Industries / Health / Healthcare / Medical spas
Consultation
This is the meeting before any treatment, where the clinic takes a history, photographs the face, measures the skin, screens for reasons not to treat, and gets the person's informed consent.
7
Milestones achieved, from 2001 to 2026.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2001
A questionnaire for dysmorphia
Body dysmorphic disorder (BDD) is a distressing preoccupation with a defect in appearance that others cannot see or find slight, and cosmetic treatment rarely relieves it. R. G. Dufresne, K. A. Phillips and colleagues at Brown University wrote a short self-report questionnaire to screen for it in a cosmetic dermatology practice. Against a clinician's diagnostic interview in 46 people it found every case (sensitivity 100%) and wrongly flagged few others (specificity 93%). Twenty-five years later it is one of the tools Australian plastic surgeons name when asked what they use.
Credit
R. G. Dufresne, K. A. Phillips, C. C. Vittorio and C. S. Wilkel, Brown University
2008
Photo scales for wrinkles
A wrinkle is hard to score by eye in the same way twice. In November 2008 the journal Dermatologic Surgery published a set of 5-point photonumeric scales, where each grade is shown by a reference photograph, for forehead lines, crow's feet, marionette lines, lip fullness, brow position and the backs of the hands, with figures for how well raters agreed. A 2026 study in Türkiye still graded 228 people's wrinkles on "the 5-point Merz scale".
Credit
Alastair and Jean Carruthers of Vancouver and co-authors
2016
The doctor takes consent
The General Medical Council, which licenses doctors in the UK, put guidance for doctors who offer cosmetic interventions into effect on 1 June 2016. A doctor must seek consent from the patient personally, discuss outcomes, benefits and risks, give the patient time to reflect before deciding, take account of the patient's vulnerabilities and psychological needs, and market responsibly.
Credit
General Medical Council
2021
Video calls bring patients in
During the COVID-19 pandemic many people spent hours a day looking at their own face on a video call. Shauna Rice, Arianne Shadi Kourosh and colleagues at Massachusetts General Hospital and elsewhere surveyed 134 dermatology providers across the US: 86.4% said patients cited video calls as a reason to seek care, and 82.7% said patients were more unhappy with their appearance. The authors named it "Zoom dysmorphia", after the earlier "Snapchat dysmorphia" of patients who bring filtered selfies to the consultation.
Credit
Shauna M. Rice, Arianne Shadi Kourosh and colleagues
2022
A skin camera, checked twice
Clinics use booth cameras such as Canfield Scientific's VISIA to count spots, wrinkles and red areas and to track change between visits. Helga Henseler, a plastic surgeon at the Klinik am Rhein in Düsseldorf, set out to check the camera independently of its maker. Henseler photographed volunteers three times in a row and again after one to four weeks. Scores for texture, UV spots, brown spots and porphyrins differed by less than 2%, and spots and wrinkles by about 6% (8 volunteers, 2022); wrinkle scores varied about 3% as absolute scores but about 9% as percentiles (19 volunteers, 2023).
LessonHenseler concluded that absolute scores should be preferred over percentiles.
Credit
Helga Henseler
2025
A consultation before each prescription
From 2 September 2025 Australia's health practitioner regulator, Ahpra, requires registered health practitioners such as nurses who perform non-surgical cosmetic procedures to follow new guidelines. A prescriber must hold an in-person or video consultation every time a cosmetic injectable is prescribed, and prescribing by text, email or online form is not acceptable. The patient's reasons and motivations must be discussed, consent must be talked through as well as written down, and people under 18 must wait seven days between the first consultation and any procedure.
LessonWhen Australia made psychological screening compulsory for cosmetic surgery in 2023, surgeons surveyed a year later rated the screening tools 3.2 out of 10 for accuracy, so a required step is not yet a working one.
Credit
Australian Health Practitioner Regulation Agency (Ahpra)
2026
AI grades wrinkles, not change
Dermatologists at Etlik City Hospital in Ankara, Türkiye, compared Haut.AI Face Skin Metrics 2.0, a commercial skin-analysis program, with four human raters on before-and-after photographs of 228 people treated with botulinum toxin for upper-face wrinkles. The program agreed closely with the humans when grading wrinkles, but the size of the improvement it measured correlated only weakly to moderately with the humans' judgment.
LessonThe authors write that a gold standard for measuring treatment effect is still needed.
Credit
Kadir Küçük, Selda Pelin Kartal and colleagues, Etlik City Hospital
Watch for
AI previews of results
Zenoti, a US maker of software for medspas, says its AI Treatment Visualizer shows a patient their own projected result at early, mid and full stages from one photo, and that such previews could lift consultation conversion, the share of consultations that turn into booked treatment, by about 10–15% in its own estimate. It quotes the imaging company Crisalix's figure of conversion rising from about 50–60% to over 85% across 229,000 consultations.
Achieved whenan outside study compares previews with photographs of the same patients' actual results and publishes how close they were.
Claimed by
Zenoti, July 2026
A questionnaire for filtered selfies
Paolo Mancin, Silvia Cerea and colleagues built the Snapchat Dysmorphia Questionnaire to measure the wish to look like one's digitally altered selfies, tested it on 163 women seeking aesthetic treatments, and found it held together as one scale and tracked BDD symptoms.
Achieved whena second group validates it against a diagnostic interview in another clinic population.
Claimed by
Mancin, Cerea and colleagues, Aesthetic Surgery Journal, January 2026
Open
01
Previews checked against results
A public test of the AI tools that show patients a predicted result before injections. Take before-and-after photo pairs that clinics have already taken in routine care, with each patient's written consent for research and an ethics approval, and note the product, amount and area treated. Feed each "before" photo and the treatment record into each preview tool, then compare the preview with the real "after" photo. No one is treated for the study.
The briefCredit · unclaimed02
Consent forms, read and scored
An open collection of the consent and intake forms that medspas publish on their websites, each scored against a checklist drawn from the GMC's 2016 guidance and Ahpra's 2025 guidelines: who will treat, named serious risks such as blindness from filler, a question on motives or BDD screening, time to reflect, cost, and how to complain.
The briefCredit · unclaimed03
Skin scores against known marks
A test of skin-analysis apps and clinic tools on a target whose marks are known: a life-size face model or print carrying dots of measured size and colour as "spots". Each tool scans it repeatedly with several phones under set lighting, so the spread of its scores and its error in counting spots can be measured. No person is scanned.
The briefCredit · unclaimed
Open 01 · the brief
Previews checked against results
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of the AI tools that show patients a predicted result before injections. Take before-and-after photo pairs that clinics have already taken in routine care, with each patient's written consent for research and an ethics approval, and note the product, amount and area treated. Feed each "before" photo and the treatment record into each preview tool, then compare the preview with the real "after" photo. No one is treated for the study.
Why it is still open
EntityMed, whose simulator BeautyFix Medspa launched in 2022, says it was "tested on more than 10,000 patients" but published no accuracy figures, and Zenoti calls its own previews "a projection, not a guarantee". A search found no published study that compares an injectable preview with the patient's actual result; the 2026 Turkish study measured real photographs only. The public before-and-after photos are few: a sample of 23 Botox pairs that the data seller UniqueData posted on Hugging Face under a non-commercial, no-derivatives licence, and a copy reformatted for training an image-editing model.
Built with what exists
Commercial preview tools such as EntityMed, Perfect Corp's aesthetic simulation and Zenoti's visualizer; Google's open MediaPipe Face Landmarker to measure distances such as lip height and cheek projection; a blinded rating form in which injectors pick the real result out of the preview, the real result and another patient's result.
First step
Write the protocol and data agreement, and ask one clinic for 20 consented, de-identified photo pairs to run through two tools.
Done looks like
At least 200 photo pairs from at least three clinics, run through at least three tools, with the landmark error in millimetres and the blinded raters' pick rate published per tool and per area treated.
Open 02 · the brief
Consent forms, read and scored
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open collection of the consent and intake forms that medspas publish on their websites, each scored against a checklist drawn from the GMC's 2016 guidance and Ahpra's 2025 guidelines: who will treat, named serious risks such as blindness from filler, a question on motives or BDD screening, time to reflect, cost, and how to complain.
Why it is still open
No systematic study of medspa consent forms was found. Pabau, a company that sells clinic software, says it checked marketing photo release forms published by medspas and found clauses missing on every one, but it gives no sample size or method.
Built with what exists
Web search and a crawler to collect PDF forms; a small language model to pull each checklist item out of each form, such as Meta's Muse Ember, which is built for extraction and long document parsing (Notes: "The jobs agents repeat"); a hand check of a sample by two readers; a public spreadsheet.
First step
Collect 30 forms from medspa websites in one US state and score them by hand to fix the checklist.
Done looks like
At least 300 forms from at least three countries, each linked to its source and scored on every item, with the language model's scores matching the two hand readers on at least 95% of items in a 50-form sample.
Open 03 · the brief
Skin scores against known marks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of skin-analysis apps and clinic tools on a target whose marks are known: a life-size face model or print carrying dots of measured size and colour as "spots". Each tool scans it repeatedly with several phones under set lighting, so the spread of its scores and its error in counting spots can be measured. No person is scanned.
Why it is still open
Which?, the UK consumer group, had 30 people run more than 120 scans on one retail tool in September 2026. Back-to-back scans in good light barely moved, but changing the light swung scores by up to +39 or −25 points for about half the testers, and testers judged accuracy against their own mirror. VISIA's precision has been measured only on volunteers in a fixed booth. No test against known marks was found.
Built with what exists
A silicone face model or a life-size print on matte paper; printed dots of known diameter and colour from a colour-managed printer; an X-Rite ColorChecker card; dimmable LED panels and a lux meter; three or more phones on a tripod; the apps and web tools that clinics and retailers offer.
First step
Check which of five common tools will accept a scan of a face model or print, then run 10 scans on the ones that do.
Done looks like
At least 10 tools, three phones and four lighting set-ups, 10 scans each, with each tool's score spread and spot-count error against the known count published in one table.