Develop and validate AI in existing patient processes for skin cancer

Purpose and goal:

The aim of our project was to create decision support using machine learning algorithms for the diagnosis of skin changes during teledermatoscopy. We have created such a decision support which has also been tested in computer experiments and a clinical validation is ongoing. We have struggled with some delays and challenges connected to new guidelines for the validation of medical devices but overall we have reached the aims of the project.

Expected results and effects:

We have created a deep learning artificial intelligence that performs well in laboratory tests and tentatively also in a retrospective validation study. The results indicate that using the AI tool improves the diagnostic accuracy of teledermatoscopy so that the risk of missed skin cancer is reduced as well as the risk of unnecessary visits or sampling.

Expected effect of the project is published articles as well as further steps via further collaborative projects for involved parties and towards productisation. Approach and implementation:

In parallel with the development of the AI tool, a research module was created in the teledermatoscopic platform. A total of 20 dermatologists were recruited from different countries. The cases were then randomly assigned to the assessors.

Information about the patient, the skin lesion and 4 photos were available to the assessors. They then did an assessment on diagnosis, confidence, image quality, and handling, first without and then with access to the AI tool. An interim analysis of the data was conducted and diagnostic test values were calculated for the assessments.