How AI Skin Checks Work and What They Can (and Cannot) Do
AI skin checks use machine learning to flag moles and rashes that need a closer look. DermaDex's CTO explains how the models work, how accurate they really are, what they cannot do, and how to use one safely in Canada.
Elham Sayyah
Computer engineer, AI/ML
As of January 3, 2026.
Artificial Intelligence (AI) skin tools have moved from research labs onto phones. As the computer engineer who builds these models at DermaDex, I want to explain in plain terms what an AI skin check measures, how close it gets to a dermatologist, and the line it cannot cross. DermaDex builds AI-assisted dermatology tools and connects patients across Canada to certified dermatologists, so this is the question I answer most.
What is an AI skin check?
Short answer: An AI skin check is software that analyzes a photo of your skin and estimates whether a spot, mole, or rash looks concerning, then suggests whether to see a clinician. It is a triage and education aid, not a diagnosis. You point a camera at a lesion, and the tool returns a score or a short list of possible matches with advice such as "monitor" or "see a clinician soon." It helps you decide whether and how quickly to seek care, which matters most for people far from a dermatology clinic. It does not write a diagnosis into your chart, and it does not prescribe.
The output is a probability, not a verdict. Good tools say so plainly. The value is in pointing you toward the right next step before a changing spot has months to grow.
How does an AI skin check actually work?
Short answer: Most AI skin checks run a convolutional neural network (CNN), a type of machine learning (ML) model trained on hundreds of thousands of labeled skin images, to score how closely your photo matches patterns linked to specific conditions. The app first crops and lighting-corrects the image. The CNN then extracts visual features such as color variation, border shape, and texture, layer by layer. Finally it outputs a probability for each condition it was trained to recognize. The model does not see the way a person does; it matches statistical patterns learned from its training set.
The 2017 Nature study by Esteva and colleagues showed a CNN trained on roughly 130,000 images could classify skin cancers at a level comparable to dermatologists, which set the template most tools still follow. The quality of those training images matters as much as the model. To see how clinicians capture the high-magnification images these systems learn from, read our explainer on dermoscopy.
How accurate are AI skin checks compared with dermatologists?
Short answer: In controlled studies the best AI models match or slightly beat dermatologists at classifying single lesion images, but accuracy falls in everyday use with varied lighting, skin tones, and camera quality. Treat any result as a prompt to act, not proof. Laboratory performance on clean, curated images is the best case. On a phone photo taken in dim light, or for a condition or skin tone the model rarely saw in training, the same tool is far less reliable. That gap between published numbers and field results is the single most important thing to understand about these tools.
Two studies anchor the evidence, and a 2019 Lancet Oncology study is one of the largest comparisons to date.
| Study | Year | What it tested | Result |
|---|---|---|---|
| Esteva et al., Nature | 2017 | CNN vs 21 dermatologists on biopsy-proven images | AI reached dermatologist-level classification |
| Tschandl et al., Lancet Oncology | 2019 | 139 human readers vs algorithms on dermoscopy images | Top algorithms outperformed most human readers on the test set |
| Consumer smartphone apps | Varies | Everyday photos across skin types | Real-world accuracy is lower and varies widely by app |
Independent reviews keep finding the same gap: lab numbers look strong, but field performance depends on image quality and on which conditions and skin tones were in the training data.
What can an AI skin check do well?
Short answer: AI skin checks are good at flagging spots that deserve a closer look, teaching the ABCDE warning signs of melanoma (Asymmetry, Border, Color, Diameter, Evolving), and helping people in high-wait regions decide when to seek care sooner. They sort routine-looking spots from ones worth checking, so a changing mole does not get ignored for months. They can track the same spot across dated photos to reveal change over time, which is one of the strongest warning signs. In rural and northern communities, a phone check is available the same day, while a specialist visit often is not.
Where these tools earn their place:
- Triage. They help you decide whether to watch a spot or have it examined.
- Education. They teach the self-exam signs the American Academy of Dermatology (AAD) recommends.
- Monitoring. Dated, repeated photos make change easy to see.
- Access. A same-day check reaches people who cannot quickly book a specialist.
What can an AI skin check not do?
Short answer: An AI skin check cannot diagnose, cannot order a biopsy, and cannot replace a clinician's judgment. It will sometimes miss real cancers (false negatives) and flag harmless spots (false positives). A model only knows what its training data taught it, so accuracy drops for conditions or skin tones that were rare in that data, and many public image sets underrepresent darker skin. A single flat photo also loses information a clinician gathers by touch, patient history, and a magnified dermatoscope view. A reassuring score on a spot that is bleeding, growing, or new is not a green light.
This is why a careful tool routes uncertain or high-risk results to a human instead of closing the loop on its own. The model narrows the question. A clinician answers it.
Can an AI skin check replace seeing a dermatologist in Canada?
Short answer: No. In Canada an AI skin check is a first step that helps you decide whether and how urgently to seek care; only a licensed clinician can diagnose and treat. Access is the real problem the technology helps with, because waits for specialty care are long and the Canadian Institute for Health Information (CIHI) tracks how many patients fall outside benchmark waits. Services covered by the Ontario Health Insurance Plan (OHIP) and other provincial plans still require a licensed physician's assessment for diagnosis. A same-day AI check plus a clear referral path beats an app that pretends to be a doctor.
The CIHI wait-time data explains why patients reach for a phone tool first. The Canadian Dermatology Association (CDA) lists certified specialists, and you can confirm a dermatologist's standing through the CDA. At DermaDex the model triages, then a certified dermatologist reviews flagged cases. You can reach our team through contact.
Is AI skin analysis regulated in Canada?
Short answer: Software that claims to diagnose or screen for disease is regulated as a medical device. In Canada, Health Canada (HC) oversees Software as a Medical Device (SaMD); in the United States the Food and Drug Administration (FDA) plays a similar role. This is why the wording on an app store matters: a tool that says it "detects skin cancer" is making a medical claim and should be licensed. Many consumer "skin analysis" apps call themselves wellness or education products to stay outside device rules, which also means no regulator has checked their accuracy against real cases.
Health Canada publishes guidance on how medical devices, including software, are reviewed and licensed. Privacy is the second question. Your photos are health data, and in Ontario they fall under the Personal Health Information Protection Act (PHIPA). Before you upload, read how the app stores and shares images; our guide on skin photo privacy in Canada walks through what to check.
How should you use an AI skin check safely?
Short answer: Use it as a prompt to act, not as a verdict. Photograph spots in good light, recheck monthly, and book a clinician for anything new, changing, bleeding, or flagged, no matter what the app says. Shoot in bright, even light with the spot in focus and the camera steady. Keep dated photos so you can compare the same spot over months. Act on the ABCDE signs and on your own instinct, not only the score. See a clinician for any sore that will not heal, or a spot that itches, bleeds, or changes shape, size, or color.
For a simple routine, follow the AAD steps on how to check your own skin once a month and after any new spot appears. This article is general information, not a diagnosis. If you are worried about a specific spot, talk to a clinician. To learn how we work, see about DermaDex. For background on a common skin cancer these tools screen for, read our piece on basal cell carcinoma.
What else do people ask about AI skin checks?
Are AI skin checks accurate?
AI skin checks can be accurate in controlled tests and much less so in everyday use. In a 2017 Nature study, a convolutional neural network classified skin cancers at a level comparable to 21 dermatologists, and a 2019 Lancet Oncology study found the best algorithms beat most of 139 human readers on a fixed image set. Those results came from clean, high-quality images. On a phone photo with poor light, an unusual skin tone, or a condition the model rarely saw in training, accuracy drops, and the tool can both miss cancers and flag harmless spots. Read any "accuracy" claim as a best case, and treat a result as a reason to get a real exam, not as a diagnosis.
Can AI analyze my skin?
Yes, AI can analyze a photo of your skin and describe visual features such as color, border shape, asymmetry, and texture, then estimate how closely a spot matches patterns it learned during training. That is what powers most skin analysis apps and the triage step at DermaDex. What it produces is a probability and a recommendation, not a medical finding. It cannot feel the lesion, take your history, or use a dermatoscope, so it works best as a first filter that tells you whether to watch a spot or get it checked. Choose a tool that routes uncertain results to a licensed clinician rather than one that delivers a confident answer with no human review.
Can AI check for skin cancer?
AI can screen images for signs that a spot may be skin cancer, but it cannot confirm it. Only a biopsy, read by a pathologist, can do that. A well-trained model can flag features linked to melanoma and other skin cancers, such as the ABCDE warning signs, and prompt you to seek care sooner. It will still produce false negatives, where a real cancer looks reassuring, and false positives, where a harmless mole looks worrying. In Canada, software that claims to detect cancer should be licensed by Health Canada as a medical device. If a spot is new, changing, bleeding, or will not heal, see a clinician regardless of what an app reports.
Is there an AI that can diagnose skin conditions?
No consumer AI can legally diagnose skin conditions on its own. Diagnosis is a regulated medical act that requires a licensed clinician, and in Canada a tool that claims to diagnose disease must be cleared by Health Canada as Software as a Medical Device. Research models can classify many conditions from images with high accuracy in studies, and clinics use AI as a decision-support aid, but the clinician makes and owns the diagnosis. Consumer apps that suggest a condition are giving you an educated guess to discuss with a professional, not a diagnosis. The safe pattern, and the one we use at DermaDex, is AI for triage and a certified dermatologist for the actual diagnosis and treatment plan.
Sources
- Esteva, A. et al. "Dermatologist-level classification of skin cancer with deep neural networks." Nature, 2017. https://www.nature.com/articles/nature21056
- Tschandl, P. et al. "Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification." Lancet Oncology, 2019. https://pubmed.ncbi.nlm.nih.gov/31201137/
- American Academy of Dermatology. "Skin cancer." https://www.aad.org/public/diseases/skin-cancer
- American Academy of Dermatology. "Melanoma (ABCDE warning signs)." https://www.aad.org/public/diseases/skin-cancer/types/common/melanoma
- American Academy of Dermatology. "How to check your skin." https://www.aad.org/public/diseases/skin-cancer/find/check-skin
- Canadian Institute for Health Information. "Wait times for priority procedures in Canada." https://www.cihi.ca/en/wait-times-for-priority-procedures-in-canada
- Health Canada. "Medical devices." https://www.canada.ca/en/health-canada/services/drugs-health-products/medical-devices.html
- Canadian Dermatology Association. https://dermatology.ca