Représentation, évaluation et utilisation de relations spatiales pour l'interprétation d'images. Application à la reconnaissance de structures anatomiques en imagerie médicale

Abstract : Spatial relations such as adjacencies, distances, directions or symmetries constitute a fundamental element in descriptions of the spatial arrangement of objects in a scene and are therefore useful for a large number of image understanding tasks. Our main application is the recognition of brain structures. The brain is a structured scene in which objects share a large number of meaningful spatial relations.

Our work deals with the following questions.
How can spatial relations be represented ? How can they be used to formalize the description of a scene, in particular to describe brain anatomy ? How can they be used for image segmentation and pattern recognition, in particular for brain structures.
First we study a particular case of relations: plane symmetries, which are important in brain imaging and have received less attention than other types of relations. We propose a method to quantify the degree of symmetry in fuzzy objects and images, defining a measure of symmetry, as well as an algorithm to find the symmetry plane of a given object. The application of this method to the brain symmetry plane is precise and robust as it has been shown through an evaluation of synthetic and real images.
Then we present how brain structures can be described with spatial relations and how this description can be formalized by a synthetic hierarchical graph. This description is then compared to the one obtained with an iconic atlas, using a structure called an example hierarchical graph.
The last two parts are devoted to the application of spatial relations in image segmentation.
We propose a method to integrate this knowledge in a deformable model through the construction of a new external force. This approach is very different from other methods available because it makes directly use of the relations to find the contours of the objects. Finally, a procedure for segmenting brain internal structures is proposed, based on the previous description and a deformable model. First results, visually evaluated, are promising.
Complete list of metadatas

https://pastel.archives-ouvertes.fr/tel-00005764
Contributor : Admin Pastel <>
Submitted on : Monday, April 5, 2004 - 6:02:30 PM
Last modification on : Friday, May 3, 2019 - 1:17:51 AM
Long-term archiving on : Friday, April 2, 2010 - 8:42:31 PM

Identifiers

  • HAL Id : tel-00005764, version 1

Collections

Citation

Olivier Colliot. Représentation, évaluation et utilisation de relations spatiales pour l'interprétation d'images. Application à la reconnaissance de structures anatomiques en imagerie médicale. Interface homme-machine [cs.HC]. Télécom ParisTech, 2003. Français. ⟨tel-00005764⟩

Share

Metrics

Record views

870

Files downloads

733