Federated Learning to Improve Prostate cancer imaging with Artificial Intelligence - MR Cancer
FLIP.AI
The increasingly important role of medical imaging in healthcare is leading to an overwhelming amount of image information. Artificial Intelligence (AI) has strong potential to help manage these imaging data, by supporting clinicians to make more accurate, reliable, and efficient diagnoses. However, to develop AI solutions that are generally applicable across hospitals, researchers and clinicians must train their models on large and diverse data sets of medical images. These are currently not available, partly because privacy regulations and data ownership concerns make the sharing of sensitive data between hospitals and across borders extremely difficult.
In FLIP.AI, we address this issue with Federated Learning, which is a novel machine learning framework that enables the development of AI on privacy-sensitive images from multiple institutions, but without the need to physically share the data between the institutions. The principle is simple: by bringing the software to the data, instead of the data to the software, Federated Learning can train AI models on patient data from multiple hospitals without compromising privacy.
The objectives of our project are to develop, test, and clinically implement a Federated Learning framework to improve diagnostic prostate cancer imaging with AI. The project builds upon the existing expertise, imaging data, AI solutions, clinical collaborations, and machine learning infrastructure from two large multicentre studies in our group: PROVIZ and 180°N. These include MRI and PET scans of more than 3500 patients from 6 hospitals in Norway, the Netherlands, and Taiwan.
The innovative technology that will be produced in FLIP.AI aims to accelerate the clinical application, validation and acceptance of AI solutions for diagnostic prostate cancer imaging. In the long term, the developed methods will help clinicians provide better care and improve patients' quality of life.
Associate Professor Mattijs Elschot is PI of FLIP.AI, which received funding from the Norwegian Cancer Society and Prostatakreft foreningen for the timeframe 2021 – 2025.
