AI identifies early risk patterns for skin cancer

The research is based on large-scale registry data covering more than six million adults in Sweden. By analysing information such as age, sex, diagnoses, medications and socioeconomic factors, the researchers trained machine learning models to predict who would develop melanoma within five years.

"This study shows how existing healthcare data can be used in new ways to better understand risk patterns and support earlier detection," says Martin Gillstedt, doctoral student at the University of Gothenburg's Sahlgrenska Academy and statistician at Sahlgrenska University Hospital.

Chalmers contribution strengthens AI modelling

Researchers at Chalmers contributed expertise in machine learning and data analysis, enabling the development and evaluation of more advanced predictive models.

"A prediction tool trained on Swedish registry data could serve as a valuable screening instrument for identifying individuals at higher risk of developing melanoma within a given time frame. This could help healthcare systems allocate resources more efficiently and work more proactively," says Lena Stempfle, researcher at the Department of Computer Science and Engineering, Chalmers.

The researchers compared different AI models to find the most accurate way to predict melanoma risk. The best results came from a model that combined several types of data, including medical and socioeconomic information.

Towards more targeted screening

Using this approach, the researchers were able to identify small groups with significantly higher risk - in some cases around a one-in-three probability of developing melanoma within five years.

"Our analyses suggest that selective screening could complement clinical assessments and contribute to more precise and efficient care," says Sam Polesie, Associate Professor at the University of Gothenburg and dermatologist at Sahlgrenska University Hospital.

Melanoma incidence has increased in many Western countries in recent decades, placing growing demands on healthcare systems. The researchers therefore see strong potential in using national registry data together with AI to support earlier detection and more personalised care. At the same time, they emphasise that further research and policy decisions are needed before such tools can be implemented in routine healthcare.

The study is published in Acta Dermato-Venereologica.

Tags: