Titre: Morphological analysis of brain subcortical structures in MR images from patients with neurological disorders

Conférencier : Mahsa Shakeri, étudiante au doctorat à Polytechnique Montréal

Résumé:
Studying morphological changes of subcortical structures often predicate neurological diseases, such as Alzheimer’s disease, schizophrenia, and subtypes of epilepsy. Over the last decades, image processing techniques, such as segmentation and registration, combined with machine learning approaches have enhanced the detection of different neuropathologies and early treatments. In this talk, we first present an automatic atlas-based segmentation method using a co-registration and co-segmentation process guided by deep priors. In this approach, for a given set of unseen magnetic resonance (MR) images, our method infers the segmentations on-the-fly by using previously learned priors, and incorporates this information to the energy formulation of a deformable registration. This results in a ground truth-free, data-driven co-registration and segmentation process that leads to more accurate results than standard multi-atlas segmentation strategies. Then, a groupwise spectral-based morphological process is presented in order to convert the output segmentations into 3D surfaces, establish one-to-one surface correspondences, and perform statistical shape analysis. The proposed morphological analysis framework can perform population-wise shape analysis in a computational efficient fashion and is highly sensitive to small morphological variations in patients with neuropathologies. Finally, a classifier based on the discriminant manifolds of the 3D surfaces will be presented. This method describes morphometric variations of 3D structures in a discriminant nonlinear graph embedding with Grassmannian manifolds, and detects the presence of a neuropathology.

Biographie :
Mahsa Shakeri is a 4th year PhD student under the supervision of Prof. Samuel Kadoury at Polytechnique Montreal in Computer Engineering. As an intern, she joined the Center for Visual Computing laboratory at INRIA Galen research team of Centrale Supeléc, France, in summer 2015. Her main research focus is image processing, morphological analysis, and machine learning.

Bienvenue à tous!