PhD defense Ismael Mounime: Accelerated Dynamic MR Imaging
Monday 5 Oct., 2026, at 14:00 (Paris time) at Télécom Paris
Télécom Paris, 19 place Marguerite Perey F-91120 Palaiseau [getting there], amphi 2 and in videoconferencing
Full title: Accelerated Dynamic MR Imaging Using Non-Linear Manifold Learning-Based Image Reconstruction Model and PET/MR Applications
Jury
-
Claude Comtat — Directeur de recherche, CEA, Université Paris-Saclay — Reviewer
-
Mathieu Hatt — Directeur de recherche INSERM, LaTIM, Université de Bretagne Occidentale — Reviewer
-
Frédérique Frouin — Chargée de recherche Hors Classe, INSERM, Institut Curie, laboratoire IRIS — Examiner
-
Philippe Ciuciu — CEA Fellow, Head of Inria-CEA MIND — Examiner
-
Kuang Gong — Assistant Professor of Biomedical Engineering, University of Florida — Examiner
-
Yoann Petibon — Associate Director, Clinical Imaging Lead, UCB — Examiner
-
Elsa Angelini — Professor, Télécom Paris (LTCI) — PhD Advisor
-
Georges El Fakhri — Professor of Radiology and Biomedical Imaging, Yale School of Medicine — PhD Advisor
-
Chao Ma — Professor, Yale University — Co-supervisor – Guest
-
Thibault Marin — Professor, Yale University — Co-supervisor – Guest
-
Pietro Gori — Professeur, Télécom Paris (LTCI) — Co-PhD Advisor – External guest
Abstract
Magnetic resonance imaging (MRI) is an essential tool for assessing cardiac anatomy, function, and tissue properties. Positron Emission Tomography (PET) provides complementary information on cardiac metabolism and function.
The intrinsically slow acquisition of MRI limits the observation of rapidly changing physiological processes such as cardiac and respiratory motion. This thesis investigates manifold-based MRI reconstruction methods for accelerated dynamic MRI and their applications in simultaneous PET/MR.
Learn more
The work builds upon the Linear Tangent Space Alignment (LTSA) model, which represents dynamic image series as lying on a low-dimensional nonlinear manifold. A sparsity-extended formulation, sparse-LTSA (sLTSA), was developed to enable high spatial- and temporal-resolution reconstructions from highly undersampled acquisitions.
The reconstructed MR images were applied to PET motion correction, enabling the estimation of respiratory and cardiac motion and the correction of motion-induced PET blurring. The method was also extended to 3D quantitative cardiac imaging, with sLTSA-based T1 and extracellular volume (ECV) mapping validated against the clinical 2D state-of-the-art method and applied to real simultaneous PET/MR acquisitions.
Several extensions of sLTSA were also explored, including tensor-based and patch-based implementations, as well as the incorporation of deep-learning-based priors.