Segmentation automatisée du ventricule gauche en IRM cardiaque : Evaluation supervisée et non supervisée de cette approche et application à l'étude de la viabilité myocardique

Abstract : The aim of this work is to perform an automated segmentation of the Left Ventricle on short-axis cardiac MR images with as few user interactions as possible. Based on a recently developed semi-automated segmentation method, a fully automated segmentation method is proposed that includes three main steps: the heart localization, the definition of a region of interest around the left ventricle, and finally its segmentation. The algorithm developed here takes into account anatomic and functional a priori information such as the temporal features of the heartbeat, the pseudo-circular shape of the LV, and the 3D continuity, combined with the image intensity features. The segmentation process is achieved using deformable models combined with morphological filters, which improve the model performances when dealing with heterogeneous gray levels within the cavity. The work achieved within the MedIEval group (Medical Imaging Evaluation) allowed to compare both proposed methods with 6 other methods, including 3 manual delineations by experts. In particular, an approach for ranking segmentation methods without using a gold standard was applied to the ejection fractions estimated by the 8 methods. Finally, the proposed segmentation method was used in a clinical research work about the regional contraction and thequantification of the myocardial infarction extent.Future work includes the automated segmentation of the right ventricle as well as the estimation of a robust mutual shape from several segmentation methods.
Complete list of metadatas

https://pastel.archives-ouvertes.fr/pastel-00982333
Contributor : Abes Star <>
Submitted on : Wednesday, April 23, 2014 - 4:37:15 PM
Last modification on : Thursday, October 17, 2019 - 12:36:08 PM
Long-term archiving on : Wednesday, July 23, 2014 - 12:50:36 PM

File

These_Constantinides_-_V2.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : pastel-00982333, version 1

Collections

Citation

Constantin Constantinides. Segmentation automatisée du ventricule gauche en IRM cardiaque : Evaluation supervisée et non supervisée de cette approche et application à l'étude de la viabilité myocardique. Médecine humaine et pathologie. Télécom ParisTech, 2012. Français. ⟨NNT : 2012ENST0034⟩. ⟨pastel-00982333⟩

Share

Metrics

Record views

461

Files downloads

2375