This thesis explores the issues involved in automatic coronary vessel analysis using deformable models and estimation algorithms. The work is carried out as a consequence of a previous international research and development project on multimedia and teleworking in medicine (radiology imaging). After developing and validating a multimedia workstation from a user perspective, the experience obtained showed the convenience to incorporate new computer vision tools to assist the physicians during the image assessment in the diagnostic process. The work presented is a part of a computer assisted diagnosis system within a telemedicine system.Focusing on cardiac imaging, the first step is to adapt the workstation to such image modality, then the goal is to build a computerised coronary tree model, able to incorporate 3D static (anatomical) and dynamic (vessel movements) data. Such a model is useful to increase the knowledge necessary when dealing with a difficult image modality in computer vision, like coronary angiography. To build the model, many computer vision problems have to be addressed. From low-level tasks as vessel detection up to high level ones like image understanding are necessarily covered. Mainly, deformable models (snakes) and estimation techniques are discussed and used with innovative ideas through the model building process. Ought to the tight dependence of the deformable models on the low level image feature detectors, new methods to learn the vessels based on statistical analysis and fine tune the detector are proposed increasing the segmentation confidence. A new statistical potential map is developed within a new energy minimising scheme. Snakes are also applied in the 3D-reconstruction process. A graph is designed and used to hold the knowledge of the complete model. The novel approach for vessel analysis and the final model were validated and the results are very encouraging.