Philippe Lambin

Maastricht University

In short, the research of Philippe Lambin entails the following:

Explainable Foundation Model-Driven Prognostic Modeling in High-Grade Serous and Rare Ovarian Cancers for Personalized Treatment

Ovarian cancer includes high grade serous disease and several rarer histological types that can behave differently but are often treated similarly. This project will develop an explainable AI tool that predicts key molecular traits and outcomes directly from routine CT scans, supporting more personalized treatment across ovarian cancer subtypes. 

We will train an ovarian focused CT foundation model using self-supervised and multi-task learning, then fine tune it on multi center datasets to predict homologous recombination deficiency (HRD) as a non-invasive alternative to costly genomic tests, and to estimate survival and relapse risk for patient stratification, including analyses for rare subtypes such as endometrioid adenocarcinoma, clear cell adenocarcinoma, and ovarian carcinosarcoma. 

To make predictions clinically transparent, generative diffusion models will create counterfactual CT images that show which minimal changes would alter HRD or risk estimates, highlighting potential imaging biomarkers. Finally, the tool will be externally validated and evaluated with clinicians using in silico workflow trials across multiple countries to ensure usability and real-world clinical value.