Image credit: AuthorsAbstract
A significant challenge in the development of Swedish AI capacity is the handling of sensitive data. Succeeding in utilizing sensitive data is a prerequisite for AI to have an impact in healthcare and other sectors of society that handle information worthy of protection or information that can not be shared. At the same time, the development of AI models often requires large amounts of both deep and broad data. The current structures for data storage in Sweden are significantly fragmented and the possibilities for collaborative use of data are limited due to current legislation, even for good analytical purposes. We therefore propose a multi-party project that aims to learn how technology can create the conditions for sharing data in practice. Sahlgrenska Hospital, Region Halland and AI Sweden have decided to lead an effort of applying Federated Learning in practice. For this use case we made use of the SIIM-ISIC 2020 dataset and performed two main tasks. First, melanoma detection, and second, skin lesion image generation.