MRI Velocimetry is a useful modality to explore cardiovascular disease. The sequence of phase contrast MRI has the characteristic of providing both anatomical information and the physiological data, thus allow to measure geometric properties of vessels and blood flows. The purpose of this thesis is to automate the segmentation of vessels and velocimetry measurements, manual processing is inadequate to operate the MRI velocimetry for a diagnostic in clinical routine. The work led to propose a segmentation method based on active contours guided by an information region ("region-based" approach), unlike conventional approaches focusing only on cross-border regions ("edge-based" approach). This "region-based" approach was evaluated on data from a phantom made to provide an objective reference. A second evaluation was conducted on the basis of 28 carotid arteries (14 patients) manually segmented by an expert radiologist. The results of the "phantom" data show that the "edge-based" approach leads to an error in measuring the area of the lumen of the segmented carotid and extent linked flows, respectively 18.4 % and 3.6%. These errors are larger than those obtained using the proposed approach (respectively 2.3 % and 0.7 %). This benefit appears much higher on the database of patients with an underestimation of areas and blood flow, respectively 40.5 % and 26.5 % for the "edge-based" approach, against 14.7 and 6.4 % for the proposed approach.
