Project 8: Development of an automated triage system for the diagnosis of skin cancer

Background

Early diagnosis of skin cancer is extremely important because if melanoma is detected in early stages, it can be cured by a simple surgical procedure. It is known that Switzerland has the highest incidence of skin cancers worldwide after Australia and New Zealand. The problem is that due to the rising incidence of skin cancers, in particular melanoma, we will be confronted with an epidemic of skin cancers that dermatologists will not be able to handle with the current procedures during routine consultation. For this reason, there is a clear need to address this problem. Since almost all skin cancers are located on the skin surface, they are easily accessible to a simple noninvasive exam called dermoscopy. This is the current standard of care for early detection of skin cancers. It has been shown that Artificial Intelligence (AI) algorithms, using such images, can perform equally or even better than dermatologists in certain classification tasks. These results suggest that it is possible to provide the dermatologists with AI-based tools that could provide invaluable inputs during the diagnosis process.

Working hypothesis

  1. Using diagnosis-aid tools based on AI algorithms in the large scale screening for skin cancers will be a valuable triage tool for both, dermatologists and general practitioners.
  2. The design criteria followed to maximize the acceptance of such tools by the community and optimize its usefulness are: acquisition of real world data plus interpretable algorithms.

Specific aims for this research project

The aim is to develop a framework of AI algorithm that ultimately could provide valuable input to the user during routine Dermatology consultation. In order to achieve this we should:

  1. Acquire and curate of a real world high quality polarized and dermoscopic images dataset.
  2. Develop a multi-purpose framework of deep learning algorithms including detection and extraction of lesions, classification into clinical categories and feature explanation.
  3. Deploy such AI based diagnosis-aid tools during routine consultation.