Graph representation learning for Pulmonary Embolism Risk Stratification Models

We are seeking a highly motivated postdoctoral researcher to join our cutting-edge project aimed
at establishing new clinical practices for Pulmonary Embolism (PE) management. The pulmonary
vascular tree, the primary site of PE, can be modeled as a graph (see Fig. 1). Relying on our unique
dataset of over 400 patients, the goal is to leverage state-of-the-art multimodal learning [3] and graph
representation learning [2, 1] to develop a graph embedding unifying CTPA images, tabular patient
data, and specific biomarkers.
This graph embedding will serve as the foundation for creating novel risk stratification models.
Given the importance of interpretability for clinical applications, a key aspect of this research will
involve exploring the learned embeddings to assess the individual contribution of each biomarker to
the stratification task. This exploration will pave the way for larger-scale clinical studies, ultimately
defining future clinical practices for PE management.

More info on the attached pdf.