The aim of this paper is to present three approaches to cardiac ventricle segmentation, which apply the potential active contour method. Two of these approaches use three-dimensional, and one of them - four-dimensional representation of data. The approaches presented simulates expert’s behaviour. They aim at image segmentation of cardiac ventricles performed at all slices simultaneously, thanks to which every slice can be analysed in the context of knowledge about other slices. The medical image understanding method is not fully automatic, however in comparison to manual segmentation performed by an expert, it saves much time, which may be of vital importance for patient’s health e.g. in pulmonary embolism diagnosis.