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

Stéphanie Jehan-Besson, Régis Clouard, Nathalie Boddaert, Jacques Grill, Frédérique Frouin

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

    Abstract

    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).

    Original languageEnglish
    Title of host publicationGeometric Science of Information - 5th International Conference, GSI 2021, Proceedings
    EditorsFrank Nielsen, Frédéric Barbaresco
    PublisherSpringer Science and Business Media Deutschland GmbH
    Pages772-780
    Number of pages9
    ISBN (Print)9783030802080
    DOIs
    Publication statusPublished - 1 Jan 2021
    Event5th International Conference on Geometric Science of Information, GSI 2021 - Paris, France
    Duration: 21 Jul 202123 Jul 2021

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume12829 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Conference

    Conference5th International Conference on Geometric Science of Information, GSI 2021
    Country/TerritoryFrance
    CityParis
    Period21/07/2123/07/21

    Keywords

    • Information theory
    • Multimodal MR imaging
    • Mutual shape
    • Neuro-oncology
    • Segmentation evaluation
    • Segmentation fusion
    • Shape optimization

    Cite this