. [6] The method may comprise determining the initial treatment in dependence on a received descriptor of the subject and an initial determined skin or hair condition of the subject, the initial skin or hair condition of the subject determined using the condition determination machine learning model processing initial dermatological data input indicative of the subject’s skin or hair prior to the treatment by the initial treatment; and providing, as output, the determined initial treatment. The descriptor of the subject may comprise data describing factors of the subject such as demographics, diet, allergies, daily routines, and other relevant information which is not necessarily a direct description of the skin or hair condition to be treated. [7] Determining the initial treatment may comprise: providing the descriptor and the determined initial skin or hair condition of the subject to a treatment determination machine learning model, and obtaining, as output from the treatment determination machine learning model, the initial treatment. [8] The dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment may comprise one or more of: an image of the subject’s skin or hair; and a textual description of the subject’s skin or hair. The textual description of the subject’s skin or hair may be provided by one or more of the subject or a clinician. [9] Determining the modified treatment for treating the updated health condition of the subject’sskin or hair in dependence on the determined updated health condition of the subject’s skin or hair maycomprise: assigning a score to the determined health condition, the score indicative of a severity of the determined health condition; assigning an updated score to the updated health condition, the updated score indicative of an updated severity of the determined health condition; and determining the modified treatment in dependence on a difference between the score and the updated score.

[0010] Determining the modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair may comprise: determining a severity of the updated health condition; comparing the severity of the updated health condition with a severity threshold; in dependence on determining that the updated health condition is equal to or greater than a severity threshold, determining the modified treatment; and in dependence on determining that the updated health condition is below the severity threshold, providing, as output, an indication that the treatment is to stop.

[0011] Determining the severity of the updated health condition may be performed by the condition determination machine learning model when processing the dermatological data to determine the updated health condition of the subject’s skin or hair.

[0012] Determining the initial treatment for treating the health condition of the subject’s skin or hair may comprise: determining a severity of the health condition; comparing the severity of the health condition with a severity threshold; in dependence on determining that the health condition is equal to or greater than the severity threshold, determining the initial treatment; and in dependence on determining that the health condition is below the severity threshold, providing, as output, an indicationthat no treatment is recommended.

[0013] Determining the severity of the health condition may be performed by the condition determination machine learning model when processing the dermatological data to determine the health condition of the subject’s skin or hair.

[0014] Processing the dermatological data using the condition determination machine learning model to determine the updated health condition of the subject’s skin or hair may comprise identifying a change in a property of the subject’s skin or hair identifiable from the dermatological data compared to the property of the subject’s skin or hair identified from the initial dermatological data using the condition determination machine learning model. Determining the modified treatment for treating the updated health condition of the subject’s skin or hair may comprise identifying a change in one or more components of the initial treatment, the one or more components corresponding to the identified changed property of the subject’s skin or hair.

[0015] The condition determination machine learning model may be trained using: a plurality of images of other subjects’ skin or hair and corresponding descriptors of the other subjects; a plurality of textual descriptions of other subjects’ skin or hair and corresponding descriptors of the other subjects dermatological data; synthetic image training data indicative of synthetic subjects’ skin or hair and corresponding descriptors of the other synthetic subjects; and synthetic textual description training data indicative of synthetic subjects’ skin or hair and corresponding descriptors of the other synthetic subjects dermatological data. The computer-implemented method may further comprise determining a DNA methylation status of the subject’s skin to determine the updated health condition of the subject’s skin or hair. The DNA methylation status may be determined through a tape-stripping process.

[0016] The condition determination machine learning model may be trained to identify one or morehealth conditions from: skin hyperpigmentation, skin inflammation, skin dryness, wrinkles, acne, rosacea, skin ageing, hair thinning, and hair loss.

[0017] One or more of the initial treatment and the modified treatment may comprise: a cleanser, a day serum, a moisturiser, a sun-protection-factor day cream, a night serum, hair vitamins, hair growth tonic, a platelet rich fibrin injectable treatment, mesotherapy (e.g. dutasteride), polynucleotides, light treatments (e.g. intense pulse light, photodynamic therapy, red / blue light, green light), laser treatments (e.g. carbon dioxide, fractional and non-fractional resurfacing, pulse dye, ND:Yag, alexandrite, erbium glass lasers), energy based devices (e.g. electromagnetic, radiofrequency including radiofrequency microneedling, ultrasound), microneedling, dermal fillers, botoxilinium toxin, fat dissolving injections, chemical peels and oral medication.

[0018] The condition determination machine learning model may comprise a neural network, for example a convolutional neural network or other neural network-based architecture.

[0019] The treatment determination machine learning model may comprise one or more of: a decision tree algorithm, a random forest algorithm (for example a random forest decision tree algorithm), a gradient boosting algorithm, a large language model, and / or any other appropriate machine learning architecture.

[0020] The method may be performed in the cloud.

[0021] In an aspect there is provided an apparatus configured to determine a treatment for treating a subject’s skin or hair, the apparatus comprising a processor and a memory having computer-readableinstructions stored thereon, wherein the processor is configured to access the memory and execute thecomputer-readable instructions to: receive, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; process the dermatological data using a condition determination machine learning model to determine an updated health condition of the subject’s skin or hair; determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment.

[0022] The processor may be configured to access the memory and execute the computer-readable instructions to: determine the initial treatment in dependence on a received descriptor of the subject and an initial determined skin or hair condition of the subject, the initial skin or hair condition of the subject determined using the condition determination machine learning model processing initial dermatological data input indicative of the subject’s skin or hair prior to the treatment by the initial treatment; and provide, as output, the determined initial treatment.

[0023] The processor may be configured to access the memory and execute the computer-readable instructions to determine the initial treatment by: providing the descriptor and the determined initial skin or hair condition of the subject to a treatment determination machine learning model, and obtaining, as output from the treatment determination machine learning model, the initial treatment.

[0024] The apparatus may be configured to receive the dermatological data of the subject’s skin or hair following treatment by an initial treatment via a web-based user portal.

[0025] The apparatus may be configured to provide, as output, the determined modified treatment toone or more of: a subject’s computing device; and a device of a third party skin or hair practitioner.

[0026] In an aspect there is provided a system comprising: a subject’s computing device; and a processing apparatus in network communication with the subject apparatus; the subject apparatus configured to: receive, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; and provide the dermatological data of the subject’s skin or hair to the processing apparatus: the processing apparatus configured to: receive, as input, the dermatological data indicative of the subject’s skin or hair; process the dermatological data using a condition determination machine learning model to determine an updated health condition of the subject’s skin or hair; determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment.

[0027] The output of the determined modified treatment may be provided to one or more of: the subject’s computing device; and a device of a third party skin or hair practitioner.

[0028] In an aspect there is provided a machine-readable medium having program code stored thereon which, when executed by a computer, causes the computer to perform any of methods disclosed herein.

[0029] In an aspect there is provided a computer-implemented method of determining a treatment for treating a subject’s skin or hair, the method comprising: receiving, as input, initial dermatological dataindicative of the subject’s skin or hair and a descriptor of the subject; processing the initialdermatological data using a trained machine learning model to determine a skin or hair condition of the subject; determining one or more initial treatments for treating the skin or hair condition of the subject in dependence on the determined skin or hair condition of the subject and the descriptor of the subject; and providing, as output, the one or more determined initial treatments; receiving, as feedback input, subsequent dermatological data indicative of the subject’s skin or hair following treatment by the one or more determined initial treatments; processing the subsequent dermatological data using the trained machine learning model to determine an updated skin or hair condition of the subject; determining one or more modified treatments for treating the updated skin or hair condition of the subject in dependence on the determined updated skin or hair condition of the subject; and providing, as output, the one or more determined modified treatments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Examples are further described hereinafter with reference to the accompanying drawings, in which:

[0031] Figure 1 illustrates a method of determining a treatment for treating a subject’s skin or hair according to examples disclosed herein;

[0032] Figure 2 shows an apparatus capable of determining a treatment for treating a subject’s skin or hair according to examples disclosed herein;

[0033] Figure 3 shows an example process of training a machine learning (ML) model to determine a treatment for treating a subject’s skin or hair according to examples disclosed herein;

[0034] Figure 4 shows a process of using a trained ML model to determine a treatment for treating asubject’s skin or hair according to examples disclosed herein;

[0035] Figure 5 shows an schematic process of using a convolutional neural network to identify a condition in an image, as an example of dermatological data, of a subject according to examples disclosed herein; and

[0036] Figure 6 shows a schematic process of using an example of a machine learning model, a random forest algorithm, to identify a treatment for a condition identified in dermatological data according to examples disclosed herein.

[0037] Throughout the description and the drawings, like reference numerals refer to like parts. DETAILED DESCRIPTION

[0038] There are many skin and hair conditions which may be improved with appropriate treatment. It can be challenging to determine the most appropriate treatment for a particular person. Current methods of obtaining a recommendation of a treatment require expertise from a dermatologist, time and expense in having an in-person consultation, and rely on the information available to the expert at the time of consultation as well as the extent and way in which the patient’s history of their condition, treatments, and relevant personal information (e.g. diet, health conditions, other skin or hair products used, and any medication being taken) are accounted for in providing the recommendation. It will be appreciated that this process can, disadvantageously, be time consuming, requires in-person consultations with an expert, and relies on an element of trial and error until a suitable treatment isidentified.

[0039] Examples disclosed herein aim to solve problems with determining a treatment for treating a subject’s skin or hair. Examples provide computer-implemented methods of determining an effective treatment for skin (dermatological) and hair conditions, such as acne, rosacea, hyperpigmentation, skin ageing and hair thinning / loss, through automated analysis of dermatological data of the patient and use of a recommender system to determine an appropriate treatment for a determined condition. Feedback from the patient is provided to the dermatological data analysis and recommender system algorithms to obtain an adjusted, or modified, treatment regimen as the patient’s condition changes over time as it is treated, to optimise the patient’s treatment in a personalised and objective, data driven way.

[0040] A patient’s skin and hair can be assessed using methods disclosed herein for the dermatological conditions. The methods may comprise a self-assessment questionnaire and AI-driven dermatological data analysis of clinical photographs of the patient’s skin or hair and / or textual descriptions describing the patient’s skin or hair . The algorithms used to determine the treatment may be web-based in some examples; for example, the subject may provide their input information (an image of their skin, or a description of their skin or hair condition, for example) through a web portal and the processing to determine an appropriate treatment may be performed elsewhere, e.g. in the cloud. In some examples, the method may include using tape stripping to assess DNA methylation status in the skin and using the DNA methylation status as input to the computer-implemented method to determine an effective treatment.

[0041] The AI-driven dermatological data analysis may generate a severity score for predeterminedconditions such as acne, rosacea, hyperpigmentation, ageing and hair thinning / loss in some examples.Meeting one or more threshold values for such conditions may then trigger inclusion of a particular skin treatment according to the severity score, resulting in a bespoke treatment regimen for the condition(s) identified and their severity. Topical treatment regimens for acne, rosacea, hyperpigmentation and ageing may include recommendation of a cleanser, prescription day serum, moisturiser, SPF day cream and prescription night serum, for example. Skin treatments may include botoxilinium injections, dermal fillers, polynucleotides, fat dissolving injections, light based treatments (e.g. red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (e.g. carbon dioxide, fractional and non-fractional, erbium glass, neodymium-doped yttrium aluminium garnet; Nd:Y3Al5O12 (ND:Yag) and alexandrite, pulse dye lasers), energy based devices (e.g. electromagnetic, radiofrequency and ultrasound), microneedling (with and without radiofrequency), or chemical peels in some examples. Treatments for hair thinning / loss may include hair vitamins, hair growth tonics and injectable treatments (such as platelet rich fibrin, dutasteride mesotherapy, polynucleotides) in some examples. In some cases, oral medications might be recommended.

[0042] For example, an online platform can take medical information from a patient regarding the status of the patient’s skin and hair health, and their overall health. The online assessment can include submission of photographs, such as clinical photographs of the patient’s skin and hair or self-captured photographs by the patient. AI-based strategies as disclosed below have been developed to analyse these photographs for conditions such as acne, rosacea, hyperpigmentation, anti-ageing and hair thinning / loss. The analysis may give a severity score of each of the conditions. In some examples, theanalysis may be supported by tape stripping DNA methylation studies. These DNA results may beprovided to the ML models to be considered as part of the severity scores of each condition. From this information, an appropriate treatment can be recommended.

[0043] Advantageously, through use of the AI-driven dermatological data analysis and treatment recommendation processes disclosed herein, the patient’s skin or hair may be later re-analysed, for example following use of a particular treatment, and thus automatically provide feedback to the methods to obtain a recommended adjustment to the patient’s treatment regimens according to any changes in the patient’s skin or hair condition. This feedback provides an effective and data-supported way for a treatment for a patient to be personalised according to how the treatment is determined to be affecting the patient’s condition. That is, the methods disclosed herein provide objective analysis of the patient’s skin or hair, which does not need in-person clinical expertise to provide the recommendation, and are advantageously able to learn for that particular patient what treatments are effective, and thus tailor treatment over time to optimise how the patient’s skin or hair condition is treated.

[0044] Figure 1 illustrates a computer-implemented method 100 of determining a treatment for treating a subject’s skin or hair. Steps 102, 104, 106, 108 relate to updating an initial treatment. Steps 112, 114, 116 and 118 relate to determining an initial treatment.

[0045] The method 100 comprises receiving, as input, dermatological data indicative of the subject’s skin or hair 120 following treatment by an initial treatment 102. The subject’s skin or hair has an associated health condition. The input may be provided by the subject themselves, for example by uploading a self-taken image to a web portal or otherwise to the software which performs the method 100 when run, and / or providing a text description of the condition or issues with their skin or hair. Thisadvantageously allow the user to have control over the images and information provided to the softwareand to make sure they are happy that the problematic areas of skin or hair are used in the treatment determination, and not necessarily visit a skin or hair specialist to show them the problem area. The input (images and / or text description) may be provided by a third party in some examples, such as by a skin or hair specialist or professional clinical photographer. In examples in which a textual description is provided as dermatological data, this data may be input through the subject or clinician providing a free-text description of the skin or hair condition (e.g. “mild redness on forehead and sides of nose” or “blackheads on the sides of the nose and whiteheads on the bridge of the nose”). As another example, a textual description may comprise responses to a multiple choice type questionnaire to capture data about the subject’s skin or hair. The textual input may indicate an outcome, for example, a description of a perceived result of using an initial treatment on the skin or hair condition.

[0046] In some examples the method may comprise a quality control step whereby, for example if images are provided, the quality of the provided image is checked, and if there is a problem with the image quality e.g. the image is blurry, too light, too dark, or otherwise unclear, the user can be notified and prompted to provide a different image. A quality control measure relating to textual input may be to ensure all questions are responded to, to request a minimum length of textual input (for example until a threshold number of descriptive words have been entered).

[0047] The method 100 comprises processing the dermatological data 120 using a condition determination machine learning (ML) model to determine an updated health condition of the subject’s skin or hair 104. The method may, for example, compare identified features in the dermatological data120 with features in previously provided dermatological data before treatment 122. In some examplesin which an image is provided as input, the image after some treatment 120 and the image taken before treatment 122 may be taken of the same area, for example to determine if the size of a blemish has reduced or see if hair thickness in the region in the images has changed. In some examples, the image after some treatment 120 and the image taken before treatment 122 may not necessarily be taken of the same area, for example to determine if the colour of a skin region exhibiting rosacea has become less red. The method 100 may perform some quality control in comparing the images, for example, determining that a similar light level and light quality is used to capture both images. The method 100 may be able to perform some image correction in some examples to synthetically adjust the image properties such as light levels or colour balance to be the same or similar in both images for better comparison of the properties of the area being treated. In some examples, the updated health condition may not necessarily be determined through comparison of two images or two similar sets of dermatological data. For example, the method 100 may compare a size of an affected skin area determined from a previous image of the patient with a recorded size value stored for that particular patient (e.g. a length in mm of a skin wrinkle, an area in mm2of a red patch of skin, or an average number of blackheads in a cm2). In examples in which a textual description is provided, the method 100 may perform a comparison between indicated severity levels in the before and after data (e.g. descriptive terms indicative of severity, affected area size, and / or symptom may be identified in the text and compared using the ML model, such as “mild redness”, “moderate blackhead presence”, “severe dryness”, “slightly thinned hair”, “very oily skin in a 5cm2region of forehead”).

[0048] The method 100 comprises determining a modified treatment for treating the updated healthcondition of the subject’s skin or hair in dependence on the determined updated health condition of thesubject’s skin or hair 106. For example, the method 100 may access data indicative of a currently used treatment for the area represented in the dermatological data 120, and determine an updated treatment according to the health condition of the subject’s skin or hair as determined from the dermatological data 120. This may be performed using a trained ML model (e.g. a recommender system) which can identify the modified treatment. That is, determining the modified treatment may comprise providing an indication of the initial treatment, and providing the determined skin or hair condition of the subject, to a treatment determination ML model, and obtaining, as output from the treatment determination ML model, the modified treatment.

[0049] As an example, a patient may be using a topical cream twice a day for a skin condition. The method 100 may determine, from the determined updated health condition of the subject’s skin or hair, that a change in treatment regimen is to be recommended, which may be a less concentrated form of treatment, may be less regular treatment (e.g. once a day), may be a change in composition of the treatment, or other change.

[0050] The method 100 then comprises providing, as output, the determined modified treatment. The output may be provided to the subject so that they can see what has been recommended. The output may be provided to the subject’s skin or hair specialist so that a new treatment may be formulated according to the recommendation for the user, and / or for a specialist opinion or quality check of the computer-implemented method generated recommendation. The output may be stored in a patient record for the subject. Advantageously, the output may be used as feedback and input to the ML modelfor use in a future iteration of the method to adjust the treatment in future. Such feedback data, whichmay be called longitudinal data, may be used to inform the ML model as to which patients have continued with a particular recommended treatment(s) and the outcomes of that treatment evidenced in dermatological data of the patient. By incorporating a feedback loop with this information, i.e. the real world effect of a recommended treatment, the ML models may be optimised to be able to recommend treatments which are suitable for the patient and for their condition, and that recommendation is based, at least in part, on evidence from previous long term use of that product providing a desirable effect.

[0051] While the above discussion considers feedback to the described system to determine any changes to the subject’s skin or hair condition and, if appropriate, determine a modified treatment, the method may also be used to determine the initial treatment (that is, before a treatment) for the subject as well in some examples. Thus, as shown in the upper portion of Figure 1, the method 100 may comprise determining the initial treatment 116 in dependence on a received descriptor 124 of the subject and an initial determined skin or hair condition of the subject. The descriptor 124 comprises data which describes one or more aspects of the subject themselves, such as the subject’s demographics, skin type, skin treatments, diet, medical information, any medication being taken, allergies and sensitivities, etc. This descriptor information may be obtained through completion of a subject questionnaire or through an initial interview with a skin or hair clinician, for example. The initial skin or hair condition of the subject may be determined using the condition determination ML model 114 processing initial dermatological data input 122 indicative of the subject’s skin or hair prior to the treatment by the initial treatment. Thus the method 100 may comprise receiving, as input, dermatological data indicative of the subject’s skin or hair 122 prior to treatment 112. The method 100may providing, as output, the determined initial treatment 118, similarly to above in respect of theprovided modified / updated treatment 108.

[0052] Determining the initial treatment 116 may comprise providing the descriptor 124 and the determined initial skin or hair condition of the subject resulting from the processing in step 114 to a treatment determination ML model, and obtaining, as output from the treatment determination ML model, the initial treatment which can then be provided 118.

[0053] Determining the modified treatment 106 for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair obtained in step 104 may comprise assigning a score to the determined health condition, wherein the score is indicative of a severity of the determined health condition. Then, assigning an updated score to the updated health condition may be performed, wherein the updated score is indicative of an updated severity of the determined health condition. Determining the modified treatment may then be performed in dependence on a difference between the score and the updated score. Advantageously, a change in score indicates a change in appearance or severity of the skin or hair condition which is of sufficient magnitude for a recommendation to be meaningful. A very minor change in appearance may not indicate a real or significant change in appearance and not warrant a change in treatment in response to improvement of the condition, for example, when the change is very minor and not a meaningful improvement.

[0054] For example, a skin redness may be scored on a scale from 1 to 5, wherein 1 means no redness, 2 means minor redness, 3 means moderate redness, 4 means strong redness and 5 meanssevere redness. If the dermatological data following treatment 120 provides a score of 3, and thedermatological data before treatment 122 had a score of 4, then this may be understood to mean that the treatment used between the two sets of dermatological data being recorded is working as intended to reduce the redness and the recommendation may be to continue with treatment, or to reduce dose or frequency of treatment application, for example. If the dermatological data recorded following treatment 120 has the same score as the dermatological before treatment 122, then the recommendation may be to try a different treatment on the basis that the current treatment is not working, or may be to try a stronger or more frequent treatment of the same formulation on the basis that the current treatment is not powerful enough, for example. It will be appreciated that these examples are very simplified and the method 100 may consider plural factors in determining a recommendation and not only a score of a single considered factor, including any change in descriptor information (e.g. a change in any other skin treatments, diet, medical information, any medication being taken, stress factors, etc.). Further, the ML model may not identify and consider only a change in severity score between initial and updated dermatological data sets, but may also detect one or more other changes between the dermatological data sets, such as a change of placement on the body of an affected area, and / or the absence or presence of a different feature of the skin or hair (e.g. dry skin may be detected before treatment and not after treatment). The ML model can be trained to advantageously consider plural such factors in an objective and data-driven way, minimising human or experience bias.

[0055] Determining the modified treatment 106 for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair may comprise: determining a severity of the updated health condition; comparing the severity of the updatedhealth condition with a severity threshold; in dependence on determining that the updated healthcondition is equal to or greater than a severity threshold, determining the modified treatment; and in dependence on determining that the updated health condition is below the severity threshold, providing, as output, an indication that the treatment is to stop.

[0056] Similarly, determining the initial treatment 116 for treating the health condition of the subject’s skin or hair may comprise: determining a severity of the health condition; comparing the severity of the health condition with a severity threshold; in dependence on determining that the health condition is equal to or greater than the severity threshold, determining the initial treatment; and in dependence on determining that the health condition is below the severity threshold, providing, as output, an indication that no treatment is recommended.

[0057] For example, a skin redness may be scored on a scale from 1 to 5 as above, and a score of 2 or higher may indicate that treatment is required whereas a score of 1 means that no treatment is recommended. In some examples, the threshold is not necessarily a score but may be another measure of the skin or hair condition to be treated. For example, the condition may relate to areas of skin and a size over a threshold area may indicate treatment is recommended whereas a size below the threshold area may indicate treatment is not recommended. As another example, the condition may relate to e.g. a discolouration, or hair thickness, and the threshold may be a difference in a measure of the colouration, or hair thickness, between the area to potentially be treated and a neighbouring area which is not to be treated. Of course there may be plural considerations which are balanced in the determination and not only a severity of a single factor, and the ML model may be trained toadvantageously consider plural such factors in an objective and data-driven way minimising human orexperience bias. Advantageously, treatment is not prioritised for less severe conditions.

[0058] Determining the severity of the updated health condition may be performed by the condition determination ML model when processing the dermatological data (whether the initial dermatological data or the updated dermatological data) to determine the updated health condition of the subject’s skin or hair. This advantageously allows the condition determination ML model to process the dermatological data consistently and account for areas not requiring treatment and areas potentially requiring treatment in a fair and objective way.

[0059] Processing the dermatological data in step 104 using the condition determination ML model to determine the updated health condition of the subject’s skin or hair may comprise identifying a change in a property of the subject’s skin or hair identifiable from the dermatological data 120 compared to the property of the subject’s skin or hair identified from the initial dermatological data 112 using the condition determination ML model. Determining the modified treatment 106 for treating the updated health condition of the subject’s skin or hair may comprise identifying a change in one or more components of the initial treatment, the one or more components corresponding to the identified changed property of the subject’s skin or hair. For example, a treatment for acne may initially be identified which comprises 0.1% tretinoin and 2% clindamycin, and after treatment, the updated health condition may indicate that the acne is less severe in a way which indicates that a weaker application of tretinoin may be recommended but the application of clindamycin should remain as it was before, thus a recommendation of a treatment comprising 0.05% tretinoin and 2% clindamycin may be recommended.Of course this is a simplified example.

[0060] It will be appreciated that the disclosed ML models may be used to refine an ongoing treatment, and / or to recommend an initial treatment. In some examples the disclosed ML models can both recommend an initial treatment and then refine that recommended treatment following receipt of updated dermatological textual description and / or images of the subject. Thus, disclosed herein is a computer-implemented method 100 of determining a treatment for treating a subject’s skin or hair, the method comprising: receiving 112, as input, initial dermatological data 122 indicative of the subject’s skin or hair and a descriptor of the subject 124; processing 114 the initial dermatological data using a trained ML model to determine a skin or hair condition of the subject; determining 116 one or more initial treatments for treating the skin or hair condition of the subject in dependence on the determined skin or hair condition of the subject and the descriptor of the subject; and providing 118, as output, the one or more determined initial treatments; receiving 102, as feedback input, subsequent dermatological data 120 indicative of the subject’s skin or hair following treatment by the one or more determined initial treatments; processing 104 the subsequent dermatological data using the trained ML model to determine an updated skin or hair condition of the subject; determining 106 one or more modified treatments for treating the updated skin or hair condition of the subject in dependence on the determined updated skin or hair condition of the subject; and providing 108, as output, the one or more determined modified treatments.

[0061] Figure 2 shows an apparatus 200 capable of determining a treatment for treating a subject’s skin or hair as described above. The apparatus 200 comprises a processor 202 and a memory 216having computer-readable instructions stored thereon. The processor 202 is configured to access thememory 216 and execute the computer-readable instructions to: receive, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment 120, the subject’s skin or hair having an associated health condition; process the dermatological data using a condition determination ML model to determine an updated health condition of the subject’s skin or hair; determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment. The input 120 may be received by an input module 204. The dermatological data 120 may be processed using a condition determination ML model in a condition determination ML module 206. The modified treatment for treating the updated health condition of the subject’s skin or hair may be determined in a treatment determination ML module 208. The determined modified treatment which is outputted may be provided by an output module 210. The output module 210 in some examples may provide the output to an output apparatus 214 such as a display screen or printer to output the recommended treatment.

[0062] The processor 202 may be configured to access the memory 216 and execute the computer- readable instructions to: determine the initial treatment in dependence on a received descriptor of the subject 124 (e.g. received as input to the input module 204) and an initial determined skin or hair condition of the subject 122 (e.g. received as input to the input module 204). The initial skin or hair condition of the subject may be determined using the condition determination ML model (e.g. in the condition determination ML module 206) processing initial dermatological data input indicative of the subject’s skin or hair prior to the treatment by the initial treatment. In this example the apparatus mayprovide, as output (e.g. by the output module 210), the determined initial treatment.

[0063] The processor 202 may be configured to access the memory 214 and execute the computer- readable instructions to determine the initial treatment by: providing the descriptor 124 and the determined initial skin or hair condition of the subject 122 to a treatment determination ML model (e.g. the treatment determination ML module 208), and obtain, as output from the treatment determination ML model (e.g. by the output module 210), the initial treatment.

[0064] The apparatus 200 may be configured to receive the dermatological data of the subject’s skin or hair following treatment by an initial treatment 120 (and in some examples prior to treatment 122) via a web-based user portal. This is advantageous for the user to be able to control the provision of dermatological data, such as medical images of their body, to the apparatus 200 from their personal device (e.g. a smartphone or laptop) running a web browser and be able to conveniently benefit from accessing and using examples disclosed herein from their own personal electronic device.

[0065] The apparatus 200 may be configured to provide, as output (e.g. by the output module 210), the determined modified treatment to one or more of: a subject’s computing device; and a device of a third party skin or hair practitioner. This is advantageous for the user to be able to control the receipt of information about their skin or hair condition, and for the results to be provided to the subject’s medical practitioner or dermatologist for preparation or review of the recommended treatment.

[0066] In some examples, there may be a system comprising: a subject’s computing device (e.g. a smartphone or personal computer); and a processing apparatus (e.g. the apparatus 200 of Figure 2) in network communication with the subject apparatus. The subject’s apparatus may be configured toreceive, as input, dermatological data indicative of the subject’s skin or hair following treatment by aninitial treatment (e.g. by upload of textual data and / or image capture by the subject), the subject’s skin or hair having an associated health condition. The subject’s apparatus may then provide the dermatological data indicative of the subject’s skin or hair to the processing apparatus 200. The processing apparatus 200 is configured to receive, as input, the dermatological data of the subject’s skin or hair 120; process the dermatological data using a condition determination ML model to determine an updated health condition of the subject’s skin or hair; determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment.

[0067] It will be appreciated that this disclosure also covers a machine-readable medium, such as memory 216, having program code stored thereon which, when executed by a computer (e.g. a processor 202), causes the computer to perform any of methods disclosed herein. The instructions may be embedded in said one or more electronic processors 202 of the apparatus 200; may be stored in a memory 216, or may be provided as software to be executed in the apparatus 200. The memory 216 may comprise any suitable memory device and may store a variety of data, data structures, and / or instructions thereon. For example, the memory 216 may store instructions for software, firmware, programs, algorithms, scripts, applications that may control or cause suitable apparatus to perform all or part of the methodology described herein. The memory 216 may comprise a computer-readable storage medium (e.g. a non-transitory, non-volatile or non-transient storage medium) that may comprise any mechanism for storing information in a form readable by a machine or electronicprocessors / computational devices, including, without limitation: cloud storage, solid state drive, amagnetic storage medium (e.g. floppy diskette); optical storage medium (e.g. CD-ROM); magneto optical storage medium; read only memory (ROM); random access memory (RAM); erasable programmable memory (e.g. EPROM and EEPROM); flash memory; or electrical or other types of medium for storing such information / instructions.

[0068] It will be appreciated that the term “module” may be considered to mean a hardware module such as a dedicated or shared processor or processors, may be considered to mean a software module such as a section of code, or may be a combination of hardware and software, which is configured to perform the described function. Where there is a discussion of processing taking place on a processor 202, it will be appreciated that there may be a plurality of processors connected to provide the described functionality.

[0069] The apparatus 200 may comprise one or more electronic processors 202 (e.g., a microprocessor, a microcontroller, an application specific integrated circuit (ASIC), Boolean logic circuitry, etc.) that are configured to execute electronic instructions.

[0070] While the examples above are described in relation to an apparatus having a memory 216 and processor 202 for performing the computer-implemented methods disclosed herein, in other examples part of all of the methods may be performed in the cloud or on a distributed computing system.

[0071] Figure 3 shows an example process of training a ML model to determine a treatment for treating a subject’s skin or hair. The ML model in this example comprises a condition determination ML model 302 and a treatment determination ML model 304. The condition determination ML model 302 is ableto be trained to analyse dermatological data and identify descriptive labels for the dermatological datato automatically identify one or more skin or hair conditions of the skin or hair presented in the input dermatological data. The ML model 302 may comprise, for example, a neural network (NN), convolutional neural network (CNN), or other suitable ML model able to analyse input data and provide an indication of skin or hair factors extracted from that input data. CNN as an example are discussed in more detail with reference to Figure 5.

[0072] The treatment determination ML model 304 is a ML model which is able to be trained to provide a recommendation of treatment for the identified skin or hair conditions identified by the condition determination ML model 302. Such recommender ML algorithms include decision trees, random forest algorithm, and other ML recommender systems. Use of the trained ML models 402, 404 is discussed in relation to Figure 4. A random forest machine learning method is discussed as an example with reference to Figure 6. Image analysis may be performed on input images provided as dermatological data, and / or textual / language analysis may be performed on input data provided as textual description dermatological data. The textual description may be provided by the subject, and / or by a physician or dermatological specialist. In some examples the dermatological data may be textual / descriptive and may not include image data.

[0073] The condition determination ML model 302 is to be trained in Figure 3 so that, in use, it is configured to process dermatological data indicative of a subject’s skin or hair to determine an updated health condition of the subject’s skin or hair. In examples it may be used initially (before treatment) to determine an initial health condition of the subject’s skin or hair. That is, the condition determination MLmodel 302 is configured to identify a condition from an dermatological data of the subject with thatcondition. The condition determination ML model 302 may be a neural network such as a convolutional neural network, which is trained with the dermatological data 320, 322 to extract features 312 such as redness, pigmentation, inflammation, dryness, wrinkles, and hair thinness, for example.

[0074] Figure 3 shows that the condition determination ML model 302 may be trained using training data 320, 322, and training data labels 306. The training data 320, 322 may comprise patient-derived images (that is, real-world images) 320, patient-derived textual dermatological descriptions, and / or synthetic training image and / or textual description data 322 (e.g. including publicly available images). The health condition labels 306 may comprise labels to categorise the identified conditions present in the training data images and / or textual descriptions 320, 322. For example, an image 320 depicting the condition “acne” of a patient’s skin may be labelled 306 with “acne”, or more specifically with the nature of the condition, e.g. “whiteheads”, “blackheads”, “acne scarring”, etc. The health condition labels may provide similar labelling data for images and textual descriptions of dermatological data from simulated patients in relation to the synthetic data 322.

[0075] The treatment determination ML model 304 is to be trained to, in use, process the updated health condition information of the subject 312 obtained from the condition determination ML model 302, and recommend a modified treatment 314. In examples it may be used initially (before treatment) to determine an initial treatment 314 for the subject’s skin or hair. That is, the treatment determination ML model is configured to recommend one or more treatments for the subject’s conditions identified by the condition determination ML model 302. Figure 3 shows that the treatment determination ML model304 may be trained using training data descriptions 124 from patient’s regarding their conditions andother personal factors 124 which may affect the recommended treatments. That is, the categorical data (descriptor information) 124 and output 312 of the convolutional neural network 302 is used for training of the treatment determination ML model 304, which may be implemented by a range of algorithms / machine learning architectures including random forest decision tree algorithm, or a large language model(s).

[0076] For example the descriptor 124 information may comprise answers to categorical questions e.g. skin type, skin concerns, diet, demographic information, from the real-world patients who have consented to their data being used to train the AI algorithm. Similar categorical questions responses may be provided as descriptor information 124 from simulated patients in relation to the synthetic data. Furthermore, recommendations 310 from expert sources such as from multiple consultant dermatologists may be provided 306 to the treatment determination ML model 304 to allow the treatment determination ML model 304 to learn what treatments to recommend for a particular identified condition 312. One or more of the initial treatment and the modified treatment 314 may comprise: a cleanser, a day serum, a moisturiser, a sun-protection-factor day cream, a night serum, hair vitamins, hair growth tonic, a platelet rich fibrin injectable treatment, mesotherapy, polynucleotides, oral medication, botoxilinium injections, dermal fillers, polynucleotides, fat dissolving injections, light based treatments (red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (carbon dioxide, fractional and non-fractional, erbium glass, ND:Yag and alexandrite, pulse dye lasers), energy based devices (electromagnetic, radiofrequency and ultrasound), microneedling(with and without radiofrequency), chemical peels in some examples.

[0077] In some examples, the computer-implemented method 100 may further comprise determining a DNA methylation status of the subject’s skin to determine the updated (and / or initial) health condition of the subject’s skin or hair. This DNA methylation status may be provided as part of the descriptor information 124 to be processed with the health condition information of the subject 312 to determine a recommended treatment 314. The DNA methylation status may be determined through a tape-stripping process. Skin changes due to ageing, disease or other conditions are linked to DNA methylation, which is a biological process in which the DNA sequence remains unchanged but the effects of the environment and patient behaviours can be correlated with DNA methylation status. DNA methylation is a epigenetic change. Tape stripping provides a non-invasive method of obtaining patient DNA for DNA methylation status determination. Tape stripping involves removing cell layers of the stratum corneum (outer layer) of the skin (epidermis) using adhesive films. The method may access the DNA methylation status information 124, and / or the categorical descriptor information 124, from a database of the information linked with the patients and their images 320.

[0078] Figure 4 shows the proposed trained ML models 402, 404 in use. The condition determination ML model 402 is trained as in Figure 3, and can take, as input, dermatological data indicative of a subject’s skin or hair (for example, an initial image 122 or an image following treatment 120, and / or a textual description before treatment 122 and a textual description following treatment 120). The trained condition determination ML model 402 determines features 412 from the dermatological data 120, 122, to provide an indication of an (updated) health condition of the subject’s skin or hair. These features412 are provided to the trained treatment determination ML model 404 which uses the features 412 anda patient descriptor 124 (e.g. the patient’s demographics, skin type, skin treatments, diet, etc.) to determine a recommended treatment 414 for the patient. This may be an initial treatment 414 in some examples, and if the patient has already undergone some treatment, that information may be provided in thew descriptor 124 with the feature information 412 so the trained treatment determination ML model 404 identified an updated treatment 414 for the patient. This advantageously allows for a patient’s treatment to be modified by providing the dermatological data 120 and descriptor information 124 about the patient back to the model 402, 404 for an updated personalised treatment to be recommended 414.

[0079] Figure 5 shows an schematic process 500 of using a convolutional neural network (CNN) such as may be used as a condition determination g ML model 402 to identify a condition in dermatological data120, 122 of a subject. CNN may be particularly useful for image analysis applications. Generally, a CNN 500 has an input layer 120, an output layer 120, and one or more hidden layers 504, 506, 508, 510, 514, 516, 518 in between.

[0080] Regarding the feature learning layers 512, these hidden layers 504, 506, 508, 510 perform operations that process the input data 120 to learn features 512 specific to the data. Three of the most common layers are convolution activation or ReLU 504, 508, and pooling 506, 510. Convolution layers 204, 208 process the input dermatological data 120 using convolutional filters, each of which activates certain features from the dermatological data. Rectified linear unit (ReLU) layers 504, 508 speed up model training by mapping negative values to zero and maintaining positive values. In this way “activated” (i.e. useful) features are passed to the next layer. Pooling 506, 510 simplifies the output byperforming downsampling to reduce the number of parameters for the network 500 to learn. Comparedto a traditional NN, a CNN can have shared weights and bias values, which are the same for all hidden neurons in a given layer, so all hidden neurons can detect the same feature in different regions of an image or different portions of a textual description, so the ML model can better manage objects at different locations in an image and / or the same feature described in different terminology in a text description. This can be useful in the application of analysing data representee of skin or hair, as a feature (e.g. a red area, wrinkle, or spot) can be identified in e.g. one location of the image or by being described one terminology, and that learning is applicable to a similar feature in a different image location or described using different terminology (e.g. “spot” vs “blackhead” or “redness” vs “rosacea”).

[0081] The following layers may be called classification layers 520, which operate on the learned fea- ture from the feature learning layers 512. These layers can flatten the data 514, connects all data in a connection layer 516 which provides the classes or labels which may be predicted, and a classification layer 518 (which may use the SoftMax approach) to provide the final classification outputs, or derma- tological data labels 412. The dermatological data labels 412 may, for example, indicate “acne”, “rosacea”, “hyperpigmentation” and “wrinkles”, for example.

[0082] Figure 6 shows a schematic process of using a random forest algorithm to identify a treatment for a condition identified in dermatological data. A decision tree 602a-d is a predictive model which can map an input (e.g. an identified condition and severity level in dermatological dataidentified from the condition determination ML model) to a predicted value (i.e. a recommended treatment or treatments) based on the input's attributes. Decision trees divide and categorise data to reduce the number ofvariables until a classification or categorisation is reached. Example portions of decision trees 602a-dare shown representing, for example, an identification of a particular treatment from an input of features identified from the dermatological data provided by the condition determination ML model. Together, plural such decision trees may be combined to form a single model which is a random forest 604 model, for recommendation making. Random forests operate by ensemble learning, to combine sub-models to provide an overall stronger (less biased or varied) model for recommendation. Each decision tree in the random forest builds on a different subset of data and makes an independent prediction (as indicated by the shaded nodes in the decision trees 602a-d of Figure 6). The final prediction, or recommended treatment, 608 (e.g. a daily cleanser and oral antibiotics) is based on the average, or weighted average, of the individual predictions made by the random forest 604 starting with the input 606 (the dermatological data feature(s) from the condition determination ML model, e.g. red oily skin). It will be appreciated that the example of a decision tree, random forest and large language model is an example of a possible recommender algorithm which may be used and other recommender algorithms may also be used to provide an indication of a recommended treatment.

[0083] Some examples of condition identification and treatment recommendation which may be provided by the described models are now provided:

[0084] Acne

[0085] Acne is becoming increasingly common. It can affect teenagers, younger children and older adults. Acne is characterized by blockage and inflammation of the hair follicles. Acne most commonly affects the face but can affect the chest and back. An associated condition is seborrhoea characterized by excessive oil production on the nose, forehead, and chin. When hair follicle openings (pores) becomeblocked the skin produces open comedones (blackheads) and closed comedones (whiteheads). Whenthese blocked pores become inflamed, they progress to papules, pustules, cysts, and nodules. As the inflamed acne recovers, post inflammatory pigmentation and scarring can be evident.

[0086] There are multiple triggers for acne including hormonal factors, inherited predisposition (acne tends to run in the family), bacterial infection and the use of heavy and occlusive skincare. Acne is usually diagnosed in clinic and treatments for acne include skincare, oral antibiotics, oral antiandrogens, and isotretinoin. The use of chemical peels and light / laser-based therapies can also be helpful in managing acne. In the model examples presented above, if a threshold severity score for acne is reached (that is, if the condition determination ML model 206 identifies the presence of acne and in this case, acne reaching a predefined severity score), this can trigger the recommendation of an acne treatment regimen. This treatment regimen may include a cleanser, an Acne Day Serum, SPF day cream and Acne Night Serum. The regimen may contain oral medications and other skin treatments including polynucleotides, light based treatments (red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (carbon dioxide, fractional and non-fractional, erbium glass, ND:Yag and alexandrite, pulse dye lasers), energy based devices (electromagnetic, radiofrequency and ultrasound), microneedling (with and without radiofrequency), chemical peels in some examples.

[0087] Acne Day Serum is a bespoke, prescription item. It may (according to the particular determination of the treatment determination ML module 208) contain niacinamide (1-15%), salicylic acid (1-5%) and / or zinc sulphate (1-5%). Acne Night Serum is a bespoke, prescription item. It maycontain tretinoin (0.01-0.1%), clindamycin (1-5%) and / or azelaic acid (1-20%). The particularingredients, concentrations of ingredients and the serum base of one or both treatments may be varied according to the severity score determined by the condition determination ML model 206. In some examples, oral medications might be recommended, including antibiotics, spironolactone or isotretinoin.

[0088] One month after a acne specific treatment regimen, re-analysis of acne severity takes place using the model 100. Re-analysis may include repeating taking medical information, taking new clinical photographs, and tape stripping for DNA methylation status. Such data from the one month re-analysis is provided as feedback to the model 100 and an updated acne treatment regimen may be obtained from the model 100, for example including a cleanser, Acne Day Serum, SPF day cream and Acne Night Serum. Oral medication recommendation may also be adjusted. Ongoing re-analysis, feedback and adjustments can be made as required. A significant reduction in severity score can be obtained in three months in this way.

[0089] Rosacea

[0090] Rosacea presents as redness, inflammation or flushing in the central face. Genetic and environmental factors are important in the development of rosacea. Rosacea can be associated with excessive sun exposure and the presence of the hair follicle mite Demodex folliculorum. Flares of rosacea can be triggered by alcohol, stress, exercise, changes in temperature and spicy foods. Patients often report that their rosacea has a detrimental effect on their self-confidence. Rosacea is a clinical diagnosis and treatments indicated by the model 100 may include avoiding triggers, skincare, oral medications and light / laser-based therapies. If a threshold severity score for rosacea is reached, this may trigger a rosacea treatment regimen. This treatment regimen may include a cleanser, RosaceaDay Serum, SPF day cream and Rosacea Night Serum. The regimen may contain oral medications.Rosacea Day Serum is a bespoke, prescription item. It may contain ivermectin (1-5%) and / or vitamin E (1-5%). Rosacea Night Serum is a bespoke, prescription item. It may contain metronidazole (1-5%) and / or azelaic acid (1-20%). The particular ingredients, concentrations of ingredients and the serum base of one or both treatments may be varied according to the severity score determined by the condition determination ML model 206. Oral medications may be recommended, and may include antibiotics, ivermectin or isotretinoin. Other skin treatments may include botoxilinium injections, dermal fillers, polynucleotides, light based treatments (red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (carbon dioxide, fractional and non-fractional, erbium glass, ND:Yag and alexandrite, pulse dye lasers), energy based devices (electromagnetic, radiofrequency and ultrasound), microneedling (with and without radiofrequency), chemical peels in some examples.

[0091] One month after a rosacea specific treatment regimen, re-analysis of the severity takes place using the model 100. Re-analysis may include repeating taking medical information, taking new clinical photographs, and tape stripping for DNA methylation status. Such data from the one month re-analysis is provided as feedback to the model 100 and an updated treatment regimen may be obtained from the model 100, for example including a cleanser, Rosacea Day Serum, SPF day cream and Rosacea Night Serum. Oral medication recommendation might also be adjusted. A significant reduction in severity score can be obtained in three months in this way.

[0092] Hyperpigmentation

[0093] Hyperpigmentation is defined as increase in pigmentation of the skin. Melasma is a cause ofhyperpigmentation that causes light brown pigmentation, usually on the face. It is more common in women, though men can also be affected. It is also more common in darker skin types, where there is often a family history of the condition. Melasma is usually more prominent in the summer months and fades in winter. There are several factors that can cause melasma; genetics, sun exposure, birth control pill, pregnancy and some medications. Melasma is usually diagnosed in clinic and treatments for melasma include avoiding hormonal triggers and sunlight, skincare, oral medications, chemical peels and laser based therapies. If a threshold severity score for hyperpigmentation is reached, this may trigger a hyperpigmentation treatment regimen. This treatment regimen may include a cleanser, Hyperpigmentation Day Serum, SPF day cream and Hyperpigmentation Night Serum. The regimen may contain oral medications including oral tranexamic acid. Other skin treatments may include botoxilinium injections, dermal fillers, polynucleotides, fat dissolving injections, light based treatments (red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (carbon dioxide, fractional and non-fractional, erbium glass, ND:Yag and alexandrite, pulse dye lasers, picolaser), energy based devices (electromagnetic, radiofrequency and ultrasound), microneedling (with and without radiofrequency), chemical peels in some examples.

[0094] Hyperpigmentation Day Serum is a bespoke, prescription item. It may contain tranexamic acid (1-10%), niacinamide (1-20%) and / or kojic acid (1-5%). Ingredients, concentrations of ingredients and the serum base may be varied according to hyperpigmentation severity score. Hyperpigmentation Night Serum is a bespoke, prescription item. It may contain hydroquinone (1-15%) and / or tretinoin (0.01- 0.1%). Ingredients, concentrations of ingredients and the serum base may be varied according torosacea severity score. Oral medication may include tranexamic acid or glutathione.

[0095] One month after the specific treatment regimen, re-analysis of the hyperpigmentation severity may take place. Re-analysis could include repeating taking medical information, clinical photographs and tape stripping for DNA methylation status. Data from one month re-analysis is provided feedback to the model 100 to identify an updated recommendation for the treatment regimen including a cleanser, Hyperpigmentation Day Serum, SPF day cream and Hyperpigmentation Night Serum. Oral medication recommendation might also be adjusted. Ongoing re-analysis, feedback and adjustments are made. Significant reduction in severity score is seen in 3 months.

[0096] Anti-ageing

[0097] The natural ageing process and excessive sun exposure can lead to significant changes in the skin. With age collagen production declines and collagen breakdown increases, leading to loss of volume in the dermis. Reduction in collagen in the dermis also leads to the development of fine lines and wrinkles. Skin ageing can be of cosmetic concern, and can be treated with skincare, injectable treatments, light / laser based therapies and surgery. If a threshold severity score for ageing is reached, this may trigger an anti-ageing treatment regimen. This treatment regimen may include a cleanser, Anti- Ageing Day Serum, SPF day cream and Anti-Ageing Night Serum. The regimen may contain oral medications.

[0098] Anti-Ageing Day Serum is a bespoke, prescription item. It may contain vitamin C (1-20%), and vitamin E (1-5%). Other anti-oxidant ingredients may be used. Ingredients, concentrations of anti- oxidant ingredients and the serum base may be varied according to ageing severity score. Anti-AgeingNight Serum is a bespoke, prescription item. It may contain tretinoin (0.01-0.1%). Ingredients,concentrations of ingredients and the serum base may be varied according to ageing severity score. Oral medication might include oral isotretinoin. Other skin treatments may include botoxilinium injections, dermal fillers, polynucleotides, fat dissolving injections, light based treatments (red / blue light, intense pulse light, photodynamic therapy, green light), laser based treatments (carbon dioxide, fractional and non-fractional, erbium glass, ND:Yag and alexandrite, pulse dye lasers), energy based devices (electromagnetic, radiofrequency and ultrasound), microneedling (with and without radiofrequency), chemical peels in some examples.

[0099] One month after the specific identified treatment regimen, re-analysis of apparent ageing severity may take place. Re-analysis could include repeating taking medical information, clinical photographs, and tape stripping for DNA methylation status. Data from one month re-analysis is feedback as an algorithm look to refresh recommendation for the treatment regimen including a cleanser, Anti-Ageing Day Serum, SPF day cream and Anti-Ageing Night Serum. Oral medication recommendation might also be adjusted. Ongoing re-analysis, feedback and adjustments are made. Significant reduction in the severity score may be seen in three months.

[0100] Hair thinning and loss

[0101] Hair thinning can be devastating for people, especially women. It is also exceedingly common, with approximately 40% of women showing signs of genetic and hormonal related hair thinning (also known as female patterned hair loss) by the age of 50. It is also associated with polycystic ovaries, hormonal contraceptives and therapies. Sometimes this form of hair thinning can be ‘unmasked’following stress related periods of hair shedding. treatment options including topical treatments,hormonal and hair growth tablets (spironolactone, oral minoxidil, finasteride, dutasteride), and platelet rich fibrin (PRF), mesotherapy and polynucleotides.

[0102] Male pattern hair loss is the most common type of hair loss in men, affecting about 50% of men over the age of 50. It reflects a combination of inherited predisposition with increased sensitivity to the effects of dihydrotestosterone that causes scalp hair to become thinner, shorter and lighter in colour until eventually the follicles shrink and stop producing hair. At any time after puberty men can become aware of a receding hairline or hair thinning. Diagnosis is usually straightforward, and there are many effective treatments. Treatments can include hair growth tablets (oral minoxidil), hormonal tablets (finasteride, dutasteride), and platelet rich plasma / platelet rich fibrin (PRF). Less common are the auto- immune and inflammatory causes of hair thinning and hair loss, such as alopecia areata and lichenplanopilaris. Treatment options used for patterned hair loss can be effective for this rare group of auto-immune and inflammatory scalp conditions.

[0103] If a threshold severity score for hair loss or hair thinning is reached, this may trigger a treatment regimen. This treatment regimen with include vitamin supplementation and a hair growth tonic. The regimen may contain oral medications. Hair vitamin regimen may be a unique combination of hair vitamins to support optimal hair maintenance and growth. Optimal concentrations of iron, zinc, vitamin B12, folate, vitamin D and selenium may be chosen by the model 100 and the recommended frequency and concentration of the vitamins may be split to help ensure adequate absorption.

[0104] Hair growth tonics may include minoxidil (1-10%), finasteride (0-1.2%) and / or melatonin (0.1-1%). Ingredients, concentrations of ingredients and the hair growth tonic base may be varied accordingto a hair loss severity score. Oral medication might include oral anti-androgens (spironolactone, finasteride, dutasteride) or oral minoxidil. Injection based treatments may include platelet rich plasma / fibrin, dutasteride mesotherapy and polynucleotides. One month after specific hair loss treatment regimen, re-analysis of hair thinning / hair loss severity may take place. Re-analysis could include repeating taking medical information, clinical photographs and tape stripping for DNA methylation status.

[0105] Data from a later date, for example any period from around one week after start of treatment onwards (e.g. three months) re-analysis is used as feedback for the algorithm 100 to refresh / update the recommendation for the hair treatment regimen including hair vitamins and hair growth tonic. Oral medication and injection based treatments (platelet rich fibrin / plasma, mesotherapy, polynucleotides) recommendation might also be adjusted. Ongoing re-analysis, feedback and adjustments may be made. Significant reduction in severity score can be seen in three months.

[0106] Each feature disclosed in this specification may be replaced by an alternative feature serving the same or equivalent purpose unless otherwise stated. The disclosure herein is not limited to the specific examples disclosed herein. The scope of protection is defined by the appended claims.

Claims

OUI-21813 CLAIMS 1. A computer-implemented method of determining a treatment for treating a subject’s skin or hair, the method comprising: receiving, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; processing the dermatological data using a condition determination machine learning model to determine an updated health condition of the subject’s skin or hair; determining a modified treatment for treating the updated health condition of the subject’s skinor hair in dependence on the determined updated health condition of the subject’s skin or hair; andproviding, as output, the determined modified treatment.

2. The computer-implemented method of claim 1, wherein determining the modified treatment comprises: providing an indication of the initial treatment, and providing the determined updated skin or hair condition of the subject, to a treatment determination machine learning model; and obtaining, as output from the treatment determination machine learning model, the modified treatment.

3. The computer-implemented method of any preceding claim, wherein the method comprises determining the initial treatment in dependence on a received descriptor of the subject and an initial determined skin or hair condition of the subject, the initial skin or hair condition of the subject determined using the condition determination machine learning model processing initial dermatological data input indicative of the subject’s skin or hair prior to the treatment by the initial treatment; and providing, as output, the determined initial treatment.

4. The computer-implemented method of claim 3, wherein determining the initial treatment comprises: providing the descriptor and the determined initial skin or hair condition of the subject to the treatment determination machine learning model, and obtaining, as output from the treatment determination machine learning model, the initial treatment.

5. A computer-implemented method of claim 1, wherein the dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment comprises one or more of: an image of the subject’s skin or hair; and a textual description of the subject’s skin or hair.OUI-21813 6. The computer-implemented method of any preceding claim, wherein determining the modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair comprises: assigning a score to the determined health condition, the score indicative of a severity of the determined health condition; assigning an updated score to the updated health condition, the updated score indicative of an updated severity of the determined health condition; and determining the modified treatment in dependence on a difference between the score and the updated score.

7. The computer-implemented method of any preceding claim, wherein determining the modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair comprises: determining a severity of the updated health condition; comparing the severity of the updated health condition with a severity threshold; in dependence on determining that the updated health condition is equal to or greater than a severity threshold, determining the modified treatment; and in dependence on determining that the updated health condition is below the severity threshold, providing, as output, an indication that the treatment is to stop.

8. The computer-implemented method of claim 7, wherein determining the severity of the updated health condition is performed by the condition determination machine learning model when processing the dermatological data to determine the updated health condition of the subject’s skin or hair.

9. The computer-implemented method of any preceding claim, wherein: processing the dermatological data using the condition determination machine learning model to determine the updated health condition of the subject’s skin or hair comprises identifying a change in a property of the subject’s skin or hair identifiable from the dermatological data compared to the property of the subject’s skin or hair identified from the initial dermatological data using the condition determination machine learning model; and determining the modified treatment for treating the updated health condition of the subject’s skin or hair comprises identifying a change in one or more components of the initial treatment, the one or more components corresponding to the identified changed property of the subject’s skin or hair.

10. The computer-implemented method of any preceding claim, wherein the condition determination machine learning model is trained using one or more of: a plurality of images of other subjects’ skin or hair and corresponding descriptors of the other subjects; a plurality of textual descriptions of other subjects’ skin or hair and corresponding descriptors of the other subjects dermatological data;OUI-21813 synthetic image training data indicative of synthetic subjects’ skin or hair and corresponding descriptors of the other synthetic subjects; and synthetic textual description training data indicative of synthetic subjects’ skin or hair and corresponding descriptors of the other synthetic subjects dermatological data.

11. The computer-implemented method of any preceding claim, further comprising determining a DNA methylation status of the subject’s skin to determine the updated health condition of the subject’s skin or hair.

12. The computer-implemented method of any preceding claim, wherein the conditiondetermination machine learning model is trained to identify one or more health conditions from: skin hyperpigmentation, skin inflammation, skin dryness, wrinkles, acne, rosacea, skin ageing, hair thinning, and hair loss.

13. The computer-implemented method of any preceding claim, wherein one or more of the initial treatment and the modified treatment comprises: a cleanser, a day serum, a moisturiser, a sun- protection-factor day cream, a night serum, hair vitamins, hair growth tonic, a platelet rich fibrin injectable treatment, mesotherapy, oral medication, botoxilinium injections, dermal fillers, polynucleotides, fat dissolving injections, red / blue light treatment, intense pulse light treatment, photodynamic therapy, green light treatment, laser based treatments, energy based devices, microneedling with radiofrequency, microneedling without radiofrequency, and chemical peels.

14. The computer-implemented method of any preceding claim, wherein the condition determination machine learning model comprises a convolutional neural network.

15. The computer-implemented method of any preceding claim, wherein the treatment determination machine learning model comprises one or more of: a gradient boosting algorithm, a decision tree algorithm, a random forest algorithm, and a large language algorithm.

16. The computer-implemented method of any preceding claim, wherein the method is performed in the cloud.

17. An apparatus configured to determine a treatment for treating a subject’s skin or hair, the apparatus comprising a processor and a memory having computer-readable instructions stored thereon, wherein the processor is configured to access the memory and execute the computer-readable instructions to: receive, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; process the dermatological data using an condition determination machine learning model to determine an updated health condition of the subject’s skin or hair;OUI-21813 determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment.

18. The apparatus of claim 17, wherein the processor is configured to access the memory and execute the computer-readable instructions to: determine the initial treatment in dependence on a received descriptor of the subject and an initial determined skin or hair condition of the subject, the initial skin or hair condition of the subject determined using the condition determination machine learning model processing initial dermatologicaldata input indicative of the subject’s skin or hair prior to the treatment by the initial treatment; andprovide, as output, the determined initial treatment.

19. The apparatus of claim 17 or claim 18, wherein the processor is configured to access the memory and execute the computer-readable instructions to determine the initial treatment by: providing the descriptor and the determined initial skin or hair condition of the subject to a treatment determination machine learning model, and obtaining, as output from the treatment determination machine learning model, the initial treatment.

20. The apparatus of any of claims 17 to 19, configured to receive the dermatological data of the subject’s skin or hair following treatment by an initial treatment via a web-based user portal.

21. The apparatus of any of claims 17 to 20, configured to provide, as output, the determined modified treatment to one or more of: a subject’s computing device; and a device of a third party skin or hair practitioner.

22. A system comprising: a subject’s computing device; and a processing apparatus in network communication with the subject apparatus; the subject apparatus configured to: receive, as input, dermatological data indicative of the subject’s skin or hair following treatment by an initial treatment, the subject’s skin or hair having an associated health condition; and provide the dermatological data of the subject’s skin or hair to the processing apparatus: the processing apparatus configured to: receive, as input, the dermatological data of the subject’s skin or hair; process the dermatological data using an condition determination machine learning model to determine an updated health condition of the subject’s skin or hair;OUI-21813 determine a modified treatment for treating the updated health condition of the subject’s skin or hair in dependence on the determined updated health condition of the subject’s skin or hair; and provide, as output, the determined modified treatment.

23. A machine-readable medium having program code stored thereon which, when executed by a computer, causes the computer to perform any of methods 1 to 16.

Images