We present an efficient pipeline for the segmentation and tracking of the Epicardium (Epi) and Left Ventricle (LV) from a set of 3D cardiac MRI sequences. The surface structure is handled as an array of planar active contours, interconnected between adjacent slices and frames, providing spatial and temporal consistency. In a given cardiac phase, the stacking of slice contours constitute a 3D mesh with a cylindrical topology. In a first stage the heart region is automatically localized by making use of the time variance of the heart region. Extraction of heart border is performed by means of energy minimization. Finally, we construct a 3D tensor array from all time surfaces in cylindrical coordinates and time and space consistency of the segmentation is enforced by fitting an m-variate tensor smoothing spline to the final 3D array.