Point your phone at a mole, tap a button, and within seconds an app tells you whether the spot is "low risk" or "high risk". It is an appealing idea: skin cancer is common, dermatology appointments can be hard to get, and artificial intelligence (AI) has matched specialists in some image-recognition studies. But an app that tells you a melanoma is harmless can do more damage than no app at all. The evidence on consumer skin-cancer apps is thinner, and the regulatory safeguards weaker, than their marketing suggests.
What the evidence shows
The most thorough independent assessment remains a systematic review by Freeman and colleagues, published in the BMJ in 2020. They found nine studies evaluating six apps. The studies were small and methodologically weak: participants were selectively recruited, many images could not be evaluated, and — importantly — the photos were taken by clinicians, not by ordinary users at home.
What do these numbers mean? A sensitivity of 80 % implies that about one in five malignant or premalignant lesions was labelled as low risk. A specificity of 78 % implies that about one in five harmless lesions triggered an alarm. And these figures come from favourable conditions; the authors expected real-world performance, with users photographing their own skin, to be lower. Their conclusions were blunt: current apps cannot be relied on to detect all melanomas or other skin cancers, and the CE-marking process did not adequately protect the public.
The CE-mark problem
In Europe, a CE mark signals that a product meets regulatory requirements — not that it has been shown to be accurate in the hands of patients. Under the old EU rules, most diagnostic software was treated as low-risk class I, which manufacturers could largely self-declare. The EU Medical Device Regulation (MDR) tightened this: under its "Rule 11", software that provides information used for diagnostic decisions is at least class IIa, and class IIb or III if wrong decisions could cause serious or irreversible harm. Higher classes require assessment by an independent notified body. SkinVision, for example, reported receiving MDR class IIa certification in 2025. That is progress, but certification still does not answer the question that matters to you: how often does the app miss a melanoma when you photograph your own skin?
AI in a dermatologist's hands is different
Much of the excitement about AI in dermatology comes from research in which algorithms analyse standardised dermatoscopic images — magnified, polarised-light pictures taken by trained staff. That is a different situation from a phone snapshot in a bathroom.
Consumer app
You choose which spot to photograph, often under poor light and without magnification. The result replaces, rather than supports, a clinical examination — and nobody looks at the rest of your skin.
AI in clinical use
Standardised dermatoscopic images, a full-body examination, and a clinician who can weigh the AI result against the history and the lesions you didn't notice.
An international study in Nature Medicine in 2020 tested how clinicians work together with AI. Good-quality AI support improved diagnostic accuracy beyond what either the AI or the physicians achieved alone, and the least experienced clinicians benefited most. But the same study showed the risk: faulty AI misled clinicians at every level of experience, including experts.
Regulators are starting to allow more. In February 2025, DERM, an AI system used in healthcare pathways, received a class III CE mark under the MDR, allowing it to clear some lesions as non-cancerous without a doctor reviewing them. The manufacturer reports very high accuracy for ruling out cancer; independent long-term data are still accumulating. Such systems operate within healthcare services, with standardised imaging and follow-up — not as free consumer apps.
The danger of false reassurance
The harms of an app are asymmetric. A false alarm costs a doctor's visit and some worry. A false "low risk" can delay the diagnosis of a melanoma, where thickness at diagnosis largely determines survival. The most dangerous melanomas are also the hardest for both apps and the ABCDE rule: nodular melanomas grow quickly, can invade deeply within months, and are often symmetrical, evenly coloured and frequently not pigmented at all. Australian dermatologists proposed a separate warning rule for them: EFG — elevated, firm and growing for more than a month.
What self-examination can do
Looking at your own skin regularly is free and sensible. The classic ABCDE checklist — asymmetry, irregular border, colour variation, diameter over 6 mm, evolution — was extended in 2004 to include "E" because change over time is one of the most important warning signs. Your phone is genuinely useful here: not as a diagnostician, but as a camera to document moles so you can compare them over months.
What this means for you
- Do not use an app to decide that a spot is harmless. A "low risk" result should never stop you from seeing a doctor about a lesion that worries you.
- See a doctor promptly for any mole that is new, changing, bleeding, itching, or simply looks different from your others — and for any firm, raised bump that keeps growing for more than a month.
- Check your skin about once a month, including back, scalp and soles; ask a partner to help, and use photos to track changes.
- If you have many moles, previous skin cancer or a family history of melanoma, ask for an examination with dermatoscopy — see why dermatoscopy beats the naked eye.
- Ask whether any AI used in your care supports the clinician or replaces them, and who reviews the result.
In Germany, people with statutory health insurance have been entitled to a skin cancer screening every two years from age 35 since 2008. In October 2026, the Federal Joint Committee (G-BA) began a legally required review of this programme, including whether a risk-based approach or a different interval would be better, and commissioned IQWiG to assess the evidence; the review is due by the end of 2027. Suspicious skin changes can be examined by a doctor at any time, independent of the screening.
Bottom line
Consumer skin-cancer apps miss a meaningful share of cancers, were tested under unrealistically good conditions, and do not replace an examination. Use your phone to document changes, not to rule out cancer. AI can improve accuracy when used by trained clinicians with dermatoscopic images, but it can also mislead. Evidence: — Not recommended (consumer apps to rule out skin cancer); ★ Emerging (AI support in clinical dermatology).
References
- Freeman K, Dinnes J, Chuchu N, et al. Algorithm based smartphone apps to assess risk of skin cancer in adults: systematic review of diagnostic accuracy studies. BMJ. 2020;368:m127. doi:10.1136/bmj.m127
- Tschandl P, Rinner C, Apalla Z, et al. Human-computer collaboration for skin cancer recognition. Nat Med. 2020;26(8):1229-1234. doi:10.1038/s41591-020-0942-0
- Abbasi NR, Shaw HM, Rigel DS, et al. Early diagnosis of cutaneous melanoma: revisiting the ABCD criteria. JAMA. 2004;292(22):2771. doi:10.1001/jama.292.22.2771
- Chamberlain AJ, Kelly JW. Nodular melanomas and older men: a major challenge for community surveillance programs. Med J Aust. 2004;180(8):432.
- Skin Analytics. DERM receives Class III CE mark. Press release, 6 February 2025.
- Gemeinsamer Bundesausschuss (G-BA). Hautkrebs-Screening und Check-up: G-BA berät über das aktuelle Früherkennungsangebot. Pressemitteilung, 15 October 2026.
This article is general information and not individual medical advice. Decisions about tests and treatment belong in a conversation with your own physician.