Skip to Main content Skip to Navigation
Theses

Modèles, primitives et méthodes de suivi pour la segmentation vasculaire. Application aux coronaires en imagerie tomodensitométrique 3D.

Abstract : The segmentation of vascular structures is a fundamental step for diagnosis assistance and treatment. In this context, image processing techniques aim at easing and speeding up reviewing tasks by medical professionals, reducing the amount of manual interaction and lowering inter-operator variability. We first present an extensive bibliographical review of existing methods for 3D vascular segmentation following three axes of study: geometric and appearance models, image features and extraction schemes. In this work, we focus on a particularly challenging problem, the delineation of coronary arteries from 3D cardiac Computed Tomography Angiographic (CTA) data. We first detail our medial-based geometric model, which we evaluate on the image thanks to a fast, oriented, gradient flux-based medialness feature. We then formulate a recursive Bayesian model, which is learned in a non-parametric fashion from a ground-truth database of manually segmented datasets. We propose two separate extraction schemes to perform the actual segmentation. Our first strategy performs a discrete, graph-based optimization propagating minimal paths on a 4D (position+radius) graph. It exploits a new cumulative cost metric derived from our seminal Bayesian formulation. Our second strategy employs a sequential, stochastic Monte-Carlo tracking approach which successively estimates the posterior distribution of our Bayesian model through a population of random samples. For both approaches, particular attention is given to robustness and computational efficiency. Qualitative and quantitative evaluation is presented on a varied database of clinical data.
Document type :
Theses
Complete list of metadatas

Cited literature [324 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/pastel-00005908
Contributor : Ecole Télécom Paristech <>
Submitted on : Wednesday, May 19, 2010 - 8:00:00 AM
Last modification on : Friday, July 31, 2020 - 10:44:05 AM
Long-term archiving on: : Friday, October 19, 2012 - 3:15:19 PM

Identifiers

  • HAL Id : pastel-00005908, version 1

Citation

David Lesage. Modèles, primitives et méthodes de suivi pour la segmentation vasculaire. Application aux coronaires en imagerie tomodensitométrique 3D.. domain_other. Télécom ParisTech, 2009. English. ⟨pastel-00005908⟩

Share

Metrics

Record views

606

Files downloads

1888