Automated learning from clinical images using AI can improve cancer screeninging, detection methods and treatment outcome prediction. We expect the insights of this study will timprove oncological decision support for patients and health care professionals.
The aim of AMICUS is to develop and utilize technology for distributed deep learning on medical images. The study will be conducted in the Netherlands using the Personal Health Train (PHT) approach. We aim to demonstrate the value of distributed deep learning on medical images for cancer care organizations and patients by showing convincing examples..
Medical imaging is the cornerstone of cancer screening, diagnosis, staging, treatment and follow-up for almost every cancer patient. However, the processing and interpretation of images is still mainly done by humans. This is time consuming and the quality of the imaging observation relies on the expertise and experience of the physician.
With AMICUS we will develop technology that supports radiologists and oncologists to extract relevant information from images.
The technology used for extracting information from images in AMICUS is called ‘deep learning’. This is a form of artificial intelligence which is a proven breakthrough in the field, but requires large volumes of imaging data to be successful. Getting access to imaging data is a problem as it is dispersed across hospitals and is very privacy sensitive.
In AMICUS we will develop technology that allows deep learning from these distributed imaging datasets where data can remain at the hospital. AMICUS will leverage the previously developed PHT approach which allows privacy-preserving learning from distributed FAIR clinical data and extend it with FAIR imaging data and deep learning.
Maastricht University, Tilburg University, UMC Groningen, University of Twente, IKNL, Varian, Medical Data Works