HARMONY Research Project

Group :  Information, Signal, Image and Life Sciences

Labelling: none

Duration: 02/19/2020 - 06/30/2024

Funding: Région Pays de la Loire with Cancéropôle du Grand Ouest (CGO)

Staff involved from LARIS: Pierre Chauvet, Nassib Abdallah (Post-doctoral researcher)

Project partners: M. Hatt (LaTIM - UBO), C. Tauber (INSERM), T. Carlier (CRCINA)

Abstract and objectives

Radiomics is the high-throughput extraction of quantitative features from medical images. Encouraging results have been obtained in predicting patient outcomes in many cancers. Radiomics requires the exploitation of robust learning techniques to select and combine the most useful and least redundant features to develop multi-parametric models. The use of convolutional neural networks (CNNs) is becoming increasingly common in computer vision. Similar to radiomics compared to the use of a few standard image metrics, CNNs have shown impressive performance compared to standard techniques, including in the field of medical imaging. Several studies have already compared conventional radiomics with the use of CNNs, however many challenges remain to be solved, in particular how to combine the two approaches and how the resulting patterns can be exploited to personalise treatments. Furthermore, whether radiomics exploits standard learning approaches or more recent deep learning methods, a major limitation to the development and validation of predictive models that are sufficiently generalisable to be transferred to clinical routine remains the wide variability of scanners, acquisition and reconstruction protocols in the real world multi-centre clinical setting. The main objective of this project is to develop harmonization methods for radiomics and then to adapt and evaluate them in four different contexts as demonstrators. Another objective of the project is to build a longer term collaborative scientific network between 4 teams that have not yet worked together in a project of this ambition.