Khotanlou, Hassan (2008) 3D brain tumors and internal brain structures segmentation in MR images. PhD thesis, Traitement du Signal et de l'Image (TSI), ENST p.280.

Full text available as: - thesis-khotanlou.pdf ( 9603 Kb )

Licence: Copyright

Abstract

The main topic of this thesis is to segment brain tumors, their components (edema and

necrosis) and internal structures of the brain in 3D MR images. For tumor segmentation

we propose a framework that is a combination of region-based and boundary-based

paradigms. In this framework, we first segment the brain using a method adapted for

pathological cases and extract some global information on the tumor by symmetrybased

histogram analysis. The second step segments the tumor and its components.

For this, we propose a new and original method that combines region and boundary

information in two phases: initialization and refinement. For initialization, which

is mostly region-based, we present two new methods. The first one is a new fuzzy

classification method which combines the membership, typicality and neighborhood

information of the voxels. The second one relies on symmetry-based histogram analysis.

The initial segmentation of the tumor is refined relying on boundary information

of the image. This method is a deformable model constrained by spatial relations.

The spatial relations are obtained based on the initial segmentation and surrounded

tissues of the tumor. The proposed method can be used for a large class of tumors in

any modality of MR images. To segment a tumor and its components full automatically

the proposed framework needs only a contrast enhanced T1-weighted image and

a FLAIR image. In the case of a contrast enhanced T1-weighted image only, some

user interaction will be needed.

We evaluated this method on a data set of 20 contrast enhanced T1-weighted and

10 FLAIR images with different types of tumors.

Another aim of this thesis is the segmentation of internal brain structures in the

presence of a tumor. For this, a priori knowledge about the anatomy and the spatial

organization of the structures is provided by an ontology. To segment each structure,

we first exploit its relative spatial position from a priori knowledge. We then select

the spatial relations which remain consistent using the information on the segmented

tumor. These spatial relations are then fuzzified and fused in a framework proposed

by our group. As for the tumor, the segmentation process of each structure has

two steps. In the first step we search the initial segmentation of the structure in a

globally segmented brain. The search process is done in the region of interest (ROI)

provided by the fused spatial relations. To globally segment the brain structures we

use two methods, the first one is the proposed fuzzy classification and the second one is a multiphase level sets. To refine the initial segmentation, we use a deformable

model which is again constrained by the fused spatial relations of the structure. This

method was also evaluated on 10 contrast enhanced T1-weighted images to segment

the ventricles, caudate nucleus and thalamus.

Statistiques de consultation

Repository Staff Only: edit this item