The present invention relates to a method of segmenting a starting image or sequence of tridimensional images for obtaining a tridimensional segmentation image comprising a partition into regions of interest, said image or sequence of images comprising measurements, for each voxel and in the course of n time intervals (n.gtoreq.1), of the real evolution of a signal representative of at least one variable of said image or sequence, which comprises essentially: a) a modeling (10) of the signal comprising the definition of a parametric model of spatio-temporal evolution of said signal, this model comprising sets of homogeneous parameters respectively specific to structures corresponding to said regions of interest; b) an extraction (30) of samples of voxels respectively included in said structures; then c) a merging (50b) of the samples grouping together those whose evolution model is specific to the same structure, said merging following, preceding or including a classification of all the voxels of said image or sequence of images or of a zone of interest of the latter by aggregation with a group of samples.