Label-efficient assessment and predictive modeling of aortic root morphology for Marfan syndrome

Marfan syndrome (MFS) is a rare hereditary disease that weakens the connective tissue of the
human body. This can lead to the splitting or bursting of the aortic wall. MFS patients can have
widely varying phenotypic expressions and rates of progression due to underlying genomic
variations, which are currently not understood. Treatment options exist, but therapeutic decisions are
entirely based on manually extracted aortic measurements that lack reproducibility and have poor
predictive power for complications.
This project aims to improve the assessment and prediction of patients at risk of rapid disease
progression. The main challenge is that Marfan patient data is scarce due to its rare nature. To cope
with this, we propose a novel label-efficient approach that incrementally learns from related tasks to
fine-tune the deep features and enhance their predictive power. We will perform a fusion of
handcrafted and deep features to increase performance in low-data settings and improve
interpretability. Finally, leveraging previous genomic research in our group, we will analyze
correlations between genomic information and radiomic features.
Our innovative research will result in an automatic aorta segmentation and measurement tool that
can be used for MFS but also other pathologies. The research on predictive markers will drive future
research and has the potential to substantially lower the burden and cost associated with Marfan and
related genetic aortic disorders