Assimilation de données images pour la personnalisation d'un modèle électromécanique du coeur
Florence Billet · HAL (Le Centre pour la Communication Scientifique Directe) · 2010
Clinical data is more and more complex, with an explosion of imaging techniques and measurements. Numeric heart models can integrate and analyse these various data by adapting (personalising) them to each patient. These personalised models aim to improve the diagnosis and to find the therapies (e.g. pacing location and delays for cardiac resynchronisation therapy) that are best suited for each patient. This thesis proposes a framework to personalise an electromechanical model from time series of 3D images (such as cine-MRI or CT-scan). This personalization consists in estimating the state of the heart model (i.e. position/velocity) together with the electrical and mechanical parameters of the given electromechanical model. In this thesis, we mainly focus on estimation of the cardiac motion and of the parameters of contractility of the model. The first part of this thesis describes our improvements of the modelisation of the cardiac phases and a sensitivity analysis. The second part reports our generalisation of the segmentation and motion tracking with deformable models commonly used for the analysis of medical images. This generalisation consists in coupling a proactive deformable model with 3D images sequences. Finally, we propose a method to automatically estimate the active mechanical parameters of the model from 3D images sequences. The main issue arises from the strong non-linearity generated by the phase switches. We address this issue using variational assimilation of image data. The gradient of the error function is computed with the adjoint state method. These methods (motion/active mechanical parameters estimation) are evaluated on synthetic data and applied to clinical cine-MRI and CT-scan. This work is a first step towards the complete personalisation of the heart model and opens up possibilities for the optimisation of a therapy for each patient.