Skip to Main content Skip to Navigation
Master Thesis

Utilisation de l’apprentissage de variété pour le diagnostic médical : application aux neuropathies optiques

Abstract : The diffusion-weighted magnetic resonance images reveal the Brownian motion of the water molecules which can be used as raw data to reconstruct various fascicles with tractography post-processing. Deep learning algorithms have recently been proposed to improve the tracking process limitations. However, the tractographic process as itself remains time consuming for clinical practice as well as highly dependent on the diffusion-signal modeling and conditions of tracking. In addition, using supervised machine learning methods raise a problem of gold standard choice. Poposing an unsupervised machine learning method with the ability to reconstruct anatomical structures directly from the Fiber Orientation Distribution (FOD) and computationally efficient hold promise to explore disease models. Statistical representation of the FOD geometry requires appropriate theories to enforce physically meaningful properties in a non-Euclidean space. Here we present a new framework for FOD analysis, as extracted by manifold Riemannian approximation. We firstly model FOD shapes as a quantitative data-driven learning task and apply this framework to classify various synthetic FODs in a manifold subspace. We then applied this algorithm on the human anterior optic pathways in normal and pathological conditions, to demonstrate the feasibility for single-subject diagnosis.of ischemic or demyelinating optic neuropathy.
Document type :
Master Thesis
Complete list of metadata
Contributor : Jean-Hugues Morneau Connect in order to contact the contributor
Submitted on : Friday, May 28, 2021 - 10:39:04 AM
Last modification on : Friday, June 4, 2021 - 3:22:27 AM
Long-term archiving on: : Sunday, August 29, 2021 - 6:29:32 PM


Files produced by the author(s)


  • HAL Id : dumas-03240571, version 1
  • PPN : 236336371


Charlotte Fabert. Utilisation de l’apprentissage de variété pour le diagnostic médical : application aux neuropathies optiques. Médecine humaine et pathologie. 2019. ⟨dumas-03240571⟩



Record views


Files downloads