Coopération entre segmentation et mouvement pour l'estimation conjointe des déplacements pariétaux et des déformations myocardiaques

cooperation between segmentation and movement for the joint estimation of the parietal displacements and myocardiac deformations

• Auteur / Autrice : Viateur Tuyisenge
• Direction : Laurent Sarry
• Type : Thèse de doctorat
• Discipline(s) : Informatique et traitement d'images médicales
• Date : Soutenance le 08/12/2014
• Etablissement(s) : Clermont-Ferrand 1
• Ecole(s) doctorale(s) : École doctorale des sciences pour l'ingénieur (Clermont-Ferrand)
• Partenaire(s) de recherche : Laboratoire : Institut des Sciences de l'image pour les techniques interventionelles (ISIT)
• Jury : Président / Présidente : Jean-René Lusson
• Jury : Examinateurs / Examinatrices : Patrick Clarysse
• Jury : Rapporteurs / Rapporteuses : Jérôme Pousin, Thomas Corpetti

Mots clés

Mots clés contrôlés

Traitement d'images -- Techniques numériques

Myocarde

Troubles moteurs

Mots clés libres

Strain cardiaque

Estimation des déformations myocardiques

Couplage mouvement segmentation

Assimilation de données

Dyskinésie myocardique

Cardiac strain

Résumé

Pas de résumé disponible

The work done in this thesis is related to the project 3DStrain the overall objective of which is to develop a generic framework for the parietal and regional tracking of the left ventricle and to adapt it the 3D + t cardiac imaging modalities used in clinical routine (3D ultrasound, SPECT, cine MRI). We worked on the parietal motion and myocardial deformation. We made the state-of-the-art on motion estimation approaches in general and on methods applied to imaging modalities in clinical practice to quantify myocardial deformation taking into account their specificities and limitations. We focused on tracking methods that optimize the similarity between the intensities between consecutive images of a sequence to estimate the spatial velocity field. They are based on the assumption of the invariance of image gray level (or optical flow) and regularization terms are used to solve the aperture problem. We proposed a regularization term well suited to physical and physiological properties of myocardial motion. The advantage of the proposed approach relies on its flexibility to estimate the dense field of myocardial motion on image sequences over the cardiac cycle. Motion is estimated while preserving myocardial wall discontinuities. However, the data similarity term used in our method is based only on the intensity of the image. It properly estimates the displacement field especially in the radial direction as the movement of circumferential twist is hardly visible on cine MRI in short axis view, the data we used for performing the experiments. To make the estimation more robust, we proposed a dynamic evolution model for the cardiac contraction and relaxation to introduce the temporal constraint ofthe dynamics of the heart. This model helps to estimate not only the dense field of myocardial displacement, but also other parameters of myocardial contractility (the contraction phase and asymmetry between systole and diastole) in variational data assimilation formalism. Automatic estimation of deformation and myocardial contractibility (the strain, phase and asymmetry) was validated against the cardiological and radiological expertise (Dr Elisabeth Coupez and Dr Lucie Cassagnes, CHU Clermont-Ferrand) through semi-quantitative scores of contraction called Wall Motion Score (WMS) and Wall Thickening Index (WTI). The proposed method provides promising results for both motion estimation results and the diagnosis indices for evaluation of myocardial dyskinesia. In order to gain in robustness and accuracy, it is necessary to perform the measurement of strain and indices of myocardial contraction precisely inside endocardial and epicardial walls. Therefore, we conducted a collaborative work with Kevin Bianchi, another PhD student on the project 3DStrain and we proposed a method of coupling of myocardial segmentation by deformable models and estimation of myocardial motion in a variational data assimilation framework.