In the field of medical imaging, radiomics is a technique that proposes to numerically study image features. Those features outline information present in medical images better than a simple visual analysis. The great advantage of using radiomics is that, combined with classical clinical features, it can lead to a new, personalised medicine in the treatment of cancer. In order to process images, the techniques often use artificial intelligence, machine learning and, more specifically, deep learning. However, with those algorithms, two challenges arise: 1) providing sufficient annotated, reliable image data; 2) secure and anonymous handling of the medical images and their associated sensitive meta-data.

Project description

The big picture is to train deep learning models with data coming from hospitals from all around the world. Combining data from many hospitals helps because we are facing a chain where many data are necessary. Nevertheless, reliability of the results as well as guarantee for patient’s privacy are really critical. The idea behind this project is to study how to :

  • build a chain of trust between the hospital providing the data and the deep learning system using the data;
  • implement a rewarding mechanism to entice hospitals in providing many and reliable data.

And all this while still being compliant with the GDPR rules for protecting patient’s privacy.

Project goals

This project aims at exploring models such as Directed Acyclic Graphs (DAGs) and blockchain in order to address the issues presented above.

Project leaders

  • Sébastien Lugan (sebastien.lugan@uclouvain.be)
  • Paul Desbordes (paul.desbordes@uclouvain.be)

Project organisation

Milestone I (week 1): definition of the general system specifications
Milestone II (week 2): investigation of distributed ledgers, automated quality evaluation
Milestone III (week 3): integration of a complete prototype (PoC)
Milestone IV (week 4): user tests and validation