Software that predicts melanoma: how AI can change prevention

In a research study, Swedish scientists analysed health data on more than six million adults

by Maria Rita Montebelli

5 May 2026

3' min read

Translated by AI

Versione italiana

Key points

Who will get sick in 6 million million?
End of 'blanket' screening?
Not only prediction: AI also improves diagnosis
Possible spillovers from this study
The real paradigm shift: the future of medicine is predictive

3' min read

Translated by AI

Versione italiana

Software capable of predicting, years in advance and with great precision, whether a person is at risk of developing melanoma. This is the new frontier of artificial intelligence applied to prevention, with a view to a medicine that does not chase the disease, but anticipates it in time. And which can predict who will develop a melanoma, up to five years in advance. This is not science fiction, but the results of Swedish research, which could revolutionise skin cancer prevention.

The study, published in Acta Dermato-Venereologica, examined the performance of an artificial intelligence software developed by researchers at the University of Gothenburg (first author, Sam Polesie, Associate Professor of Dermatology) together with the Chalmers University of Technology. And the results certainly do not leave one indifferent.

Who will get sick in 6 million millions?

Swedish scientists analysed health data on more than six million Swedish adults. In this huge sample, approximately 0.64% (or more than 38,000 people) developed melanoma within the five-year survey period.

Machine learning algorithms carefully studied a range of information (age, gender, medical diagnosis, medication taken, socio-economic conditions) in the Swedish registry's huge database. In this way, the AI learnt to recognise hidden patterns, weak but decisive signals that precede the development of melanoma, invisible to the naked eye, but clearly recognisable to the AI's probing eye.

The most advanced AI model achieved an accuracy of about 73% in predicting who would develop melanoma, while a basic model trained on fewer elements (age and gender) stopped at 64%. More educated algorithms, integrating more data, showed a distinct leap in quality, even managing to identify small groups of people at very high risk, in whom the probability of developing melanoma was as high as 33% within five years. That is, 1 in 3 people.

End of 'blanket' screening?

Today, melanoma prevention is based on periodic dermatological checks of moles, often generalised. But this approach, although valuable (it is still the only one available in the clinic), has obvious limitations: it is expensive and time-consuming and leads to many unnecessary examinations (including biopsies).

But with tools like AI, the future could be different. The idea is to use AI as an initial filter to select high-risk patients, direct them to more frequent dermatological check-ups, intervene before the tumour develops or in its early stages. In other words, it could help implement personalised prevention, tailored to the individual patient.

This is the logic of precision medicine applied to prevention.

Not only prediction: AI also improves diagnosis

But the AI algorithm revolution is not limited to prediction. Another research, coordinated by the Karolinska Institutet in collaboration with Yale University, has shown that AI can also make diagnosis more reliable.

In this case, the algorithms focused on the analysis of so-called tumour-infiltrating lymphocytes (TILs), important indicators of melanoma aggressiveness. Put to the test in the field, they produced more reliable assessments and more accurate prognoses. Translated into clinical practice, this could lead to fewer errors in diagnosis and more effective treatment decisions.

The possible fallout of this study

Melanoma is the most aggressive and deadly skin cancer and its incidence is increasing in Western countries. Predictive tools such as those based on AI could help optimise resources, reduce the costs of prevention and diagnosis and, more importantly, increase early diagnosis.

But it is from the authors of this research themselves that comes an invitation for caution: these tools are not yet ready for large-scale clinical use. Crucial knots remain to be unravelled: the AI algorithms will have to be validated on other populations (these are built on Swedish databases) and integrated into the diagnostic flowchart. Then there always remains the stumbling block of ethical issues and data privacy. But the direction now seems to be set.

The real paradigm shift: the future of medicine is predictive

This research comes at a very important historical junction: from reactive to predictive medicine. The basic idea is that it will no longer be necessary to wait for a tumour to appear. A personalised risk profile will have to be constructed, which will allow earlier intervention.

All thanks to well-educated algorithms that will silently read the numerous data that define a person's health profile, warning years in advance that something is not going right. AI could go far beyond early diagnosis, moving us towards intelligent prevention with selective screening.