Optimization of a Shape Metric Based on Information Theory Applied to Segmentation Fusion and Evaluation in Multimodal MRI for DIPG Tumor Analysis - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

Optimization of a Shape Metric Based on Information Theory Applied to Segmentation Fusion and Evaluation in Multimodal MRI for DIPG Tumor Analysis

Optimisation d'une métrique de forme basée sur la théorie de l'information pour la fusion et l'évaluation de méthodes de segmentation de tumeurs GITC en IRM multimodale 3D

Résumé

In medical imaging, the construction of a reference shape from a set of segmentation results from different algorithms or image modalities is an important issue when dealing with the evaluation of segmentation without knowing the gold standard or when an evaluation of the inter or intra expert variability is needed. It is also interesting to build this consensus shape to merge the results obtained for the same target object from automatic or semi-automatic segmentation methods. In this paper, to deal with both segmentation fusion and evaluation, we propose to define such a “mutual shape” as the optimum of a criterion using both the mutual information and the joint entropy of the segmentation methods. This energy criterion is justified using the similarities between quantities of information theory and area measures and is presented in a continuous variational framework. We investigate the applicability of our framework for the fusion and evaluation of segmentation methods in multimodal MR images of diffuse intrinsic pontine glioma (DIPG).
Fichier non déposé

Dates et versions

hal-03357235 , version 1 (28-09-2021)

Identifiants

Citer

Stéphanie Jehan-Besson, Régis Clouard, Nathalie Boddaert, Jacques Grill, Frédérique Frouin. Optimization of a Shape Metric Based on Information Theory Applied to Segmentation Fusion and Evaluation in Multimodal MRI for DIPG Tumor Analysis. International Conference on Geometric Science of Information, Jul 2021, Paris, France. pp.772-780, ⟨10.1007/978-3-030-80209-7_83⟩. ⟨hal-03357235⟩
56 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More