I'm Emile Gaudinot

Segmentation of coronary arteries from 800 CTA images of the heart, in 3D. Extraction of a graph from the arteries, to automate early detection of Coronary Artery Disease. The whole standardized Deep Learning pipeline we developed was deployed to production. Models tested: U-Net, nnU-Net, UNETR, MedSAM. Best Dice-score of 0.77.

Winner among 24 competitors! Implementation, training, and fine-tuning of various proof-of-concept ML models to tackle cybersecurity challenges: adversarial attacks/defenses, network intrusion detection, malware detection in documents, etc. Models leveraged: SVM, Random Forests, Gradient Boosting, LOF. Feature extraction: Bag-of-Words and PCA.

Direct collaboration with the core team of MNE-Python, a Python package for EEG and MEG analysis. Expanded the SPoC (Dähne et al., 2014) and SSD (Nikulin et al., 2011) functions. Contributions merged via pull requests, strengthening this open-source neuroscience tool.

This project bridges computational geometry and cardiac research. It provides a robust method to generate 3D geodesic meshes of heart valves, to study their deformations. This is critical for advancing our understanding of pathologies like Mitral Valve Prolapse (~3% of the population).