Federated learning on cardiac magnetic resonance imaging data. Federated learning on cardiac magnetic resonance imaging data.

The majority of AI algorithms are data-hungry, requiring large databases with annotated labels, which is feasible in typical non-medical computer vision applications (e.g. cat detection in natural images) or with medical images that are easy to acquire (e.g. 2D skin images for cancer detection). This is not the case in cardiology, where it is not straightforward to collect complete databases, robustly annotated by several observers, capturing the large inter-human variability and the complementary of different sources of information. To generate the required large databases, combination of data acquired in multiple clinical sites will be needed, which impose several challenges. It is particularly critical the sharing of patient data due to the compliance of each country GDPR laws, requiring significant resources for a secure and fully anonymized data transmission. Recently, the concept of federated learning has emerged as a technological framework to process data with machine algorithms without compromising privacy and it is starting to make its way in healthcare applications. The main goal of this project is to apply federated learning strategies for the processing of cardiac magnetic resonance imaging (MRI) data such as 4D-flow MRI. This project will be performed in collaboration with researchers at Hospital de la Vall d’Hebron and from the Auckland Bioengineering Institute.

Supervisor: Oscar Camara, Gonzalo Talou (Auckland Bioengineering Institute), Andrea Guala (Vall d’Hebron)