What this Trial Is About

Background: Artificial intelligence has in numerous studies shown high accuracy in detecting skin cancer when trained on various databases of dermatoscopic images. However, there are very few prospective studies conducted in real clinical settings directed at patients seeking healthcare for assessment of skin lesions, and nosuch studies at all in primary care, where the majority of patients are managed. Project aim: To study the accuracy, reliability, and clinical utility of an AI-based decision support system (Dermalyser), developed for primary care, in distinguishing skin cancer from benign lesions. Method: Cluster-randomized controlled trial at approx. 30 primary care centers in Sweden, Germany, Scotland, the Netherlands and Australia. At study start, the participating primary care centres in each country are equally randomised to either be enabled to use the Dermalyser (intervention phase) or to assess patients according to the standard clinical procedure (control phase). When half of the intended sample size (i.e. 1500 of 3000 participants) have been included, the primary care centres switch phase from intervention to control, or vice versa. During the intervention phase, the physicians may use (if found indicated) Dermalyser as a part of their clinical evaluation, whereas during the control phase the physicians follow their ordinary diagnostic routine without support from Dermalyser. This will direct the participants to either an intervention or a control cohort. Both groups will be followed for up to 5 years, with regard to the tumour diagnoses, proportions of skin cancer/benign lesions, and morbidity and mortality in skin cancer. Possible between-group differences will be investigated statistically. Potential benefits: If the Dermalyser prooves to be safe and diagnotically reliability, it could enhance the chance of detecting skin cancer in early stage in primary care, and to reduce the proportion of benign skin lesion unnecessarily excised or referred to dermatologist.

Artificial Intelligent Decision Support for Skin Cancer Diagnostics in Primary Care