The skin is visible, and skin lesions can be photographed – ideal conditions for artificial intelligence. Indeed, dermatology is one of the specialties in which AI achieved impressive results early on. For lay people, however, the situation is confusing.
How AI assesses skin lesions
Most systems have been trained on tens of thousands of images of moles and skin tumours, often with images from a dermatoscope (epiluminescence microscope). In the process, they learn patterns such as irregular borders, colour distribution or vascular structures that point to a melanoma, a basal cell carcinoma or a squamous cell carcinoma. Other devices do not analyse the image but measure physical properties of the tissue, such as how light is scattered in the skin.
What the studies show
In comparative studies under laboratory conditions, algorithms achieved an accuracy similar to that of dermatologists. Collaboration is particularly interesting: a study in Nature Medicine (Tschandl and colleagues, 2020) showed that doctors supported by AI made better diagnoses than either human or machine alone – especially the less experienced ones. At the same time, however, they could also be misled by incorrect AI suggestions.
In January 2024, the US Food and Drug Administration (FDA) cleared DermaSensor, the first AI device for detecting the common types of skin cancer in primary care. In the pivotal study with 1,005 patients at 22 centres, sensitivity was 96 per cent. The downside: specificity was low at around 21 per cent – meaning the device frequently raises the alarm for harmless lesions too.
Smartphone apps: big differences
Numerous apps promise to check moles using a mobile phone photo. A systematic review in the BMJ (Freeman and colleagues, 2020) concluded that the evidence for such apps is patchy and that their reliability had not been adequately demonstrated. Problems include poor photos, changing lighting, a lack of data on darker skin types and the risk of false reassurance: an app that reports “unremarkable” can lead to a melanoma being examined too late.
Anyone using an app should check whether it is CE-marked as a medical device and should regard it at most as a supplement – never as a substitute for a medical examination.
Teledermatology with AI
Models in which patients send in photos, an AI carries out pre-sorting and a specialist reviews and signs off the findings are promising. In this way, urgent cases can be identified more quickly and waiting times shortened. Good image quality, clear rules on when an in-person examination is needed and data protection in accordance with the GDPR are crucial.
Warning signs remain important
Regardless of any technology, the ABCDE rule serves as a simple guide: Asymmetry, irregular Border, different Colours, Diameter over about five millimetres and Evolution (change). New, growing, bleeding or itchy skin lesions should be checked by a doctor. In Germany, people with statutory health insurance aged 35 and over are entitled to skin cancer screening every two years.
Conclusion
AI can improve skin cancer detection, above all as a support for doctors. For self-diagnosis via app, the evidence has so far been weak. The best combination remains: paying attention to your own skin, sun protection and specialist advice if anything looks abnormal.
More on this topic: AI in Medicine 2026 – an overview.
Note: This article is for general information only and does not replace medical advice. As of: October 2026.
Sources
- MedTech Dive: Dermasensor gets FDA clearance for AI skin cancer detection device (January 2024)
- Tschandl P et al.: Human–computer collaboration for skin cancer recognition. Nature Medicine 2020
- Freeman K et al.: Algorithm based smartphone apps to assess risk of skin cancer in adults: systematic review of diagnostic accuracy studies. BMJ 2020
- Deutsche Dermatologische Gesellschaft u. a.: S3-Leitlinie Prävention von Hautkrebs
Leave a Reply