Abstract : The motivation of this thesis is the study of the spatial organization of cardiac muscle fibers from a series of three-dimensional images acquired by Diffusion Tensor MRI (DT-MRI). This organization is a fundamental property underlying the heart contractile function. However it is very difficult to obtain considering the difficulties inherent to cardiac and respiratory motion. Our goal is to develop new approaches that can cope with physiological motion and noise sensititvity, for the estimation, the analysis and the visualization of myocardial fibers. My work is composed of three main axis. The first compares, in the context of ex vivo clinical studies, the main regularization approaches that operate either on diffusion weighted images or on diffusion tensors. The differences are small enough to conclude that the quality of our DT-MRI data is sufficient to consider all regularization methods as equivalent. The second concerns a new tractography method especially designed for cardiac specificity. It is guided by a global cost functional which allows automatic estimation of cardiac fibers in one shot, without using any initialization points. The latest axis consists in distinguishing a cardiac fibre population into clusters. It is based on the comparison of two classification methods (geometrical and topological type) using three different fibre representation modes. Our results establish that classification may allow automatic identification of myocardial regions from DT-MRI images, which could greatly ease analysis and comparison of these images towards the design of patient-specific therapies.