Skip to Main content Skip to Navigation
Theses

Indexation et fusion multimodale pour la recherche d'information par le contenu. Application aux bases de données d'images médicales.

Abstract : In this Ph.D. thesis, we study methods for information retrieval in databases made up of multimedia documents. Our objective is to select in a database documents similar to a query document. The aimed application is computer aided diagnosis in a medical framework: the database is made up of several images together with clinical contextual information about the patient. We first try to characterize each image in the patient file individually. We have thus proposed two original indexing methods derived from the wavelet transform of images: 1) a global method, modeling the distribution of wavelet coefficients in the image, 2) a local method, based on the extraction of lesions. Once images are characterized, we try to fuse all the information in the file to retrieve the closest patient files. In addition to the heterogeneity of the data, with have to cope with missing information in patient files. We propose three new approaches, derived from data mining and information fusion theory. The first approach is based on decision trees, the second one on Bayesian networks and the third one on the Dezert-Smarandache theory (DSmT). The results obtained on two multimodal medical databases are satisfying and superior to existing methods. Thus, the mean precision at five reaches 81.78% on a retinal image database and 92.90% on a mammography database.
Complete list of metadatas

Cited literature [142 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00326131
Contributor : Gwénolé Quellec <>
Submitted on : Wednesday, October 1, 2008 - 6:53:26 PM
Last modification on : Friday, September 25, 2020 - 3:35:49 AM
Long-term archiving on: : Monday, October 8, 2012 - 1:50:39 PM

Identifiers

  • HAL Id : tel-00326131, version 1

Citation

Gwénolé Quellec. Indexation et fusion multimodale pour la recherche d'information par le contenu. Application aux bases de données d'images médicales.. Traitement du signal et de l'image [eess.SP]. TELECOM Bretagne, 2008. Français. ⟨tel-00326131⟩

Share

Metrics

Record views

692

Files downloads

1994