Skip to Main content Skip to Navigation

Segmentation d'images multispectrales basée sur
la fusion d'informations : application aux images IRM

Abstract : The objective of this work consists of developing architecture of information fusion based on the fuzzy theory in order to segment a target from multiple sources of image. Our principal application carries on the image segmentation of multispectral cerebral MRI. We propose an approach of automatic segmentation based on the fusion of characteristics extracted from each source of image. These characteristics are modellized by some membership functions obtained from analytic functions, that not only takes into account some a priori knowledge from an expert about the possibility of target (tumor or cerebral tissues) membership, but also deals with graduality of target relative to signal intensity. Finally, the target segmentation consists on the fusion of different membership degrees of the target. An additional step based on 3D fuzzy region growing is proposed to improve the result of fusion. To evaluate these results of segmentation represented by a fuzzy set, an extension of Cohen's Kappa coefficient is proposed and named "fuzzy Kappa" which is a method of global evaluation concerning the agreement proportion of a fuzzy classification.
This developed architecture is performed for the segmentation of cerebral tumors from MRI images that presently include these routine sequences: T1, T2 and proton density. The results obtained from seven patients with tumor show the efficacy of our system.
Document type :
Complete list of metadatas

Cited literature [157 references]  Display  Hide  Download
Contributor : Hal System <>
Submitted on : Monday, November 6, 2006 - 3:15:44 PM
Last modification on : Friday, October 23, 2020 - 4:37:18 PM
Long-term archiving on: : Thursday, September 20, 2012 - 2:27:17 PM


  • HAL Id : tel-00111904, version 1


Weibei Dou. Segmentation d'images multispectrales basée sur
la fusion d'informations : application aux images IRM. Informatique. Université de Caen, 2006. Français. ⟨tel-00111904⟩



Record views


Files downloads