Skip to Main content Skip to Navigation

Fusion d'informations et segmentation d'images basées sur la théorie des fonctions de croyance : Application à l'imagerie médicale TEP multi-traceurs

Benoît Lelandais 1 
1 QuantIF-LITIS - Equipe Quantification en Imagerie Fonctionnelle
LITIS - Laboratoire d'Informatique, de Traitement de l'Information et des Systèmes
Abstract : Multi-tracer Positron Emission Tomography (PET) functional imaging could have a prominent effect for the treatment of cancer by radiotherapy. PET images using 18Fluoro-Deoxy-Glucose (18 FDG), 18F-Fluoro-L Thymidine (18 FLT) and 18Fluoro-Misonidazole (18 FMiso) tracers are respectively indicators of glucose cell consumption, cell proliferation and hypoxia (cell lack of oxygen). The joint use of these three tracers could helps us to define sub-volumes lea- ding to an adequate treatment. For this purpose, it is imperative to provide a medical tool of segmentation and fusion of these images. PET images have the characteristic of being very noisy and having a low spatial resolution. These imperfections result in the presence of uncertain and imprecise information in the images respectively. Our contribution consists in proposing a method, called EVEII for Evidential Voxel-based Estimation of Imperfect Information, based on belief function theory, for providing a reliable and accurate segmentation in the context of imperfect images. It also lies in proposing a method for the fusion of multi-tracer PET images. The study of EVEII on simulated images reveals that it is the best suited compared to other methods based on belief function theory, giving a good recognition rate of almost 100 % of pixels when the signal-to-noise ratio is greater than 2.5. On PET physical phantoms, simulating the characteristics of 18FDG, 18FLT and 18FMiso PET images, the results show that our method gives the better estimation of sphere volumes to segment compared to methods of the literature proposed for this purpose. On both lowly and highly noisy phantoms respectively, mean error bias of volume estimation are only of -0.27 and 3.89 mL, demonstrating its suitability for PET image segmentation task. Finally, our method is applied to the segmentation of multi-tracer PET images for three patients. The results show that our method is well suited for the fusion of multi-tracer PET images, giving for this purpose a set of parametric images for differentiating the different biological tissues.
Document type :
Complete list of metadata

Cited literature [189 references]  Display  Hide  Download
Contributor : Benoit Lelandais Connect in order to contact the contributor
Submitted on : Tuesday, December 3, 2013 - 4:53:17 PM
Last modification on : Wednesday, March 2, 2022 - 10:10:09 AM
Long-term archiving on: : Monday, March 3, 2014 - 10:10:11 PM


  • HAL Id : tel-00912534, version 1


Benoît Lelandais. Fusion d'informations et segmentation d'images basées sur la théorie des fonctions de croyance : Application à l'imagerie médicale TEP multi-traceurs. Traitement des images [eess.IV]. Université de Rouen, 2013. Français. ⟨tel-00912534⟩



Record views


Files downloads