Method for identifying skin profiles, each associated with at least one cosmetic skincare product intended for a patient, and its application device.

Abstract

The invention relates to a computer-implemented method for identifying skin profiles associated with cosmetic skincare, comprising the following steps: Obtaining, via a computer system, input data (100) comprising: - patient data (101) representing a patient's assessment; - clinical data (102) representing a clinical assessment by a healthcare professional; - selected image data (103) representing a portion of a patient's skin; - skincare data (104) representing the nature, effects, and conditions of treatments affecting the skin; and its implementation device. Figure for the abstract: Figure 1

Description

Title of the invention: Method for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient and its implementation device

[0001] The present invention relates to a computer-implemented method for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient and the associated device.

[0002] More specifically, the invention relates to a computer-implemented method that uses machine learning artificial intelligence techniques for the accurate identification of an individual's skin profile from a combination of patient personal, clinical and dermatological image information and also makes it possible to recommend personalized skin care based on the identified skin profile.

[0003] Skin health is of paramount importance for the quality of life of individuals. Skin care must be tailored to each individual based on their skin type, skin color and skin condition, specific needs and clinical status. However, accurately identifying the skin profile and recommending appropriate care have long been challenges for healthcare professionals and patients themselves.

[0004] Using a paragraph end Q). When using it (visible if the formatting marks are displayed) the following block will be considered as a new paragraph.

[0005] Furthermore, the method according to the invention makes it possible to establish a correlation between the skin profile and the ingredients, products and / or treatments forming care available on the market. Thus, it offers personalized recommendations to help individuals maintain healthy skin and resolve the dermatological problems that concern them.

[0006] The applicant observed that the method provides a reliable and effective means of identifying individual skin care needs and recommending appropriate solutions.

[0007] A method is known as taught by document WO2021050928A1, using clinical data of the patient to be treated and image analysis technologies allowing the identification of cutaneous events from image data.

[0008] This method has the disadvantage of only providing a diagnostic indication of the condition of a patient's skin and not what the potential treatment(s) would be to remedy the identified skin events.

[0009] Furthermore, this method has the disadvantage of not taking into account potential cross-reactions which could bias the diagnosis and any potential resulting skin care recommendations.

[0010] The present invention overcomes these drawbacks.

[0011] The invention relates to a method for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient, said method being implemented by computer.

[0012] According to a general definition of the invention, the method comprises the following steps:

[0013] -obtain input data comprising:

[0014] - patient metadata which represents the transformation of patient data and clinical data; - image data which respectively represents a part of the patient's skin;

[0015] -skin care data that represents the nature, effects, and conditions of skin care;

[0016] -determining, by means of a first skin classification model, a plurality of integrations respectively for the patient metadata, the image data, and the skin care data, comprising the respective integration of each type of input data by an instantiation of the first skin classification model;

[0017] -learning a second skin profile classification model associated with at least one skin care treatment by respectively processing the patient metadata, the image data, and the skin care data;

[0018] -combine the plurality of integrations to obtain a unified representation associating a patient's skin profile with a treatment or combination of treatments; and

[0019] -determine, using the second classification model, a skin profile and at least one associated skin treatment from input data not included in patient metadata and image data.

[0020] In practice, the skin profile includes classifications belonging to the group formed by skin type, skin class, skin conditions.

[0021] For example, patient metadata is data obtained by transforming patient data and clinical data into Boolean or binary semantic data obtained from an identification of keywords in a patient questionnaire.

[0022] In practice, the image data is raw image data captured under a Wood's Lamp or any similar ultraviolet light source having an emission wavelength of 365 nm plus or minus one nanometer.

[0023] According to an embodiment in accordance with the invention, the clinical data are binary data obtained from the Boolean processing of transcription data. written of a visual and / or tactile clinical assessment by detecting the presence of key words.

[0024] According to one embodiment, the skin care data comprises at least one list of treatments belonging to the group formed by local care, general care, compatibility with skin profile parameters, compatibility between treatments.

[0025] In practice, the first classification model 202 is based on a convolutional neural network.

[0026] Further, the combining step is implemented by processing via at least one FC Layer a unified image representation.

[0027] The invention further relates to a device for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient comprising:

[0028] - processing means configured to execute programmed instructions;

[0029] -data storage means capable of storing instructions which, when they are executed by the processing means, cause the device to perform operations, the operations including:

[0030] -determining, by means of a first skin classification model, a plurality of integrations respectively for the patient metadata, the image data, and the skin care data, comprising the respective integration of each type of input data by an instantiation of the first skin classification model;

[0031] -learning a second skin profile classification model P associated with at least one skin care treatment by respectively processing the patient metadata, the image data, and the skin care data;

[0032] - combine the plurality of integrations to obtain a unified representation associating a skin profile P of the patient with a treatment or combination of treatments; and

[0033] -determine, using a classification model, a skin profile P and at least one associated skin treatment from input data not included in patient metadata and image data.

[0034] Advantageously, the method according to the invention makes it possible to identify a skin profile according to chosen parameters as well as any skin care adapted to said skin profile, and this from data taken from questionnaires completed by the patient and / or a health professional, in combination with associated images.

[0035] The Applicant observed that this identification also makes it possible to limit the potential number of cross-reactions between the different treatments that a patient could use and thus globally adapt the precision of said treatment(s).

[0036] Other advantages and characteristics of the invention will appear on examining the description and the drawings in which:

[0037] [Fig.l] schematically represents the steps of processing the input data of the method according to the invention;

[0038] [Fig.2] schematically represents the implementation device according to the invention;

[0039] [Fig.3] schematically represents the implementation device with the classification results in accordance with the invention;

[0040] [Fig.4] schematically represents a detailed view of the software structure of the method according to the invention;

[0041] [Fig.5] schematically represents a table corresponding to a skin profile of the process according to the invention; and

[0042] [Fig.6] schematically represents a table corresponding to treatments associated with the parameters of a skin profile of the process according to the invention.

[0043] With reference to Figures 1 to 6, the method according to the invention is a first step of preparing the input data 100.

[0044] In practice, the input data includes patient metadata 105, image data 103 and skin care data 104.

[0045] The patient metadata 105 is defined as the transformation into Boolean data 100a of patient data 101 and clinical data 102.

[0046] In practice, the patient data 101 and clinical data 102 can be processed by a transformation of characteristics to generate a representation of characteristics hereinafter called patient metadata 105 associating a Boolean operator (true, false) with each parameter provided by the data so as to be usable during learning.

[0047] By way of non-limiting example, the patient data 101 includes information belonging to the group formed by hereditary history, lifestyle, general health, taking vitamin, mineral or protein supplements, smoking, sleeping habits, stress level, nutritional diet, coffee / tea consumption, sugar consumption, exercise, current skin care, taking medical treatments, health problems, previous cosmetic care or medication or any other data that may come from at least one questionnaire comprising Yes / No type data 102a or semantic data attached to weightings of keywords other than yes / no 102b.

[0048] By way of non-limiting example, the clinical data 102 are written transcription data 102a of a visual and / or tactile clinical evaluation carried out by a health professional or the like.

[0049] For example, the clinical visual data of the clinical data 102 are such as skin color, presence of freckles, texture and surface, deep structure, excess sebum, dilated pores, hyperplasia of the sebaceous glands, photo aging manifested as fine lines and wrinkles, fine lines due to photoaging, dynamic crow's feet, melasma, fine lines due to sagging and photoaging, rosacea and enlarged pores, wrinkles resulting from loss of volume, loss of volume, chin lines and pits, enlarged pores and rosacea, pigmentation spots, vertical wrinkles in front of the ears due to photodamage, photodamage, elastosis, poikiloderma, pale spot under the chin, loss of collagen causing sagging of the skin, photoaging.

[0050] For example, the clinical tactile data of the clinical data 102 are of the type skin elasticity, vascular supply, presence of nodular areas, presence of sebaceous gland retention, masses or bumps.

[0051] The selected image data 103 which respectively represent a part of the skin of a patient, are raw image data captured under Wood's Lamp 103a or any similar ultraviolet light source having an emission wavelength of 365 nm plus or minus one nanometer.

[0052] Advantageously, such image data 103 makes it possible to identify superficial cutaneous pigmentation, porphyrins, fungal conditions, the depth of cutaneous pigmentation and subcutaneous vascular disorders which are not visible to the naked eye.

[0053] In practice, the image data is processed to allow the recording of skin sectors associated with the patient's body.

[0054] With reference to [Fig.6], as an example the skin locations are defined for the face of a patient such that:

[0055] 1. T-zone, the group formed by the forehead and the nose

[0056] 2. Upper eyelids and temples

[0057] 3. Lower eyelids

[0058] 4. Central area of ​​the cheeks

[0059] 5. Upper lip

[0060] 6. Lower lip and chin 7. Cheeks

[0061] 8, 9, 10. Neck and sides of the neck

[0062] 11. Lower central part of the neck

[0063] Finally, the input data 100 comprises the skin care data 104 having at least one list of treatments belonging to the group formed by local care, general care, compatibility with skin profile parameters, compatibility between treatments associated with a skin profile P.

[0064] In practice, skin care is divided into two categories: local care, general care.

[0065] Local care means any care applied to the surface of a patient's skin.

[0066] By way of non-limiting example, local treatments belong to the group formed by: topical creams, gels and ointments, cosmetic needling, gentle peels, sun protection, cleaning, electro-sonic care, cool peels, needling, or any similar care.

[0067] General treatment means any compound or combination of compounds intended to be absorbed by the patient in oral or analogous form.

[0068] By way of non-limiting example, general treatments belong to the group formed by: Additives of the vitamin, peptide, anti-oxidant type, lifestyle changes (high protein, low protein, low sugar, high sugar, low in oilseeds, high in oilseeds, and any similar change having known cutaneous effects.

[0069] With reference to [Fig.5], each skin care element is associated with at least one parameter of the skin profile P, and at least one treatment is associated with a Skin profile P.

[0070] CLASSIFICATION AND LEARNING MODEL

[0071] The training objective of the model according to the invention is the determination of a skin profile P associated with at least one adapted skin care product from patient metadata 105, and image data 103.

[0072] The skin profile (P) includes classifications belonging to the group formed by Skin Class, Skin Type, Skin Conditions.

[0073] Skin Classes are constructed according to three classes A, B, and C.

[0074] In practice, skin classes are defined as class A: Sensitive to light, class B: More tolerant to light, Class C: More resistant to light.

[0075] Each Skin class is defined by patient metadata 105 and image data 103 chosen which are, by way of non-limiting example:

[0076] Class A:

[0077] This group is most at risk for sun damage and skin cancer.

[0078] Metadata 105: Cannot stay in the sun for a long time without getting sunburned, does not tan at all, lightly, Patient with pale skin, blue or pale green eyes and red or pale blond hair, signs of photoaging such as brown spots, wrinkles and freckles.

[0079] Class B:

[0080] This group has a moderate risk of sun damage and skin cancer.

[0081] Metadata 105: Can stay in the sun for some time without sunburn, skin blisters and will peel only if exposed for a long time without sunscreen, tans easily, light brown skin, brown eyes and brown hair, less signs of fine lines than in group A, but may have freckles and dark spots or larger areas of pigmentation, such as melasma.

[0082] Class C:

[0083] This group is at the lowest risk of sun damage and skin cancer.

[0084] Metadata 105: Can stay in the sun longer without burning, skin never blisters or peels, very dark skin, black or dark brown eyes and black hair, potential pigmentation problems, wrinkles.

[0085] The skin profile (P) further includes classifications of skin types belonging to the group formed by Normal Skin, Combination Skin, Dry Skin, Oily Skin, and Red / Sensitive Skin.

[0086] By way of non-limiting example, each skin type is defined as:

[0087] Normal Skin:

[0088] Patient 105 Metadata: No oily skin, no dry skin, few blemishes, barely visible pores, no fine or deep wrinkles and no red tender spots, no unusual bumps or masses, smooth, elastic and well-hydrated skin, fine and firm texture.

[0089] Image Data 103: Wood's Lamp will show that most of the skin is light blue-grey.

[0090] Combination Skin:

[0091] Patient Metadata 105: Dry or normal skin on the cheeks and an oily, shiny T-zone with enlarged pores on the nose, forehead, and chin. There may even be pimples. The cheeks may have fine lines or be smooth, the T-zone is oily, and the side areas are dry or normal.

[0092] Image Data 103: Wood's Lamp will show large purple or light blue-gray areas on the cheeks and yellow on the T-zone.

[0093] Oily Skin:

[0094] Patient Metadata 105: Skin is oily, potential breakouts, enlarged pores and shine, enlarged pores, congested areas of blackheads will not be uncommon, and whiteheads will be apparent, acne pustules may be present. Skin is shiny, and may appear thickened or dull, greasy to the touch, and lumps and bumps under the skin will likely be present, due to blocked sebaceous glands.

[0095] Image Data 103: Wood's Lamp will show large yellow areas. Porphyrins will be pinkish-red and may appear as small spots in acne derived from porphyrins in p.acnes (bacteria).

[0096] Dry skin:

[0097] Patient Metadata 105: Skin feels tight, taut, and rough, with flaking and wrinkles. Skin may be oily, then dry, and return to its original state. Skin may be sore and uncomfortable, with fine lines around the eyes and on the cheeks, and skin may appear whitish and flaky. Skin will appear dull and sometimes grayish, and skin will feel rough instead of smooth.

[0098] Image Data 103: Wood's Lamp will reveal large purple areas and some light white-grey spots.

[0099] Red / sensitive skin:

[0100] Patient Metadata 105: Burning or itching sensations with conventional cosmetic products, the skin may even burn and itch without the presence of cosmetic products, red spots or patches, usually on the cheeks, areas of dryness and itching may be present, potential rashes on the eyelids. The skin may appear fragile, the skin may feel rough because the sensitivity has caused peeling. It may even appear thinned.

[0101] Image data 103: presence of large purple areas.

[0102] The skin profile (P) further includes classifications of skin conditions further including photoaging, acne, pigmentation, rosacea, surface irregularities, poikiloderma of civatte, and redness.

[0103] By way of non-limiting example, each skin type is defined as:

[0104] PHOTO-AGING

[0105] LIGHT PHOTO-AGING

[0106] Patient 105 metadata: sun exposure, freckles on cheeks, nose and arms, few pigment spots, low chance of skin irregularities.

[0107] Image data 103: Sun-induced epidermal pigmentation tends to be dark, superficial, and well-defined on Wood's lamp where it shows up well. Skin pigmentation, on the other hand, is dark (almost undetectable).

[0108] Moderate PHOTO-AGING

[0109] Patient 105 Metadata: Wrinkles, brown spots, white bumps in the skin of the neck, loose skin, Freckles, Pigmentation spots, Fine lines, Smile lines, Dry and ashy skin, Uneven skin color, Loss of skin tone, Dryness.

[0110] Image data 103: Sun-induced epidermal pigmentation tends to be dark, superficial, and well-defined on Wood's lamp, where it shows up well. Skin pigmentation, on the other hand, is dark (almost undetectable).

[0111] Severe PHOTO-AGING

[0112] Patient 105 Metadata: Wrinkles, brown spots, white bumps in the skin of the neck, sagging skin, keratoses and solar lentigines, Dark spots, Static wrinkles, Deep wrinkles, Easy bruising, Elastosis, Sagging, Keratosis, Skin cancer, Dry and rough skin, Poor circulation, Low elastic rebound

[0113] Image data 103: Sun-induced pigmentation of the epidermis tends to be dark, superficial, and well-defined on Wood's lamp where it shows up well. Skin pigmentation, on the other hand, is dark (almost undetectable). Keratoses or thickening of the stratum corneum appear white.

[0114] ACNE:

[0115] Moderate Acne

[0116] Patient Metadata 105: Pimples, scars, pimples, blackheads, whiteheads, mild inflammation, oily T-zone, painful skin

[0117] Image data 103: Under Wood's Lamp, porphyrins will be pinkish-red and may appear as small spots derived from porphyrins in p.acnes (bacteria).

[0118] Severe Acne

[0119] Patient Metadata 105: Pimples, scars, Pimples, Blackheads, whiteheads, solar blackheads, Boils and abscesses, oily T-zone, Pus, Swollen nodules, subcutaneous cysts, scars, Loss of skin tone, Dry and rough, Thick and disordered horny layer, Poor circulation, Low elastic rebound

[0120] Image data 103: On Wood's Lamp, porphyrins will be pink-red and may appear as small spots derived from porphyrins in p.acnes (bacteria).

[0121] PIGMENTATION

[0122] SUN PIGMENTATION

[0123] Patient 105 Metadata: After severe sunburn or after hormonal change, Brown spots, only in sun-exposed areas.

[0124] Image data 103: Sun-induced pigmentation of the epidermis tends to be dark, superficial, and well-defined on Wood's lamp where it shows up well. Keratoses or thickening of the stratum corneum appear white.

[0125] PIGMENTATION - HORMONAL MELASMA

[0126] Patient 105 Metadata: After severe sunburn or after a hormonal change, dark circles around the eyes. Particularly in people with skin type B, pigmentation in the armpits, under the breasts and in the groin.

[0127] Image data 103: None

[0128] PIGMENTATION - POST-TRAUMATIC INFLAMMATION

[0129] Patient 105 metadata: accident or infection, brown and irregular pigmentation at the wound, rough surface.

[0130] ROSACEA

[0131] Patient 105 Metadata: Skin is red, tender and dry on the cheeks, nose, chin and between the eyebrows, Family history of rosacea, Phenomenon occurs when exposed to extreme temperatures, eats spicy food, exercises or drinks alcoholic beverages, Dryness, Redness, Inflammation, Drunk nose, Small pimples, bumps in the red areas

[0132] SURFACE IRREGULARITIES

[0133] SURFACE IRREGULARITIES - BEAUTY MOLES

[0134] Patient Metadata 105: Mole has changed in size, shape, or color, or has become painful or itchy. It may have started to ooze, bleed, or swell.

[0135] Image data 103: brown, tan or black mole, flat or raised, round or oval, uniform in color and less than 6 mm in diameter, abnormal mole identifiable via method A,B,C,D,E: • A for Asymmetry: One half of a mole or birthmark does not match the other. • B is for Edge: The edges are irregular, jagged, serrated or blurred. • C stands for Color: The color is not the same everywhere and can include different shades of brown or black, or sometimes include spots of pink, red, white or blue. • D corresponds to Diameter: The spot has a diameter of more than 6 mm. • E corresponds to Evolving: The mole changes in size, shape or of color. SURFACE IRREGULARITIES - SCARS

[0136] Patient 105 Metadata: Past presence of acne, inflammation, pustules and painful nodules.

[0137] Image data 103: Acne scars: atrophic or hypertrophic, Trauma scarring: variety of scars of different shapes from injuries and burns, Keloid scars: thick, discolored scars that appear on the skin and enlarge over time.

[0138] SURFACE IRREGULARITIES - STRETCH MARKS

[0139] Patient 105 metadata: multiple pregnancies, rapid weight gain and / or growth spurts.

[0140] Image data 103: Silver lines or stripes indicate old stretch marks. Red lines will indicate new stretch marks.

[0141] CIVATTE POIKILODERMA

[0142] Patient 105 Metadata: Sensitive red neck, or pigmentation on the sides of his neck and a white or pale area under his chin, Hyperpigmentation with a rash (pink), Small patches of hypopigmentation on the sides of his neck and lower forehead, sometimes extending to the décolletage, Visible fine blood vessels, roughness and slight inflammation of the surface.

[0143] REDNESS

[0144] Patient Metadata 105: Reaction to a chemical, allergy, red and sensitive skin, with possible raised bumps, rash and scaly areas.

[0145] In practice, all the listed classifications have a number of keywords or singular characteristics alone or in combinations making it possible to classify the input data 100 so that a complete classification according to a Skin profile P is determined.

[0146] The computer-implemented method according to the invention further comprises the following steps:

[0147] According to a first step, determining, by means of a first skin classification model 202, a plurality of integrations respectively for the patient metadata 105, the image data 103, and the skin care data 104, comprising the respective integration of each type of input data 100 by an instantiation of the first skin classification model 202.

[0148] In practice, to generate skin profile classifications P based on image data 103, the skin profile classification model 202 may accept a plurality of images captured of the patient under Wood's lamp as input. Each patient image may be processed by an instance of a convolutional neural network CNN. Each convolutional neural network produces an EMBED embedding representation of each image. The plurality of embeddings may, for example, be averaged to produce a combined embedding representation.

[0149] According to a second step, learning a second classification model 202 of skin profile P associated with at least one skin care treatment by respectively processing the patient metadata (105), the image data (103), and the skin care data (104).

[0150] In practice, the classification model 202 concatenates the combined embedding representation of the image data 103 with the patient metadata 105 into an aggregated unified image representation of the patient metadata 105 and the image data 103.

[0151] According to a third step, combine the plurality of integrations to obtain a unified representation associating a skin profile (P) of the patient with a treatment or combination of treatments.

[0152] In practice, the unified image representation thus obtained is subjected to processing via at least one FC Layer (from the English Fully Connected for fully connected layer).

[0153] FC layer or fully connected layer is a neural network in which each neuron applies a linear transformation to the input vector through a weight matrix. As a result, all possible layer-to-layer connections are present, meaning that every input of the input vector influences every output of the output vector.

[0154] According to a final step, determining, using the second classification model 202, a skin profile P and at least one associated skin treatment from input data not included in patient metadata 105 and image data 103.

[0155] The classification model 202 may use the aggregated unified image representation to generate a differential skin profile classification P.

[0156] The differential skin profile P may comprise a plurality of skin conditions, each with a specific confidence value associated with at least one skin care treatment.

[0157] The invention further relates to a cosmetic skin treatment identification device comprising processing means 200 configured to execute a series of instructions 204 stored in storage means 201.

[0158] In practice, the storage means 201 are capable of storing instructions which, when executed by the processing means, cause the device to carry out operations, the operations comprising:

[0159] -determining, by means of a first skin classification model 202, a plurality of integrations respectively for the patient metadata 105, the image data 103, and the skin care data 104, comprising the respective integration of each type of input data 200 by an instantiation of the first skin classification model 202; -learning a second skin profile classification model P associated with at least one skin care by processing respectively the patient metadata 105, the image data 103, and the skin care data 104;

[0160] - combine the plurality of integrations to obtain a unified representation associating a skin profile P of the patient with a treatment or combination of treatments; and

[0161] -determine, using a classification model 302, a skin profile P and at least one associated skin treatment from input data not included in patient metadata 105 and image data 103.

[0162] Although the subject matter of the present invention has been set forth in detail with respect to various specific exemplary embodiments, each example is provided for the purpose of illustrating, and not limiting, the disclosure. Those skilled in the art, after understanding the foregoing, can readily make modifications, variations, and / or equivalents to such embodiments. Therefore, the This disclosure does not preclude the inclusion of such modifications, variations and / or additions to the subject matter of this disclosure as would readily appear to a person of ordinary skill in the art. For example, features illustrated and / or described as part of one embodiment may be used with another embodiment to give rise to a still different embodiment. Thus, it is intended that this disclosure cover such modifications, variations and / or equivalents.

Claims

Claims

1. A method for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient, said method being computer-implemented and comprising the following steps: obtaining input data (100) comprising: - patient metadata (105) which represents the transformation of patient data (101) and clinical data (102); - image data (103) which respectively represents a portion of the patient's skin; - skin care data (104) which represents the nature, effects, and conditions of skin care; determining, by means of a first skin classification model (202), a plurality of integrations (I) respectively for the patient metadata (105), the image data (103), and the skin care data (104), comprising the respective integration (I) of each type of input data (200) by an instantiation of the first skin classification model (202);learning a second skin profile classification model (P) associated with at least one skin treatment by respectively processing the patient metadata (105), the image data (103), and the skin care data (104); combining the plurality of integrations (I) to obtain a unified representation associating a skin profile (P) of the patient with a treatment or combination of treatments; and determining, using the second classification model (202), a skin profile (P) and at least one associated skin treatment from input data not included in the patient metadata (105) and image data (103).;

2. Identification method according to claim 1, characterized in that the skin profile (P) comprises classifications belonging to the group formed by Skin type, Skin class, Skin conditions.

3. Identification method according to claim 1 or 2, characterized in that the patient metadata (105) are data obtained by transforming patient data (101) and clinical data (102) into binary Boolean or semantic data obtained from an identification of keywords in a patient questionnaire.

4. Identification method according to one of claims 1 to 3, characterized in that characterized in that image data (103) are raw image data (103a) captured under a lamp of Wood or any similar ultraviolet light source with an emission wavelength of 365 nm plus or minus one nanometer.

5. Identification method according to one of claims 1 to 4, characterized in that the clinical data (102) are binary data obtained from the Boolean processing of written transcription data (102a) of a visual and / or tactile clinical evaluation by detecting the presence of keywords.

6. Identification method according to one of claims 1 to 5, characterized in that the skin care data (104) comprises at least one list of treatments belonging to the group formed by local care, general care, compatibility with skin profile parameters, compatibility between treatments.

7. An identification method according to any preceding claim, wherein the first classification model (202) is based on a convolutional neural network.

8. An identification method according to any preceding claim, wherein the combining step is implemented by processing via at least one FC Layer a unified image representation.

9. Device for identifying skin profiles each associated with at least one cosmetic skin care product intended for a patient comprising: - processing means (200) configured to execute programmed instructions (204); - data storage means (201) capable of storing the instructions (204) which, when executed by the processing means, cause the device to perform operations, the operations comprising: determining, by means of a first skin classification model (202), a plurality of integrations respectively for the patient metadata (105), the image data (103), and the skin care data (104), comprising the respective integration of each type of input data (200) by an instantiation of the first skin classification model (202);learning a second skin profile classification model (P) associated with at least one skin care by respectively processing the patient metadata (105), the image data (103), and the skin care data (104); combining the plurality of integrations to obtain a unified representation associating a skin profile (P) of the patient with a treatment or combination of treatments; and determining, using a classification model (302), a skin profile (P) and at least one associated skin treatment from input data not included in patient metadata (105) and image data (103).

Images