Thesis on "Quantifying and understanding uncertainty in deep-learning-based medical imagesegmentation"• Development of novel, efficient methods to evaluate the uncertainty of deep neural networks in the context of 3D medical image processing (MRI, CT-scan).• Extension of voxel-wise uncertainty estimates to structure-wise metrics (e.g lesions, ROI), in particular using Graph Neural Networks. • Automatic detection of out-of-distribution images from the feature space of trained models• Construction of predictive intervals associated to volume estimation (e.g tumor volume estimation). • Calibration of Deep Learning model. • Conformal prediction applied to 2D and 3D medical image segmentation.• Generation of synthetic images using Diffusion models

Deep Learning approaches for anomaly detection in 3D medical images, based on Unsupervised Autoencoders.

Deep Learning approaches for medical image synthesis and vessels segmentation using Conditional Generative Adversarial Networks. Application to stroke detection in head CT scans.

Creation of a Human-Machine Interface dedicated to PET tomographic reconstruction.